12 Commits

Author SHA1 Message Date
Yeachan-Heo
40008b6513 wip: grok provider abstraction 2026-04-01 06:00:48 +00:00
Yeachan-Heo
f477dde4a6 feat: provider tests + grok integration 2026-04-01 05:45:27 +00:00
Yeachan-Heo
178934a9a0 feat: grok provider tests + cargo fmt 2026-04-01 04:20:15 +00:00
Yeachan-Heo
2a0f4b677a feat: provider abstraction layer + Grok API support 2026-04-01 04:10:46 +00:00
Yeachan-Heo
cbc0a83059 auto: save WIP progress from rcc session 2026-04-01 04:01:37 +00:00
Yeachan-Heo
ac6c5d00a8 Enable Claude-compatible tool hooks in the Rust runtime
This threads typed hook settings through runtime config, adds a shell-based hook runner, and executes PreToolUse/PostToolUse around each tool call in the conversation loop. The CLI now rebuilds runtimes with settings-derived hook configuration so user-defined Claude hook commands actually run before and after tools.

Constraint: Hook behavior needed to match Claude-style settings.json hooks without broad plugin/MCP parity work in this change
Rejected: Delay hook loading to the tool executor layer | would miss denied tool calls and duplicate runtime policy plumbing
Confidence: medium
Scope-risk: moderate
Reversibility: clean
Directive: Keep hook execution in the runtime loop so permission decisions and tool results remain wrapped by the same conversation semantics
Tested: cargo test; cargo build --release
Not-tested: Real user hook scripts outside the test harness; broader plugin/skills parity
2026-04-01 03:35:25 +00:00
Yeachan-Heo
a94ef61b01 feat: -p flag compat, --print flag, OAuth defaults, UI rendering merge 2026-04-01 03:22:34 +00:00
Yeachan-Heo
a9ac7e5bb8 feat: default OAuth config for claude.com, merge UI polish rendering 2026-04-01 03:20:26 +00:00
Yeachan-Heo
0175ee0a90 Merge remote-tracking branch 'origin/rcc/ui-polish' into dev/rust 2026-04-01 03:17:16 +00:00
Yeachan-Heo
1bd0eef368 Merge remote-tracking branch 'origin/rcc/subagent' into dev/rust 2026-04-01 03:12:25 +00:00
Yeachan-Heo
ba220d210e Enable real Agent tool delegation in the Rust CLI
The Rust Agent tool only persisted queued metadata, so delegated work never actually ran. This change wires Agent into a detached background conversation path with isolated runtime, API client, session state, restricted tool subsets, and file-backed lifecycle/result updates.

Constraint: Keep the tool entrypoint in the tools crate and avoid copying the upstream TypeScript implementation
Rejected: Spawn an external claw process | less aligned with the requested in-process runtime/client design
Rejected: Leave execution in the CLI crate only | would keep tools::Agent as a metadata-only stub
Confidence: medium
Scope-risk: moderate
Reversibility: clean
Directive: Tool subset mappings are curated guardrails; revisit them before enabling recursive Agent access or richer agent definitions
Tested: cargo build --release --manifest-path rust/Cargo.toml
Tested: cargo test --manifest-path rust/Cargo.toml
Not-tested: Live end-to-end background sub-agent run against Anthropic API credentials
2026-04-01 03:10:20 +00:00
Yeachan-Heo
04b1f1e85d docs: rewrite rust/ README with full feature matrix and usage guide 2026-04-01 02:59:05 +00:00
25 changed files with 4800 additions and 1290 deletions

5
.claude.json Normal file
View File

@@ -0,0 +1,5 @@
{
"permissions": {
"defaultMode": "dontAsk"
}
}

3
.gitignore vendored
View File

@@ -2,3 +2,6 @@ __pycache__/
archive/
.omx/
.clawd-agents/
# Claude Code local artifacts
.claude/settings.local.json
.claude/sessions/

21
CLAUDE.md Normal file
View File

@@ -0,0 +1,21 @@
# CLAUDE.md
This file provides guidance to Claude Code (claude.ai/code) when working with code in this repository.
## Detected stack
- Languages: Rust.
- Frameworks: none detected from the supported starter markers.
## Verification
- Run Rust verification from `rust/`: `cargo fmt`, `cargo clippy --workspace --all-targets -- -D warnings`, `cargo test --workspace`
- `src/` and `tests/` are both present; update both surfaces together when behavior changes.
## Repository shape
- `rust/` contains the Rust workspace and active CLI/runtime implementation.
- `src/` contains source files that should stay consistent with generated guidance and tests.
- `tests/` contains validation surfaces that should be reviewed alongside code changes.
## Working agreement
- Prefer small, reviewable changes and keep generated bootstrap files aligned with actual repo workflows.
- Keep shared defaults in `.claude.json`; reserve `.claude/settings.local.json` for machine-local overrides.
- Do not overwrite existing `CLAUDE.md` content automatically; update it intentionally when repo workflows change.

View File

@@ -0,0 +1 @@
{"messages":[{"blocks":[{"text":"clear","type":"text"}],"role":"user"},{"blocks":[{"text":"\n\nI've cleared the conversation. How can I help you today?","type":"text"}],"role":"assistant","usage":{"cache_creation_input_tokens":0,"cache_read_input_tokens":0,"input_tokens":4272,"output_tokens":17}}],"version":1}

View File

@@ -0,0 +1 @@
{"messages":[{"blocks":[{"text":"exit","type":"text"}],"role":"user"},{"blocks":[{"text":"\n\nGoodbye! 👋","type":"text"}],"role":"assistant","usage":{"cache_creation_input_tokens":0,"cache_read_input_tokens":0,"input_tokens":4272,"output_tokens":10}}],"version":1}

View File

@@ -0,0 +1 @@
{"messages":[],"version":1}

View File

@@ -1,22 +1,27 @@
[
{
"content": "Phase 0: Structural Cleanup — spawn 4 agents for 0.1-0.4",
"activeForm": "Executing Phase 0: Structural Cleanup via sub-agents",
"content": "Architecture & dependency analysis",
"activeForm": "Complete",
"status": "completed"
},
{
"content": "Runtime crate deep analysis",
"activeForm": "Complete",
"status": "completed"
},
{
"content": "CLI & Tools analysis",
"activeForm": "Complete",
"status": "completed"
},
{
"content": "Code quality verification",
"activeForm": "Complete",
"status": "completed"
},
{
"content": "Synthesize findings into unified report",
"activeForm": "Writing report",
"status": "in_progress"
},
{
"content": "Phase 1.1-1.2: Status bar with live HUD and token counter",
"activeForm": "Awaiting Phase 0",
"status": "pending"
},
{
"content": "Phase 2.4: Remove artificial 8ms stream delay",
"activeForm": "Awaiting Phase 0",
"status": "pending"
},
{
"content": "Phase 3.1: Collapsible tool output",
"activeForm": "Awaiting Phase 0",
"status": "pending"
}
]

2
rust/Cargo.lock generated
View File

@@ -1545,10 +1545,12 @@ dependencies = [
name = "tools"
version = "0.1.0"
dependencies = [
"api",
"reqwest",
"runtime",
"serde",
"serde_json",
"tokio",
]
[[package]]

View File

@@ -1,230 +1,149 @@
# Rusty Claude CLI
# 🦞 Claw Code — Rust Implementation
`rust/` contains the Rust workspace for the integrated `rusty-claude-cli` deliverable.
It is intended to be something you can clone, build, and run directly.
A high-performance Rust rewrite of the Claude Code CLI agent harness. Built for speed, safety, and native tool execution.
## Workspace layout
## Quick Start
```text
```bash
# Build
cd rust/
cargo build --release
# Run interactive REPL
./target/release/claw
# One-shot prompt
./target/release/claw prompt "explain this codebase"
# With specific model
./target/release/claw --model sonnet prompt "fix the bug in main.rs"
```
## Configuration
Set your API credentials:
```bash
export ANTHROPIC_API_KEY="sk-ant-..."
# Or use a proxy
export ANTHROPIC_BASE_URL="https://your-proxy.com"
```
Or authenticate via OAuth:
```bash
claw login
```
## Features
| Feature | Status |
|---------|--------|
| Anthropic API + streaming | ✅ |
| OAuth login/logout | ✅ |
| Interactive REPL (rustyline) | ✅ |
| Tool system (bash, read, write, edit, grep, glob) | ✅ |
| Web tools (search, fetch) | ✅ |
| Sub-agent orchestration | ✅ |
| Todo tracking | ✅ |
| Notebook editing | ✅ |
| CLAUDE.md / project memory | ✅ |
| Config file hierarchy (.claude.json) | ✅ |
| Permission system | ✅ |
| MCP server lifecycle | ✅ |
| Session persistence + resume | ✅ |
| Extended thinking (thinking blocks) | ✅ |
| Cost tracking + usage display | ✅ |
| Git integration | ✅ |
| Markdown terminal rendering (ANSI) | ✅ |
| Model aliases (opus/sonnet/haiku) | ✅ |
| Slash commands (/status, /compact, /clear, etc.) | ✅ |
| Hooks (PreToolUse/PostToolUse) | 🔧 Config only |
| Plugin system | 📋 Planned |
| Skills registry | 📋 Planned |
## Model Aliases
Short names resolve to the latest model versions:
| Alias | Resolves To |
|-------|------------|
| `opus` | `claude-opus-4-6` |
| `sonnet` | `claude-sonnet-4-6` |
| `haiku` | `claude-haiku-4-5-20251213` |
## CLI Flags
```
claw [OPTIONS] [COMMAND]
Options:
--model MODEL Set the model (alias or full name)
--dangerously-skip-permissions Skip all permission checks
--permission-mode MODE Set read-only, workspace-write, or danger-full-access
--allowedTools TOOLS Restrict enabled tools
--output-format FORMAT Output format (text or json)
--version, -V Print version info
Commands:
prompt <text> One-shot prompt (non-interactive)
login Authenticate via OAuth
logout Clear stored credentials
init Initialize project config
doctor Check environment health
self-update Update to latest version
```
## Slash Commands (REPL)
| Command | Description |
|---------|-------------|
| `/help` | Show help |
| `/status` | Show session status (model, tokens, cost) |
| `/cost` | Show cost breakdown |
| `/compact` | Compact conversation history |
| `/clear` | Clear conversation |
| `/model [name]` | Show or switch model |
| `/permissions` | Show or switch permission mode |
| `/config [section]` | Show config (env, hooks, model) |
| `/memory` | Show CLAUDE.md contents |
| `/diff` | Show git diff |
| `/export [path]` | Export conversation |
| `/session [id]` | Resume a previous session |
| `/version` | Show version |
## Workspace Layout
```
rust/
├── Cargo.toml
├── Cargo.toml # Workspace root
├── Cargo.lock
├── README.md
└── crates/
├── api/ # Anthropic API client + SSE streaming support
├── commands/ # Shared slash-command metadata/help surfaces
├── compat-harness/ # Upstream TS manifest extraction harness
├── runtime/ # Session/runtime/config/prompt orchestration
├── rusty-claude-cli/ # Main CLI binary
├── api/ # Anthropic API client + SSE streaming
├── commands/ # Shared slash-command registry
├── compat-harness/ # TS manifest extraction harness
├── runtime/ # Session, config, permissions, MCP, prompts
├── rusty-claude-cli/ # Main CLI binary (`claw`)
└── tools/ # Built-in tool implementations
```
## Prerequisites
### Crate Responsibilities
- Rust toolchain installed (`rustup`, stable toolchain)
- Network access and Anthropic credentials for live prompt/REPL usage
- **api** — HTTP client, SSE stream parser, request/response types, auth (API key + OAuth bearer)
- **commands** — Slash command definitions and help text generation
- **compat-harness** — Extracts tool/prompt manifests from upstream TS source
- **runtime** — `ConversationRuntime` agentic loop, `ConfigLoader` hierarchy, `Session` persistence, permission policy, MCP client, system prompt assembly, usage tracking
- **rusty-claude-cli** — REPL, one-shot prompt, streaming display, tool call rendering, CLI argument parsing
- **tools** — Tool specs + execution: Bash, ReadFile, WriteFile, EditFile, GlobSearch, GrepSearch, WebSearch, WebFetch, Agent, TodoWrite, NotebookEdit, Skill, ToolSearch, REPL runtimes
## Build
## Stats
From the repository root:
- **~20K lines** of Rust
- **6 crates** in workspace
- **Binary name:** `claw`
- **Default model:** `claude-opus-4-6`
- **Default permissions:** `danger-full-access`
```bash
cd rust
cargo build --release -p rusty-claude-cli
```
## License
The optimized binary will be written to:
```bash
./target/release/rusty-claude-cli
```
## Test
Run the verified workspace test suite used for release-readiness:
```bash
cd rust
cargo test --workspace --exclude compat-harness
```
## Quick start
### Show help
```bash
cd rust
cargo run -p rusty-claude-cli -- --help
```
### Print version
```bash
cd rust
cargo run -p rusty-claude-cli -- --version
```
### Login with OAuth
Configure `settings.json` with an `oauth` block containing `clientId`, `authorizeUrl`, `tokenUrl`, optional `callbackPort`, and optional `scopes`, then run:
```bash
cd rust
cargo run -p rusty-claude-cli -- login
```
This opens the browser, listens on the configured localhost callback, exchanges the auth code for tokens, and stores OAuth credentials in `~/.claude/credentials.json` (or `$CLAUDE_CONFIG_HOME/credentials.json`).
### Logout
```bash
cd rust
cargo run -p rusty-claude-cli -- logout
```
This removes only the stored OAuth credentials and preserves unrelated JSON fields in `credentials.json`.
### Self-update
```bash
cd rust
cargo run -p rusty-claude-cli -- self-update
```
The command checks the latest GitHub release for `instructkr/clawd-code`, compares it to the current binary version, downloads the matching binary asset plus checksum manifest, verifies SHA-256, replaces the current executable, and prints the release changelog. If no published release or matching asset exists, it exits safely with an explanatory message.
## Usage examples
### 1) Prompt mode
Send one prompt, stream the answer, then exit:
```bash
cd rust
cargo run -p rusty-claude-cli -- prompt "Summarize the architecture of this repository"
```
Use a specific model:
```bash
cd rust
cargo run -p rusty-claude-cli -- --model claude-sonnet-4-20250514 prompt "List the key crates in this workspace"
```
Restrict enabled tools in an interactive session:
```bash
cd rust
cargo run -p rusty-claude-cli -- --allowedTools read,glob
```
Bootstrap Claude project files for the current repo:
```bash
cd rust
cargo run -p rusty-claude-cli -- init
```
### 2) REPL mode
Start the interactive shell:
```bash
cd rust
cargo run -p rusty-claude-cli --
```
Inside the REPL, useful commands include:
```text
/help
/status
/model claude-sonnet-4-20250514
/permissions workspace-write
/cost
/compact
/memory
/config
/init
/diff
/version
/export notes.txt
/sessions
/session list
/exit
```
### 3) Resume an existing session
Inspect or maintain a saved session file without entering the REPL:
```bash
cd rust
cargo run -p rusty-claude-cli -- --resume session-123456 /status /compact /cost
```
You can also inspect memory/config state for a restored session:
```bash
cd rust
cargo run -p rusty-claude-cli -- --resume ~/.claude/sessions/session-123456.json /memory /config
```
## Available commands
### Top-level CLI commands
- `prompt <text...>` — run one prompt non-interactively
- `--resume <session-id-or-path> [/commands...]` — inspect or maintain a saved session stored under `~/.claude/sessions/`
- `dump-manifests` — print extracted upstream manifest counts
- `bootstrap-plan` — print the current bootstrap skeleton
- `system-prompt [--cwd PATH] [--date YYYY-MM-DD]` — render the synthesized system prompt
- `self-update` — update the installed binary from the latest GitHub release when a matching asset is available
- `--help` / `-h` — show CLI help
- `--version` / `-V` — print the CLI version and build info locally (no API call)
- `--output-format text|json` — choose non-interactive prompt output rendering
- `--allowedTools <tool[,tool...]>` — restrict enabled tools for interactive sessions and prompt-mode tool use
### Interactive slash commands
- `/help` — show command help
- `/status` — show current session status
- `/compact` — compact local session history
- `/model [model]` — inspect or switch the active model
- `/permissions [read-only|workspace-write|danger-full-access]` — inspect or switch permissions
- `/clear [--confirm]` — clear the current local session
- `/cost` — show token usage totals
- `/resume <session-id-or-path>` — load a saved session into the REPL
- `/config [env|hooks|model]` — inspect discovered Claude config
- `/memory` — inspect loaded instruction memory files
- `/init` — bootstrap `.claude.json`, `.claude/`, `CLAUDE.md`, and local ignore rules
- `/diff` — show the current git diff for the workspace
- `/version` — print version and build metadata locally
- `/export [file]` — export the current conversation transcript
- `/sessions` — list recent managed local sessions from `~/.claude/sessions/`
- `/session [list|switch <session-id>]` — inspect or switch managed local sessions
- `/exit` — leave the REPL
## Environment variables
### Anthropic/API
- `ANTHROPIC_API_KEY` — highest-precedence API credential
- `ANTHROPIC_AUTH_TOKEN` — bearer-token override used when no API key is set
- Persisted OAuth credentials in `~/.claude/credentials.json` — used when neither env var is set
- `ANTHROPIC_BASE_URL` — override the Anthropic API base URL
- `ANTHROPIC_MODEL` — default model used by selected live integration tests
### CLI/runtime
- `RUSTY_CLAUDE_PERMISSION_MODE` — default REPL permission mode (`read-only`, `workspace-write`, or `danger-full-access`)
- `CLAUDE_CONFIG_HOME` — override Claude config discovery root
- `CLAUDE_CODE_REMOTE` — enable remote-session bootstrap handling when supported
- `CLAUDE_CODE_REMOTE_SESSION_ID` — remote session identifier when using remote mode
- `CLAUDE_CODE_UPSTREAM` — override the upstream TS source path for compat-harness extraction
- `CLAWD_WEB_SEARCH_BASE_URL` — override the built-in web search service endpoint used by tooling
## Notes
- `compat-harness` exists to compare the Rust port against the upstream TypeScript codebase and is intentionally excluded from the requested release test run.
- The CLI currently focuses on a practical integrated workflow: prompt execution, REPL operation, session inspection/resume, config discovery, and tool/runtime plumbing.
See repository root.

File diff suppressed because it is too large Load Diff

View File

@@ -4,7 +4,10 @@ use std::time::Duration;
#[derive(Debug)]
pub enum ApiError {
MissingApiKey,
MissingCredentials {
provider: &'static str,
env_vars: &'static [&'static str],
},
ExpiredOAuthToken,
Auth(String),
InvalidApiKeyEnv(VarError),
@@ -30,13 +33,21 @@ pub enum ApiError {
}
impl ApiError {
#[must_use]
pub const fn missing_credentials(
provider: &'static str,
env_vars: &'static [&'static str],
) -> Self {
Self::MissingCredentials { provider, env_vars }
}
#[must_use]
pub fn is_retryable(&self) -> bool {
match self {
Self::Http(error) => error.is_connect() || error.is_timeout() || error.is_request(),
Self::Api { retryable, .. } => *retryable,
Self::RetriesExhausted { last_error, .. } => last_error.is_retryable(),
Self::MissingApiKey
Self::MissingCredentials { .. }
| Self::ExpiredOAuthToken
| Self::Auth(_)
| Self::InvalidApiKeyEnv(_)
@@ -51,12 +62,11 @@ impl ApiError {
impl Display for ApiError {
fn fmt(&self, f: &mut Formatter<'_>) -> std::fmt::Result {
match self {
Self::MissingApiKey => {
write!(
Self::MissingCredentials { provider, env_vars } => write!(
f,
"ANTHROPIC_AUTH_TOKEN or ANTHROPIC_API_KEY is not set; export one before calling the Anthropic API"
)
}
"missing {provider} credentials; export {} before calling the {provider} API",
env_vars.join(" or ")
),
Self::ExpiredOAuthToken => {
write!(
f,
@@ -65,10 +75,7 @@ impl Display for ApiError {
}
Self::Auth(message) => write!(f, "auth error: {message}"),
Self::InvalidApiKeyEnv(error) => {
write!(
f,
"failed to read ANTHROPIC_AUTH_TOKEN / ANTHROPIC_API_KEY: {error}"
)
write!(f, "failed to read credential environment variable: {error}")
}
Self::Http(error) => write!(f, "http error: {error}"),
Self::Io(error) => write!(f, "io error: {error}"),
@@ -81,20 +88,14 @@ impl Display for ApiError {
..
} => match (error_type, message) {
(Some(error_type), Some(message)) => {
write!(
f,
"anthropic api returned {status} ({error_type}): {message}"
)
write!(f, "api returned {status} ({error_type}): {message}")
}
_ => write!(f, "anthropic api returned {status}: {body}"),
_ => write!(f, "api returned {status}: {body}"),
},
Self::RetriesExhausted {
attempts,
last_error,
} => write!(
f,
"anthropic api failed after {attempts} attempts: {last_error}"
),
} => write!(f, "api failed after {attempts} attempts: {last_error}"),
Self::InvalidSseFrame(message) => write!(f, "invalid sse frame: {message}"),
Self::BackoffOverflow {
attempt,

View File

@@ -1,13 +1,19 @@
mod client;
mod error;
mod providers;
mod sse;
mod types;
pub use client::{
oauth_token_is_expired, read_base_url, resolve_saved_oauth_token,
resolve_startup_auth_source, AnthropicClient, AuthSource, MessageStream, OAuthTokenSet,
oauth_token_is_expired, read_base_url, read_xai_base_url, resolve_saved_oauth_token,
resolve_startup_auth_source, MessageStream, OAuthTokenSet, ProviderClient,
};
pub use error::ApiError;
pub use providers::anthropic::{AnthropicClient, AuthSource};
pub use providers::openai_compat::{OpenAiCompatClient, OpenAiCompatConfig};
pub use providers::{
detect_provider_kind, max_tokens_for_model, resolve_model_alias, ProviderKind,
};
pub use sse::{parse_frame, SseParser};
pub use types::{
ContentBlockDelta, ContentBlockDeltaEvent, ContentBlockStartEvent, ContentBlockStopEvent,

File diff suppressed because it is too large Load Diff

View File

@@ -0,0 +1,216 @@
use std::future::Future;
use std::pin::Pin;
use crate::error::ApiError;
use crate::types::{MessageRequest, MessageResponse};
pub mod anthropic;
pub mod openai_compat;
pub type ProviderFuture<'a, T> = Pin<Box<dyn Future<Output = Result<T, ApiError>> + Send + 'a>>;
pub trait Provider {
type Stream;
fn send_message<'a>(
&'a self,
request: &'a MessageRequest,
) -> ProviderFuture<'a, MessageResponse>;
fn stream_message<'a>(
&'a self,
request: &'a MessageRequest,
) -> ProviderFuture<'a, Self::Stream>;
}
#[derive(Debug, Clone, Copy, PartialEq, Eq)]
pub enum ProviderKind {
Anthropic,
Xai,
OpenAi,
}
#[derive(Debug, Clone, Copy, PartialEq, Eq)]
pub struct ProviderMetadata {
pub provider: ProviderKind,
pub auth_env: &'static str,
pub base_url_env: &'static str,
pub default_base_url: &'static str,
}
const MODEL_REGISTRY: &[(&str, ProviderMetadata)] = &[
(
"opus",
ProviderMetadata {
provider: ProviderKind::Anthropic,
auth_env: "ANTHROPIC_API_KEY",
base_url_env: "ANTHROPIC_BASE_URL",
default_base_url: anthropic::DEFAULT_BASE_URL,
},
),
(
"sonnet",
ProviderMetadata {
provider: ProviderKind::Anthropic,
auth_env: "ANTHROPIC_API_KEY",
base_url_env: "ANTHROPIC_BASE_URL",
default_base_url: anthropic::DEFAULT_BASE_URL,
},
),
(
"haiku",
ProviderMetadata {
provider: ProviderKind::Anthropic,
auth_env: "ANTHROPIC_API_KEY",
base_url_env: "ANTHROPIC_BASE_URL",
default_base_url: anthropic::DEFAULT_BASE_URL,
},
),
(
"grok",
ProviderMetadata {
provider: ProviderKind::Xai,
auth_env: "XAI_API_KEY",
base_url_env: "XAI_BASE_URL",
default_base_url: openai_compat::DEFAULT_XAI_BASE_URL,
},
),
(
"grok-3",
ProviderMetadata {
provider: ProviderKind::Xai,
auth_env: "XAI_API_KEY",
base_url_env: "XAI_BASE_URL",
default_base_url: openai_compat::DEFAULT_XAI_BASE_URL,
},
),
(
"grok-mini",
ProviderMetadata {
provider: ProviderKind::Xai,
auth_env: "XAI_API_KEY",
base_url_env: "XAI_BASE_URL",
default_base_url: openai_compat::DEFAULT_XAI_BASE_URL,
},
),
(
"grok-3-mini",
ProviderMetadata {
provider: ProviderKind::Xai,
auth_env: "XAI_API_KEY",
base_url_env: "XAI_BASE_URL",
default_base_url: openai_compat::DEFAULT_XAI_BASE_URL,
},
),
(
"grok-2",
ProviderMetadata {
provider: ProviderKind::Xai,
auth_env: "XAI_API_KEY",
base_url_env: "XAI_BASE_URL",
default_base_url: openai_compat::DEFAULT_XAI_BASE_URL,
},
),
];
#[must_use]
pub fn resolve_model_alias(model: &str) -> String {
let trimmed = model.trim();
let lower = trimmed.to_ascii_lowercase();
MODEL_REGISTRY
.iter()
.find_map(|(alias, metadata)| {
(*alias == lower).then_some(match metadata.provider {
ProviderKind::Anthropic => match *alias {
"opus" => "claude-opus-4-6",
"sonnet" => "claude-sonnet-4-6",
"haiku" => "claude-haiku-4-5-20251213",
_ => trimmed,
},
ProviderKind::Xai => match *alias {
"grok" | "grok-3" => "grok-3",
"grok-mini" | "grok-3-mini" => "grok-3-mini",
"grok-2" => "grok-2",
_ => trimmed,
},
ProviderKind::OpenAi => trimmed,
})
})
.map_or_else(|| trimmed.to_string(), ToOwned::to_owned)
}
#[must_use]
pub fn metadata_for_model(model: &str) -> Option<ProviderMetadata> {
let canonical = resolve_model_alias(model);
if canonical.starts_with("claude") {
return Some(ProviderMetadata {
provider: ProviderKind::Anthropic,
auth_env: "ANTHROPIC_API_KEY",
base_url_env: "ANTHROPIC_BASE_URL",
default_base_url: anthropic::DEFAULT_BASE_URL,
});
}
if canonical.starts_with("grok") {
return Some(ProviderMetadata {
provider: ProviderKind::Xai,
auth_env: "XAI_API_KEY",
base_url_env: "XAI_BASE_URL",
default_base_url: openai_compat::DEFAULT_XAI_BASE_URL,
});
}
None
}
#[must_use]
pub fn detect_provider_kind(model: &str) -> ProviderKind {
if let Some(metadata) = metadata_for_model(model) {
return metadata.provider;
}
if anthropic::has_auth_from_env_or_saved().unwrap_or(false) {
return ProviderKind::Anthropic;
}
if openai_compat::has_api_key("OPENAI_API_KEY") {
return ProviderKind::OpenAi;
}
if openai_compat::has_api_key("XAI_API_KEY") {
return ProviderKind::Xai;
}
ProviderKind::Anthropic
}
#[must_use]
pub fn max_tokens_for_model(model: &str) -> u32 {
let canonical = resolve_model_alias(model);
if canonical.contains("opus") {
32_000
} else {
64_000
}
}
#[cfg(test)]
mod tests {
use super::{detect_provider_kind, max_tokens_for_model, resolve_model_alias, ProviderKind};
#[test]
fn resolves_grok_aliases() {
assert_eq!(resolve_model_alias("grok"), "grok-3");
assert_eq!(resolve_model_alias("grok-mini"), "grok-3-mini");
assert_eq!(resolve_model_alias("grok-2"), "grok-2");
}
#[test]
fn detects_provider_from_model_name_first() {
assert_eq!(detect_provider_kind("grok"), ProviderKind::Xai);
assert_eq!(
detect_provider_kind("claude-sonnet-4-6"),
ProviderKind::Anthropic
);
}
#[test]
fn keeps_existing_max_token_heuristic() {
assert_eq!(max_tokens_for_model("opus"), 32_000);
assert_eq!(max_tokens_for_model("grok-3"), 64_000);
}
}

File diff suppressed because it is too large Load Diff

View File

@@ -3,9 +3,9 @@ use std::sync::Arc;
use std::time::Duration;
use api::{
AnthropicClient, ApiError, ContentBlockDelta, ContentBlockDeltaEvent, ContentBlockStartEvent,
InputContentBlock, InputMessage, MessageDeltaEvent, MessageRequest, OutputContentBlock,
StreamEvent, ToolChoice, ToolDefinition,
AnthropicClient, ApiError, AuthSource, ContentBlockDelta, ContentBlockDeltaEvent,
ContentBlockStartEvent, InputContentBlock, InputMessage, MessageDeltaEvent, MessageRequest,
OutputContentBlock, ProviderClient, StreamEvent, ToolChoice, ToolDefinition,
};
use serde_json::json;
use tokio::io::{AsyncReadExt, AsyncWriteExt};
@@ -195,6 +195,47 @@ async fn retries_retryable_failures_before_succeeding() {
assert_eq!(state.lock().await.len(), 2);
}
#[tokio::test]
async fn provider_client_dispatches_anthropic_requests() {
let state = Arc::new(Mutex::new(Vec::<CapturedRequest>::new()));
let server = spawn_server(
state.clone(),
vec![http_response(
"200 OK",
"application/json",
"{\"id\":\"msg_provider\",\"type\":\"message\",\"role\":\"assistant\",\"content\":[{\"type\":\"text\",\"text\":\"Dispatched\"}],\"model\":\"claude-3-7-sonnet-latest\",\"stop_reason\":\"end_turn\",\"stop_sequence\":null,\"usage\":{\"input_tokens\":3,\"output_tokens\":2}}",
)],
)
.await;
let client = ProviderClient::from_model_with_anthropic_auth(
"claude-sonnet-4-6",
Some(AuthSource::ApiKey("test-key".to_string())),
)
.expect("anthropic provider client should be constructed");
let client = match client {
ProviderClient::Anthropic(client) => {
ProviderClient::Anthropic(client.with_base_url(server.base_url()))
}
other => panic!("expected anthropic provider, got {other:?}"),
};
let response = client
.send_message(&sample_request(false))
.await
.expect("provider-dispatched request should succeed");
assert_eq!(response.total_tokens(), 5);
let captured = state.lock().await;
let request = captured.first().expect("server should capture request");
assert_eq!(request.path, "/v1/messages");
assert_eq!(
request.headers.get("x-api-key").map(String::as_str),
Some("test-key")
);
}
#[tokio::test]
async fn surfaces_retry_exhaustion_for_persistent_retryable_errors() {
let state = Arc::new(Mutex::new(Vec::<CapturedRequest>::new()));

View File

@@ -0,0 +1,415 @@
use std::collections::HashMap;
use std::ffi::OsString;
use std::sync::Arc;
use std::sync::{Mutex as StdMutex, OnceLock};
use api::{
ContentBlockDelta, ContentBlockDeltaEvent, ContentBlockStartEvent, ContentBlockStopEvent,
InputContentBlock, InputMessage, MessageRequest, OpenAiCompatClient, OpenAiCompatConfig,
OutputContentBlock, ProviderClient, StreamEvent, ToolChoice, ToolDefinition,
};
use serde_json::json;
use tokio::io::{AsyncReadExt, AsyncWriteExt};
use tokio::net::TcpListener;
use tokio::sync::Mutex;
#[tokio::test]
async fn send_message_uses_openai_compatible_endpoint_and_auth() {
let state = Arc::new(Mutex::new(Vec::<CapturedRequest>::new()));
let body = concat!(
"{",
"\"id\":\"chatcmpl_test\",",
"\"model\":\"grok-3\",",
"\"choices\":[{",
"\"message\":{\"role\":\"assistant\",\"content\":\"Hello from Grok\",\"tool_calls\":[]},",
"\"finish_reason\":\"stop\"",
"}],",
"\"usage\":{\"prompt_tokens\":11,\"completion_tokens\":5}",
"}"
);
let server = spawn_server(
state.clone(),
vec![http_response("200 OK", "application/json", body)],
)
.await;
let client = OpenAiCompatClient::new("xai-test-key", OpenAiCompatConfig::xai())
.with_base_url(server.base_url());
let response = client
.send_message(&sample_request(false))
.await
.expect("request should succeed");
assert_eq!(response.model, "grok-3");
assert_eq!(response.total_tokens(), 16);
assert_eq!(
response.content,
vec![OutputContentBlock::Text {
text: "Hello from Grok".to_string(),
}]
);
let captured = state.lock().await;
let request = captured.first().expect("server should capture request");
assert_eq!(request.path, "/chat/completions");
assert_eq!(
request.headers.get("authorization").map(String::as_str),
Some("Bearer xai-test-key")
);
let body: serde_json::Value = serde_json::from_str(&request.body).expect("json body");
assert_eq!(body["model"], json!("grok-3"));
assert_eq!(body["messages"][0]["role"], json!("system"));
assert_eq!(body["tools"][0]["type"], json!("function"));
}
#[tokio::test]
async fn send_message_accepts_full_chat_completions_endpoint_override() {
let state = Arc::new(Mutex::new(Vec::<CapturedRequest>::new()));
let body = concat!(
"{",
"\"id\":\"chatcmpl_full_endpoint\",",
"\"model\":\"grok-3\",",
"\"choices\":[{",
"\"message\":{\"role\":\"assistant\",\"content\":\"Endpoint override works\",\"tool_calls\":[]},",
"\"finish_reason\":\"stop\"",
"}],",
"\"usage\":{\"prompt_tokens\":7,\"completion_tokens\":3}",
"}"
);
let server = spawn_server(
state.clone(),
vec![http_response("200 OK", "application/json", body)],
)
.await;
let endpoint_url = format!("{}/chat/completions", server.base_url());
let client = OpenAiCompatClient::new("xai-test-key", OpenAiCompatConfig::xai())
.with_base_url(endpoint_url);
let response = client
.send_message(&sample_request(false))
.await
.expect("request should succeed");
assert_eq!(response.total_tokens(), 10);
let captured = state.lock().await;
let request = captured.first().expect("server should capture request");
assert_eq!(request.path, "/chat/completions");
}
#[tokio::test]
async fn stream_message_normalizes_text_and_multiple_tool_calls() {
let state = Arc::new(Mutex::new(Vec::<CapturedRequest>::new()));
let sse = concat!(
"data: {\"id\":\"chatcmpl_stream\",\"model\":\"grok-3\",\"choices\":[{\"delta\":{\"content\":\"Hello\"}}]}\n\n",
"data: {\"id\":\"chatcmpl_stream\",\"choices\":[{\"delta\":{\"tool_calls\":[{\"index\":0,\"id\":\"call_1\",\"function\":{\"name\":\"weather\",\"arguments\":\"{\\\"city\\\":\\\"Paris\\\"}\"}},{\"index\":1,\"id\":\"call_2\",\"function\":{\"name\":\"clock\",\"arguments\":\"{\\\"zone\\\":\\\"UTC\\\"}\"}}]}}]}\n\n",
"data: {\"id\":\"chatcmpl_stream\",\"choices\":[{\"delta\":{},\"finish_reason\":\"tool_calls\"}]}\n\n",
"data: [DONE]\n\n"
);
let server = spawn_server(
state.clone(),
vec![http_response_with_headers(
"200 OK",
"text/event-stream",
sse,
&[("x-request-id", "req_grok_stream")],
)],
)
.await;
let client = OpenAiCompatClient::new("xai-test-key", OpenAiCompatConfig::xai())
.with_base_url(server.base_url());
let mut stream = client
.stream_message(&sample_request(false))
.await
.expect("stream should start");
assert_eq!(stream.request_id(), Some("req_grok_stream"));
let mut events = Vec::new();
while let Some(event) = stream.next_event().await.expect("event should parse") {
events.push(event);
}
assert!(matches!(events[0], StreamEvent::MessageStart(_)));
assert!(matches!(
events[1],
StreamEvent::ContentBlockStart(ContentBlockStartEvent {
content_block: OutputContentBlock::Text { .. },
..
})
));
assert!(matches!(
events[2],
StreamEvent::ContentBlockDelta(ContentBlockDeltaEvent {
delta: ContentBlockDelta::TextDelta { .. },
..
})
));
assert!(matches!(
events[3],
StreamEvent::ContentBlockStart(ContentBlockStartEvent {
index: 1,
content_block: OutputContentBlock::ToolUse { .. },
})
));
assert!(matches!(
events[4],
StreamEvent::ContentBlockDelta(ContentBlockDeltaEvent {
index: 1,
delta: ContentBlockDelta::InputJsonDelta { .. },
})
));
assert!(matches!(
events[5],
StreamEvent::ContentBlockStart(ContentBlockStartEvent {
index: 2,
content_block: OutputContentBlock::ToolUse { .. },
})
));
assert!(matches!(
events[6],
StreamEvent::ContentBlockDelta(ContentBlockDeltaEvent {
index: 2,
delta: ContentBlockDelta::InputJsonDelta { .. },
})
));
assert!(matches!(
events[7],
StreamEvent::ContentBlockStop(ContentBlockStopEvent { index: 1 })
));
assert!(matches!(
events[8],
StreamEvent::ContentBlockStop(ContentBlockStopEvent { index: 2 })
));
assert!(matches!(
events[9],
StreamEvent::ContentBlockStop(ContentBlockStopEvent { index: 0 })
));
assert!(matches!(events[10], StreamEvent::MessageDelta(_)));
assert!(matches!(events[11], StreamEvent::MessageStop(_)));
let captured = state.lock().await;
let request = captured.first().expect("captured request");
assert_eq!(request.path, "/chat/completions");
assert!(request.body.contains("\"stream\":true"));
}
#[tokio::test]
async fn provider_client_dispatches_xai_requests_from_env() {
let _lock = env_lock();
let _api_key = ScopedEnvVar::set("XAI_API_KEY", "xai-test-key");
let state = Arc::new(Mutex::new(Vec::<CapturedRequest>::new()));
let server = spawn_server(
state.clone(),
vec![http_response(
"200 OK",
"application/json",
"{\"id\":\"chatcmpl_provider\",\"model\":\"grok-3\",\"choices\":[{\"message\":{\"role\":\"assistant\",\"content\":\"Through provider client\",\"tool_calls\":[]},\"finish_reason\":\"stop\"}],\"usage\":{\"prompt_tokens\":9,\"completion_tokens\":4}}",
)],
)
.await;
let _base_url = ScopedEnvVar::set("XAI_BASE_URL", server.base_url());
let client =
ProviderClient::from_model("grok").expect("xAI provider client should be constructed");
assert!(matches!(client, ProviderClient::Xai(_)));
let response = client
.send_message(&sample_request(false))
.await
.expect("provider-dispatched request should succeed");
assert_eq!(response.total_tokens(), 13);
let captured = state.lock().await;
let request = captured.first().expect("captured request");
assert_eq!(request.path, "/chat/completions");
assert_eq!(
request.headers.get("authorization").map(String::as_str),
Some("Bearer xai-test-key")
);
}
#[derive(Debug, Clone, PartialEq, Eq)]
struct CapturedRequest {
path: String,
headers: HashMap<String, String>,
body: String,
}
struct TestServer {
base_url: String,
join_handle: tokio::task::JoinHandle<()>,
}
impl TestServer {
fn base_url(&self) -> String {
self.base_url.clone()
}
}
impl Drop for TestServer {
fn drop(&mut self) {
self.join_handle.abort();
}
}
async fn spawn_server(
state: Arc<Mutex<Vec<CapturedRequest>>>,
responses: Vec<String>,
) -> TestServer {
let listener = TcpListener::bind("127.0.0.1:0")
.await
.expect("listener should bind");
let address = listener.local_addr().expect("listener addr");
let join_handle = tokio::spawn(async move {
for response in responses {
let (mut socket, _) = listener.accept().await.expect("accept");
let mut buffer = Vec::new();
let mut header_end = None;
loop {
let mut chunk = [0_u8; 1024];
let read = socket.read(&mut chunk).await.expect("read request");
if read == 0 {
break;
}
buffer.extend_from_slice(&chunk[..read]);
if let Some(position) = find_header_end(&buffer) {
header_end = Some(position);
break;
}
}
let header_end = header_end.expect("headers should exist");
let (header_bytes, remaining) = buffer.split_at(header_end);
let header_text = String::from_utf8(header_bytes.to_vec()).expect("utf8 headers");
let mut lines = header_text.split("\r\n");
let request_line = lines.next().expect("request line");
let path = request_line
.split_whitespace()
.nth(1)
.expect("path")
.to_string();
let mut headers = HashMap::new();
let mut content_length = 0_usize;
for line in lines {
if line.is_empty() {
continue;
}
let (name, value) = line.split_once(':').expect("header");
let value = value.trim().to_string();
if name.eq_ignore_ascii_case("content-length") {
content_length = value.parse().expect("content length");
}
headers.insert(name.to_ascii_lowercase(), value);
}
let mut body = remaining[4..].to_vec();
while body.len() < content_length {
let mut chunk = vec![0_u8; content_length - body.len()];
let read = socket.read(&mut chunk).await.expect("read body");
if read == 0 {
break;
}
body.extend_from_slice(&chunk[..read]);
}
state.lock().await.push(CapturedRequest {
path,
headers,
body: String::from_utf8(body).expect("utf8 body"),
});
socket
.write_all(response.as_bytes())
.await
.expect("write response");
}
});
TestServer {
base_url: format!("http://{address}"),
join_handle,
}
}
fn find_header_end(bytes: &[u8]) -> Option<usize> {
bytes.windows(4).position(|window| window == b"\r\n\r\n")
}
fn http_response(status: &str, content_type: &str, body: &str) -> String {
http_response_with_headers(status, content_type, body, &[])
}
fn http_response_with_headers(
status: &str,
content_type: &str,
body: &str,
headers: &[(&str, &str)],
) -> String {
let mut extra_headers = String::new();
for (name, value) in headers {
use std::fmt::Write as _;
write!(&mut extra_headers, "{name}: {value}\r\n").expect("header write");
}
format!(
"HTTP/1.1 {status}\r\ncontent-type: {content_type}\r\n{extra_headers}content-length: {}\r\nconnection: close\r\n\r\n{body}",
body.len()
)
}
fn sample_request(stream: bool) -> MessageRequest {
MessageRequest {
model: "grok-3".to_string(),
max_tokens: 64,
messages: vec![InputMessage {
role: "user".to_string(),
content: vec![InputContentBlock::Text {
text: "Say hello".to_string(),
}],
}],
system: Some("Use tools when needed".to_string()),
tools: Some(vec![ToolDefinition {
name: "weather".to_string(),
description: Some("Fetches weather".to_string()),
input_schema: json!({
"type": "object",
"properties": {"city": {"type": "string"}},
"required": ["city"]
}),
}]),
tool_choice: Some(ToolChoice::Auto),
stream,
}
}
fn env_lock() -> std::sync::MutexGuard<'static, ()> {
static LOCK: OnceLock<StdMutex<()>> = OnceLock::new();
LOCK.get_or_init(|| StdMutex::new(()))
.lock()
.unwrap_or_else(|poisoned| poisoned.into_inner())
}
struct ScopedEnvVar {
key: &'static str,
previous: Option<OsString>,
}
impl ScopedEnvVar {
fn set(key: &'static str, value: impl AsRef<std::ffi::OsStr>) -> Self {
let previous = std::env::var_os(key);
std::env::set_var(key, value);
Self { key, previous }
}
}
impl Drop for ScopedEnvVar {
fn drop(&mut self) {
match &self.previous {
Some(value) => std::env::set_var(self.key, value),
None => std::env::remove_var(self.key),
}
}
}

View File

@@ -0,0 +1,86 @@
use std::ffi::OsString;
use std::sync::{Mutex, OnceLock};
use api::{read_xai_base_url, ApiError, AuthSource, ProviderClient, ProviderKind};
#[test]
fn provider_client_routes_grok_aliases_through_xai() {
let _lock = env_lock();
let _xai_api_key = EnvVarGuard::set("XAI_API_KEY", Some("xai-test-key"));
let client = ProviderClient::from_model("grok-mini").expect("grok alias should resolve");
assert_eq!(client.provider_kind(), ProviderKind::Xai);
}
#[test]
fn provider_client_reports_missing_xai_credentials_for_grok_models() {
let _lock = env_lock();
let _xai_api_key = EnvVarGuard::set("XAI_API_KEY", None);
let error = ProviderClient::from_model("grok-3")
.expect_err("grok requests without XAI_API_KEY should fail fast");
match error {
ApiError::MissingCredentials { provider, env_vars } => {
assert_eq!(provider, "xAI");
assert_eq!(env_vars, &["XAI_API_KEY"]);
}
other => panic!("expected missing xAI credentials, got {other:?}"),
}
}
#[test]
fn provider_client_uses_explicit_anthropic_auth_without_env_lookup() {
let _lock = env_lock();
let _anthropic_api_key = EnvVarGuard::set("ANTHROPIC_API_KEY", None);
let _anthropic_auth_token = EnvVarGuard::set("ANTHROPIC_AUTH_TOKEN", None);
let client = ProviderClient::from_model_with_anthropic_auth(
"claude-sonnet-4-6",
Some(AuthSource::ApiKey("anthropic-test-key".to_string())),
)
.expect("explicit anthropic auth should avoid env lookup");
assert_eq!(client.provider_kind(), ProviderKind::Anthropic);
}
#[test]
fn read_xai_base_url_prefers_env_override() {
let _lock = env_lock();
let _xai_base_url = EnvVarGuard::set("XAI_BASE_URL", Some("https://example.xai.test/v1"));
assert_eq!(read_xai_base_url(), "https://example.xai.test/v1");
}
fn env_lock() -> std::sync::MutexGuard<'static, ()> {
static LOCK: OnceLock<Mutex<()>> = OnceLock::new();
LOCK.get_or_init(|| Mutex::new(()))
.lock()
.unwrap_or_else(|poisoned| poisoned.into_inner())
}
struct EnvVarGuard {
key: &'static str,
original: Option<OsString>,
}
impl EnvVarGuard {
fn set(key: &'static str, value: Option<&str>) -> Self {
let original = std::env::var_os(key);
match value {
Some(value) => std::env::set_var(key, value),
None => std::env::remove_var(key),
}
Self { key, original }
}
}
impl Drop for EnvVarGuard {
fn drop(&mut self) {
match &self.original {
Some(value) => std::env::set_var(self.key, value),
None => std::env::remove_var(self.key),
}
}
}

View File

@@ -37,6 +37,7 @@ pub struct RuntimeConfig {
#[derive(Debug, Clone, PartialEq, Eq, Default)]
pub struct RuntimeFeatureConfig {
hooks: RuntimeHookConfig,
mcp: McpConfigCollection,
oauth: Option<OAuthConfig>,
model: Option<String>,
@@ -44,6 +45,12 @@ pub struct RuntimeFeatureConfig {
sandbox: SandboxConfig,
}
#[derive(Debug, Clone, PartialEq, Eq, Default)]
pub struct RuntimeHookConfig {
pre_tool_use: Vec<String>,
post_tool_use: Vec<String>,
}
#[derive(Debug, Clone, PartialEq, Eq, Default)]
pub struct McpConfigCollection {
servers: BTreeMap<String, ScopedMcpServerConfig>,
@@ -221,6 +228,7 @@ impl ConfigLoader {
let merged_value = JsonValue::Object(merged.clone());
let feature_config = RuntimeFeatureConfig {
hooks: parse_optional_hooks_config(&merged_value)?,
mcp: McpConfigCollection {
servers: mcp_servers,
},
@@ -278,6 +286,11 @@ impl RuntimeConfig {
&self.feature_config.mcp
}
#[must_use]
pub fn hooks(&self) -> &RuntimeHookConfig {
&self.feature_config.hooks
}
#[must_use]
pub fn oauth(&self) -> Option<&OAuthConfig> {
self.feature_config.oauth.as_ref()
@@ -300,6 +313,17 @@ impl RuntimeConfig {
}
impl RuntimeFeatureConfig {
#[must_use]
pub fn with_hooks(mut self, hooks: RuntimeHookConfig) -> Self {
self.hooks = hooks;
self
}
#[must_use]
pub fn hooks(&self) -> &RuntimeHookConfig {
&self.hooks
}
#[must_use]
pub fn mcp(&self) -> &McpConfigCollection {
&self.mcp
@@ -326,6 +350,26 @@ impl RuntimeFeatureConfig {
}
}
impl RuntimeHookConfig {
#[must_use]
pub fn new(pre_tool_use: Vec<String>, post_tool_use: Vec<String>) -> Self {
Self {
pre_tool_use,
post_tool_use,
}
}
#[must_use]
pub fn pre_tool_use(&self) -> &[String] {
&self.pre_tool_use
}
#[must_use]
pub fn post_tool_use(&self) -> &[String] {
&self.post_tool_use
}
}
impl McpConfigCollection {
#[must_use]
pub fn servers(&self) -> &BTreeMap<String, ScopedMcpServerConfig> {
@@ -424,6 +468,22 @@ fn parse_optional_model(root: &JsonValue) -> Option<String> {
.map(ToOwned::to_owned)
}
fn parse_optional_hooks_config(root: &JsonValue) -> Result<RuntimeHookConfig, ConfigError> {
let Some(object) = root.as_object() else {
return Ok(RuntimeHookConfig::default());
};
let Some(hooks_value) = object.get("hooks") else {
return Ok(RuntimeHookConfig::default());
};
let hooks = expect_object(hooks_value, "merged settings.hooks")?;
Ok(RuntimeHookConfig {
pre_tool_use: optional_string_array(hooks, "PreToolUse", "merged settings.hooks")?
.unwrap_or_default(),
post_tool_use: optional_string_array(hooks, "PostToolUse", "merged settings.hooks")?
.unwrap_or_default(),
})
}
fn parse_optional_permission_mode(
root: &JsonValue,
) -> Result<Option<ResolvedPermissionMode>, ConfigError> {
@@ -836,6 +896,8 @@ mod tests {
.and_then(JsonValue::as_object)
.expect("hooks object")
.contains_key("PostToolUse"));
assert_eq!(loaded.hooks().pre_tool_use(), &["base".to_string()]);
assert_eq!(loaded.hooks().post_tool_use(), &["project".to_string()]);
assert!(loaded.mcp().get("home").is_some());
assert!(loaded.mcp().get("project").is_some());

View File

@@ -4,6 +4,8 @@ use std::fmt::{Display, Formatter};
use crate::compact::{
compact_session, estimate_session_tokens, CompactionConfig, CompactionResult,
};
use crate::config::RuntimeFeatureConfig;
use crate::hooks::{HookRunResult, HookRunner};
use crate::permissions::{PermissionOutcome, PermissionPolicy, PermissionPrompter};
use crate::session::{ContentBlock, ConversationMessage, Session};
use crate::usage::{TokenUsage, UsageTracker};
@@ -94,6 +96,7 @@ pub struct ConversationRuntime<C, T> {
system_prompt: Vec<String>,
max_iterations: usize,
usage_tracker: UsageTracker,
hook_runner: HookRunner,
}
impl<C, T> ConversationRuntime<C, T>
@@ -108,6 +111,25 @@ where
tool_executor: T,
permission_policy: PermissionPolicy,
system_prompt: Vec<String>,
) -> Self {
Self::new_with_features(
session,
api_client,
tool_executor,
permission_policy,
system_prompt,
&RuntimeFeatureConfig::default(),
)
}
#[must_use]
pub fn new_with_features(
session: Session,
api_client: C,
tool_executor: T,
permission_policy: PermissionPolicy,
system_prompt: Vec<String>,
feature_config: &RuntimeFeatureConfig,
) -> Self {
let usage_tracker = UsageTracker::from_session(&session);
Self {
@@ -118,6 +140,7 @@ where
system_prompt,
max_iterations: usize::MAX,
usage_tracker,
hook_runner: HookRunner::from_feature_config(feature_config),
}
}
@@ -185,19 +208,41 @@ where
let result_message = match permission_outcome {
PermissionOutcome::Allow => {
let pre_hook_result = self.hook_runner.run_pre_tool_use(&tool_name, &input);
if pre_hook_result.is_denied() {
let deny_message = format!("PreToolUse hook denied tool `{tool_name}`");
ConversationMessage::tool_result(
tool_use_id,
tool_name,
format_hook_message(&pre_hook_result, &deny_message),
true,
)
} else {
let (mut output, mut is_error) =
match self.tool_executor.execute(&tool_name, &input) {
Ok(output) => ConversationMessage::tool_result(
Ok(output) => (output, false),
Err(error) => (error.to_string(), true),
};
output = merge_hook_feedback(pre_hook_result.messages(), output, false);
let post_hook_result = self
.hook_runner
.run_post_tool_use(&tool_name, &input, &output, is_error);
if post_hook_result.is_denied() {
is_error = true;
}
output = merge_hook_feedback(
post_hook_result.messages(),
output,
post_hook_result.is_denied(),
);
ConversationMessage::tool_result(
tool_use_id,
tool_name,
output,
false,
),
Err(error) => ConversationMessage::tool_result(
tool_use_id,
tool_name,
error.to_string(),
true,
),
is_error,
)
}
}
PermissionOutcome::Deny { reason } => {
@@ -290,6 +335,32 @@ fn flush_text_block(text: &mut String, blocks: &mut Vec<ContentBlock>) {
}
}
fn format_hook_message(result: &HookRunResult, fallback: &str) -> String {
if result.messages().is_empty() {
fallback.to_string()
} else {
result.messages().join("\n")
}
}
fn merge_hook_feedback(messages: &[String], output: String, denied: bool) -> String {
if messages.is_empty() {
return output;
}
let mut sections = Vec::new();
if !output.trim().is_empty() {
sections.push(output);
}
let label = if denied {
"Hook feedback (denied)"
} else {
"Hook feedback"
};
sections.push(format!("{label}:\n{}", messages.join("\n")));
sections.join("\n\n")
}
type ToolHandler = Box<dyn FnMut(&str) -> Result<String, ToolError>>;
#[derive(Default)]
@@ -329,6 +400,7 @@ mod tests {
StaticToolExecutor,
};
use crate::compact::CompactionConfig;
use crate::config::{RuntimeFeatureConfig, RuntimeHookConfig};
use crate::permissions::{
PermissionMode, PermissionPolicy, PermissionPromptDecision, PermissionPrompter,
PermissionRequest,
@@ -503,6 +575,141 @@ mod tests {
));
}
#[test]
fn denies_tool_use_when_pre_tool_hook_blocks() {
struct SingleCallApiClient;
impl ApiClient for SingleCallApiClient {
fn stream(&mut self, request: ApiRequest) -> Result<Vec<AssistantEvent>, RuntimeError> {
if request
.messages
.iter()
.any(|message| message.role == MessageRole::Tool)
{
return Ok(vec![
AssistantEvent::TextDelta("blocked".to_string()),
AssistantEvent::MessageStop,
]);
}
Ok(vec![
AssistantEvent::ToolUse {
id: "tool-1".to_string(),
name: "blocked".to_string(),
input: r#"{"path":"secret.txt"}"#.to_string(),
},
AssistantEvent::MessageStop,
])
}
}
let mut runtime = ConversationRuntime::new_with_features(
Session::new(),
SingleCallApiClient,
StaticToolExecutor::new().register("blocked", |_input| {
panic!("tool should not execute when hook denies")
}),
PermissionPolicy::new(PermissionMode::DangerFullAccess),
vec!["system".to_string()],
&RuntimeFeatureConfig::default().with_hooks(RuntimeHookConfig::new(
vec![shell_snippet("printf 'blocked by hook'; exit 2")],
Vec::new(),
)),
);
let summary = runtime
.run_turn("use the tool", None)
.expect("conversation should continue after hook denial");
assert_eq!(summary.tool_results.len(), 1);
let ContentBlock::ToolResult {
is_error, output, ..
} = &summary.tool_results[0].blocks[0]
else {
panic!("expected tool result block");
};
assert!(
*is_error,
"hook denial should produce an error result: {output}"
);
assert!(
output.contains("denied tool") || output.contains("blocked by hook"),
"unexpected hook denial output: {output:?}"
);
}
#[test]
fn appends_post_tool_hook_feedback_to_tool_result() {
struct TwoCallApiClient {
calls: usize,
}
impl ApiClient for TwoCallApiClient {
fn stream(&mut self, request: ApiRequest) -> Result<Vec<AssistantEvent>, RuntimeError> {
self.calls += 1;
match self.calls {
1 => Ok(vec![
AssistantEvent::ToolUse {
id: "tool-1".to_string(),
name: "add".to_string(),
input: r#"{"lhs":2,"rhs":2}"#.to_string(),
},
AssistantEvent::MessageStop,
]),
2 => {
assert!(request
.messages
.iter()
.any(|message| message.role == MessageRole::Tool));
Ok(vec![
AssistantEvent::TextDelta("done".to_string()),
AssistantEvent::MessageStop,
])
}
_ => Err(RuntimeError::new("unexpected extra API call")),
}
}
}
let mut runtime = ConversationRuntime::new_with_features(
Session::new(),
TwoCallApiClient { calls: 0 },
StaticToolExecutor::new().register("add", |_input| Ok("4".to_string())),
PermissionPolicy::new(PermissionMode::DangerFullAccess),
vec!["system".to_string()],
&RuntimeFeatureConfig::default().with_hooks(RuntimeHookConfig::new(
vec![shell_snippet("printf 'pre hook ran'")],
vec![shell_snippet("printf 'post hook ran'")],
)),
);
let summary = runtime
.run_turn("use add", None)
.expect("tool loop succeeds");
assert_eq!(summary.tool_results.len(), 1);
let ContentBlock::ToolResult {
is_error, output, ..
} = &summary.tool_results[0].blocks[0]
else {
panic!("expected tool result block");
};
assert!(
!*is_error,
"post hook should preserve non-error result: {output:?}"
);
assert!(
output.contains('4'),
"tool output missing value: {output:?}"
);
assert!(
output.contains("pre hook ran"),
"tool output missing pre hook feedback: {output:?}"
);
assert!(
output.contains("post hook ran"),
"tool output missing post hook feedback: {output:?}"
);
}
#[test]
fn reconstructs_usage_tracker_from_restored_session() {
struct SimpleApi;
@@ -581,4 +788,14 @@ mod tests {
MessageRole::System
);
}
#[cfg(windows)]
fn shell_snippet(script: &str) -> String {
script.replace('\'', "\"")
}
#[cfg(not(windows))]
fn shell_snippet(script: &str) -> String {
script.to_string()
}
}

View File

@@ -0,0 +1,357 @@
use std::ffi::OsStr;
use std::process::Command;
use serde_json::json;
use crate::config::{RuntimeFeatureConfig, RuntimeHookConfig};
#[derive(Debug, Clone, Copy, PartialEq, Eq)]
pub enum HookEvent {
PreToolUse,
PostToolUse,
}
impl HookEvent {
fn as_str(self) -> &'static str {
match self {
Self::PreToolUse => "PreToolUse",
Self::PostToolUse => "PostToolUse",
}
}
}
#[derive(Debug, Clone, PartialEq, Eq)]
pub struct HookRunResult {
denied: bool,
messages: Vec<String>,
}
impl HookRunResult {
#[must_use]
pub fn allow(messages: Vec<String>) -> Self {
Self {
denied: false,
messages,
}
}
#[must_use]
pub fn is_denied(&self) -> bool {
self.denied
}
#[must_use]
pub fn messages(&self) -> &[String] {
&self.messages
}
}
#[derive(Debug, Clone, PartialEq, Eq, Default)]
pub struct HookRunner {
config: RuntimeHookConfig,
}
#[derive(Debug, Clone, Copy)]
struct HookCommandRequest<'a> {
event: HookEvent,
tool_name: &'a str,
tool_input: &'a str,
tool_output: Option<&'a str>,
is_error: bool,
payload: &'a str,
}
impl HookRunner {
#[must_use]
pub fn new(config: RuntimeHookConfig) -> Self {
Self { config }
}
#[must_use]
pub fn from_feature_config(feature_config: &RuntimeFeatureConfig) -> Self {
Self::new(feature_config.hooks().clone())
}
#[must_use]
pub fn run_pre_tool_use(&self, tool_name: &str, tool_input: &str) -> HookRunResult {
self.run_commands(
HookEvent::PreToolUse,
self.config.pre_tool_use(),
tool_name,
tool_input,
None,
false,
)
}
#[must_use]
pub fn run_post_tool_use(
&self,
tool_name: &str,
tool_input: &str,
tool_output: &str,
is_error: bool,
) -> HookRunResult {
self.run_commands(
HookEvent::PostToolUse,
self.config.post_tool_use(),
tool_name,
tool_input,
Some(tool_output),
is_error,
)
}
fn run_commands(
&self,
event: HookEvent,
commands: &[String],
tool_name: &str,
tool_input: &str,
tool_output: Option<&str>,
is_error: bool,
) -> HookRunResult {
if commands.is_empty() {
return HookRunResult::allow(Vec::new());
}
let payload = json!({
"hook_event_name": event.as_str(),
"tool_name": tool_name,
"tool_input": parse_tool_input(tool_input),
"tool_input_json": tool_input,
"tool_output": tool_output,
"tool_result_is_error": is_error,
})
.to_string();
let mut messages = Vec::new();
for command in commands {
match Self::run_command(
command,
HookCommandRequest {
event,
tool_name,
tool_input,
tool_output,
is_error,
payload: &payload,
},
) {
HookCommandOutcome::Allow { message } => {
if let Some(message) = message {
messages.push(message);
}
}
HookCommandOutcome::Deny { message } => {
let message = message.unwrap_or_else(|| {
format!("{} hook denied tool `{tool_name}`", event.as_str())
});
messages.push(message);
return HookRunResult {
denied: true,
messages,
};
}
HookCommandOutcome::Warn { message } => messages.push(message),
}
}
HookRunResult::allow(messages)
}
fn run_command(command: &str, request: HookCommandRequest<'_>) -> HookCommandOutcome {
let mut child = shell_command(command);
child.stdin(std::process::Stdio::piped());
child.stdout(std::process::Stdio::piped());
child.stderr(std::process::Stdio::piped());
child.env("HOOK_EVENT", request.event.as_str());
child.env("HOOK_TOOL_NAME", request.tool_name);
child.env("HOOK_TOOL_INPUT", request.tool_input);
child.env(
"HOOK_TOOL_IS_ERROR",
if request.is_error { "1" } else { "0" },
);
if let Some(tool_output) = request.tool_output {
child.env("HOOK_TOOL_OUTPUT", tool_output);
}
match child.output_with_stdin(request.payload.as_bytes()) {
Ok(output) => {
let stdout = String::from_utf8_lossy(&output.stdout).trim().to_string();
let stderr = String::from_utf8_lossy(&output.stderr).trim().to_string();
let message = (!stdout.is_empty()).then_some(stdout);
match output.status.code() {
Some(0) => HookCommandOutcome::Allow { message },
Some(2) => HookCommandOutcome::Deny { message },
Some(code) => HookCommandOutcome::Warn {
message: format_hook_warning(
command,
code,
message.as_deref(),
stderr.as_str(),
),
},
None => HookCommandOutcome::Warn {
message: format!(
"{} hook `{command}` terminated by signal while handling `{}`",
request.event.as_str(),
request.tool_name
),
},
}
}
Err(error) => HookCommandOutcome::Warn {
message: format!(
"{} hook `{command}` failed to start for `{}`: {error}",
request.event.as_str(),
request.tool_name
),
},
}
}
}
enum HookCommandOutcome {
Allow { message: Option<String> },
Deny { message: Option<String> },
Warn { message: String },
}
fn parse_tool_input(tool_input: &str) -> serde_json::Value {
serde_json::from_str(tool_input).unwrap_or_else(|_| json!({ "raw": tool_input }))
}
fn format_hook_warning(command: &str, code: i32, stdout: Option<&str>, stderr: &str) -> String {
let mut message =
format!("Hook `{command}` exited with status {code}; allowing tool execution to continue");
if let Some(stdout) = stdout.filter(|stdout| !stdout.is_empty()) {
message.push_str(": ");
message.push_str(stdout);
} else if !stderr.is_empty() {
message.push_str(": ");
message.push_str(stderr);
}
message
}
fn shell_command(command: &str) -> CommandWithStdin {
#[cfg(windows)]
let mut command_builder = {
let mut command_builder = Command::new("cmd");
command_builder.arg("/C").arg(command);
CommandWithStdin::new(command_builder)
};
#[cfg(not(windows))]
let command_builder = {
let mut command_builder = Command::new("sh");
command_builder.arg("-lc").arg(command);
CommandWithStdin::new(command_builder)
};
command_builder
}
struct CommandWithStdin {
command: Command,
}
impl CommandWithStdin {
fn new(command: Command) -> Self {
Self { command }
}
fn stdin(&mut self, cfg: std::process::Stdio) -> &mut Self {
self.command.stdin(cfg);
self
}
fn stdout(&mut self, cfg: std::process::Stdio) -> &mut Self {
self.command.stdout(cfg);
self
}
fn stderr(&mut self, cfg: std::process::Stdio) -> &mut Self {
self.command.stderr(cfg);
self
}
fn env<K, V>(&mut self, key: K, value: V) -> &mut Self
where
K: AsRef<OsStr>,
V: AsRef<OsStr>,
{
self.command.env(key, value);
self
}
fn output_with_stdin(&mut self, stdin: &[u8]) -> std::io::Result<std::process::Output> {
let mut child = self.command.spawn()?;
if let Some(mut child_stdin) = child.stdin.take() {
use std::io::Write;
child_stdin.write_all(stdin)?;
}
child.wait_with_output()
}
}
#[cfg(test)]
mod tests {
use super::{HookRunResult, HookRunner};
use crate::config::{RuntimeFeatureConfig, RuntimeHookConfig};
#[test]
fn allows_exit_code_zero_and_captures_stdout() {
let runner = HookRunner::new(RuntimeHookConfig::new(
vec![shell_snippet("printf 'pre ok'")],
Vec::new(),
));
let result = runner.run_pre_tool_use("Read", r#"{"path":"README.md"}"#);
assert_eq!(result, HookRunResult::allow(vec!["pre ok".to_string()]));
}
#[test]
fn denies_exit_code_two() {
let runner = HookRunner::new(RuntimeHookConfig::new(
vec![shell_snippet("printf 'blocked by hook'; exit 2")],
Vec::new(),
));
let result = runner.run_pre_tool_use("Bash", r#"{"command":"pwd"}"#);
assert!(result.is_denied());
assert_eq!(result.messages(), &["blocked by hook".to_string()]);
}
#[test]
fn warns_for_other_non_zero_statuses() {
let runner = HookRunner::from_feature_config(&RuntimeFeatureConfig::default().with_hooks(
RuntimeHookConfig::new(
vec![shell_snippet("printf 'warning hook'; exit 1")],
Vec::new(),
),
));
let result = runner.run_pre_tool_use("Edit", r#"{"file":"src/lib.rs"}"#);
assert!(!result.is_denied());
assert!(result
.messages()
.iter()
.any(|message| message.contains("allowing tool execution to continue")));
}
#[cfg(windows)]
fn shell_snippet(script: &str) -> String {
script.replace('\'', "\"")
}
#[cfg(not(windows))]
fn shell_snippet(script: &str) -> String {
script.to_string()
}
}

View File

@@ -4,6 +4,7 @@ mod compact;
mod config;
mod conversation;
mod file_ops;
mod hooks;
mod json;
mod mcp;
mod mcp_client;
@@ -26,8 +27,8 @@ pub use config::{
ConfigEntry, ConfigError, ConfigLoader, ConfigSource, McpClaudeAiProxyServerConfig,
McpConfigCollection, McpOAuthConfig, McpRemoteServerConfig, McpSdkServerConfig,
McpServerConfig, McpStdioServerConfig, McpTransport, McpWebSocketServerConfig, OAuthConfig,
ResolvedPermissionMode, RuntimeConfig, RuntimeFeatureConfig, ScopedMcpServerConfig,
CLAUDE_CODE_SETTINGS_SCHEMA_NAME,
ResolvedPermissionMode, RuntimeConfig, RuntimeFeatureConfig, RuntimeHookConfig,
ScopedMcpServerConfig, CLAUDE_CODE_SETTINGS_SCHEMA_NAME,
};
pub use conversation::{
ApiClient, ApiRequest, AssistantEvent, ConversationRuntime, RuntimeError, StaticToolExecutor,
@@ -38,6 +39,7 @@ pub use file_ops::{
GrepSearchInput, GrepSearchOutput, ReadFileOutput, StructuredPatchHunk, TextFilePayload,
WriteFileOutput,
};
pub use hooks::{HookEvent, HookRunResult, HookRunner};
pub use mcp::{
mcp_server_signature, mcp_tool_name, mcp_tool_prefix, normalize_name_for_mcp,
scoped_mcp_config_hash, unwrap_ccr_proxy_url,

View File

@@ -12,8 +12,9 @@ use std::process::Command;
use std::time::{SystemTime, UNIX_EPOCH};
use api::{
resolve_startup_auth_source, AnthropicClient, AuthSource, ContentBlockDelta, InputContentBlock,
InputMessage, MessageRequest, MessageResponse, OutputContentBlock,
detect_provider_kind, max_tokens_for_model, resolve_model_alias, resolve_startup_auth_source,
AnthropicClient, AuthSource, ContentBlockDelta, InputContentBlock, InputMessage,
MessageRequest, MessageResponse, OutputContentBlock, ProviderClient, ProviderKind,
StreamEvent as ApiStreamEvent, ToolChoice, ToolDefinition, ToolResultContentBlock,
};
@@ -27,7 +28,7 @@ use runtime::{
clear_oauth_credentials, generate_pkce_pair, generate_state, load_system_prompt,
parse_oauth_callback_request_target, save_oauth_credentials, ApiClient, ApiRequest,
AssistantEvent, CompactionConfig, ConfigLoader, ConfigSource, ContentBlock,
ConversationMessage, ConversationRuntime, MessageRole, OAuthAuthorizationRequest,
ConversationMessage, ConversationRuntime, MessageRole, OAuthAuthorizationRequest, OAuthConfig,
OAuthTokenExchangeRequest, PermissionMode, PermissionPolicy, ProjectContext, RuntimeError,
Session, TokenUsage, ToolError, ToolExecutor, UsageTracker,
};
@@ -35,13 +36,6 @@ use serde_json::json;
use tools::{execute_tool, mvp_tool_specs, ToolSpec};
const DEFAULT_MODEL: &str = "claude-opus-4-6";
fn max_tokens_for_model(model: &str) -> u32 {
if model.contains("opus") {
32_000
} else {
64_000
}
}
const DEFAULT_DATE: &str = "2026-03-31";
const DEFAULT_OAUTH_CALLBACK_PORT: u16 = 4545;
const VERSION: &str = env!("CARGO_PKG_VERSION");
@@ -196,6 +190,25 @@ fn parse_args(args: &[String]) -> Result<CliAction, String> {
permission_mode = PermissionMode::DangerFullAccess;
index += 1;
}
"-p" => {
// Claude Code compat: -p "prompt" = one-shot prompt
let prompt = args[index + 1..].join(" ");
if prompt.trim().is_empty() {
return Err("-p requires a prompt string".to_string());
}
return Ok(CliAction::Prompt {
prompt,
model: resolve_model_alias(&model).to_string(),
output_format,
allowed_tools: normalize_allowed_tools(&allowed_tool_values)?,
permission_mode,
});
}
"--print" => {
// Claude Code compat: --print makes output non-interactive
output_format = CliOutputFormat::Text;
index += 1;
}
"--allowedTools" | "--allowed-tools" => {
let value = args
.get(index + 1)
@@ -269,15 +282,6 @@ fn parse_args(args: &[String]) -> Result<CliAction, String> {
}
}
fn resolve_model_alias(model: &str) -> &str {
match model {
"opus" => "claude-opus-4-6",
"sonnet" => "claude-sonnet-4-6",
"haiku" => "claude-haiku-4-5-20251213",
_ => model,
}
}
fn normalize_allowed_tools(values: &[String]) -> Result<Option<AllowedToolSet>, String> {
if values.is_empty() {
return Ok(None);
@@ -428,15 +432,26 @@ fn print_bootstrap_plan() {
}
}
fn default_oauth_config() -> OAuthConfig {
OAuthConfig {
client_id: String::from("9d1c250a-e61b-44d9-88ed-5944d1962f5e"),
authorize_url: String::from("https://platform.claude.com/oauth/authorize"),
token_url: String::from("https://platform.claude.com/v1/oauth/token"),
callback_port: None,
manual_redirect_url: None,
scopes: vec![
String::from("user:profile"),
String::from("user:inference"),
String::from("user:sessions:claude_code"),
],
}
}
fn run_login() -> Result<(), Box<dyn std::error::Error>> {
let cwd = env::current_dir()?;
let config = ConfigLoader::default_for(&cwd).load()?;
let oauth = config.oauth().ok_or_else(|| {
io::Error::new(
io::ErrorKind::NotFound,
"OAuth config is missing. Add settings.oauth.clientId/authorizeUrl/tokenUrl first.",
)
})?;
let default_oauth = default_oauth_config();
let oauth = config.oauth().unwrap_or(&default_oauth);
let callback_port = oauth.callback_port.unwrap_or(DEFAULT_OAUTH_CALLBACK_PORT);
let redirect_uri = runtime::loopback_redirect_uri(callback_port);
let pkce = generate_pkce_pair()?;
@@ -950,7 +965,7 @@ struct LiveCli {
allowed_tools: Option<AllowedToolSet>,
permission_mode: PermissionMode,
system_prompt: Vec<String>,
runtime: ConversationRuntime<AnthropicRuntimeClient, CliToolExecutor>,
runtime: ConversationRuntime<ProviderRuntimeClient, CliToolExecutor>,
session: SessionHandle,
}
@@ -1873,6 +1888,15 @@ fn build_system_prompt() -> Result<Vec<String>, Box<dyn std::error::Error>> {
)?)
}
fn build_runtime_feature_config(
) -> Result<runtime::RuntimeFeatureConfig, Box<dyn std::error::Error>> {
let cwd = env::current_dir()?;
Ok(ConfigLoader::default_for(cwd)
.load()?
.feature_config()
.clone())
}
fn build_runtime(
session: Session,
model: String,
@@ -1881,14 +1905,16 @@ fn build_runtime(
emit_output: bool,
allowed_tools: Option<AllowedToolSet>,
permission_mode: PermissionMode,
) -> Result<ConversationRuntime<AnthropicRuntimeClient, CliToolExecutor>, Box<dyn std::error::Error>>
) -> Result<ConversationRuntime<ProviderRuntimeClient, CliToolExecutor>, Box<dyn std::error::Error>>
{
Ok(ConversationRuntime::new(
let feature_config = build_runtime_feature_config()?;
Ok(ConversationRuntime::new_with_features(
session,
AnthropicRuntimeClient::new(model, enable_tools, emit_output, allowed_tools.clone())?,
ProviderRuntimeClient::new(model, enable_tools, emit_output, allowed_tools.clone())?,
CliToolExecutor::new(allowed_tools, emit_output),
permission_policy(permission_mode),
system_prompt,
&feature_config,
))
}
@@ -1938,26 +1964,33 @@ impl runtime::PermissionPrompter for CliPermissionPrompter {
}
}
struct AnthropicRuntimeClient {
struct ProviderRuntimeClient {
runtime: tokio::runtime::Runtime,
client: AnthropicClient,
client: ProviderClient,
model: String,
enable_tools: bool,
emit_output: bool,
allowed_tools: Option<AllowedToolSet>,
}
impl AnthropicRuntimeClient {
impl ProviderRuntimeClient {
fn new(
model: String,
enable_tools: bool,
emit_output: bool,
allowed_tools: Option<AllowedToolSet>,
) -> Result<Self, Box<dyn std::error::Error>> {
let model = resolve_model_alias(&model).to_string();
let client = match detect_provider_kind(&model) {
ProviderKind::Anthropic => ProviderClient::from_model_with_anthropic_auth(
&model,
Some(resolve_cli_auth_source()?),
)?,
ProviderKind::Xai | ProviderKind::OpenAi => ProviderClient::from_model(&model)?,
};
Ok(Self {
runtime: tokio::runtime::Runtime::new()?,
client: AnthropicClient::from_auth(resolve_cli_auth_source()?)
.with_base_url(api::read_base_url()),
client,
model,
enable_tools,
emit_output,
@@ -1976,7 +2009,7 @@ fn resolve_cli_auth_source() -> Result<AuthSource, Box<dyn std::error::Error>> {
})?)
}
impl ApiClient for AnthropicRuntimeClient {
impl ApiClient for ProviderRuntimeClient {
#[allow(clippy::too_many_lines)]
fn stream(&mut self, request: ApiRequest) -> Result<Vec<AssistantEvent>, RuntimeError> {
let message_request = MessageRequest {
@@ -2014,7 +2047,7 @@ impl ApiClient for AnthropicRuntimeClient {
let renderer = TerminalRenderer::new();
let mut markdown_stream = MarkdownStreamState::default();
let mut events = Vec::new();
let mut pending_tool: Option<(String, String, String)> = None;
let mut pending_tools: BTreeMap<u32, (String, String, String)> = BTreeMap::new();
let mut saw_stop = false;
while let Some(event) = stream
@@ -2025,15 +2058,23 @@ impl ApiClient for AnthropicRuntimeClient {
match event {
ApiStreamEvent::MessageStart(start) => {
for block in start.message.content {
push_output_block(block, out, &mut events, &mut pending_tool, true)?;
push_output_block(
block,
0,
out,
&mut events,
&mut pending_tools,
true,
)?;
}
}
ApiStreamEvent::ContentBlockStart(start) => {
push_output_block(
start.content_block,
start.index,
out,
&mut events,
&mut pending_tool,
&mut pending_tools,
true,
)?;
}
@@ -2049,18 +2090,18 @@ impl ApiClient for AnthropicRuntimeClient {
}
}
ContentBlockDelta::InputJsonDelta { partial_json } => {
if let Some((_, _, input)) = &mut pending_tool {
if let Some((_, _, input)) = pending_tools.get_mut(&delta.index) {
input.push_str(&partial_json);
}
}
},
ApiStreamEvent::ContentBlockStop(_) => {
ApiStreamEvent::ContentBlockStop(stop) => {
if let Some(rendered) = markdown_stream.flush(&renderer) {
write!(out, "{rendered}")
.and_then(|()| out.flush())
.map_err(|error| RuntimeError::new(error.to_string()))?;
}
if let Some((id, name, input)) = pending_tool.take() {
if let Some((id, name, input)) = pending_tools.remove(&stop.index) {
// Display tool call now that input is fully accumulated
writeln!(out, "\n{}", format_tool_call_start(&name, &input))
.and_then(|()| out.flush())
@@ -2524,9 +2565,10 @@ fn truncate_for_summary(value: &str, limit: usize) -> String {
fn push_output_block(
block: OutputContentBlock,
block_index: u32,
out: &mut (impl Write + ?Sized),
events: &mut Vec<AssistantEvent>,
pending_tool: &mut Option<(String, String, String)>,
pending_tools: &mut BTreeMap<u32, (String, String, String)>,
streaming_tool_input: bool,
) -> Result<(), RuntimeError> {
match block {
@@ -2551,7 +2593,7 @@ fn push_output_block(
} else {
input.to_string()
};
*pending_tool = Some((id, name, initial_input));
pending_tools.insert(block_index, (id, name, initial_input));
}
}
Ok(())
@@ -2562,11 +2604,13 @@ fn response_to_events(
out: &mut (impl Write + ?Sized),
) -> Result<Vec<AssistantEvent>, RuntimeError> {
let mut events = Vec::new();
let mut pending_tool = None;
let mut pending_tools = BTreeMap::new();
for block in response.content {
push_output_block(block, out, &mut events, &mut pending_tool, false)?;
if let Some((id, name, input)) = pending_tool.take() {
for (index, block) in response.content.into_iter().enumerate() {
let index =
u32::try_from(index).map_err(|_| RuntimeError::new("response block index overflow"))?;
push_output_block(block, index, out, &mut events, &mut pending_tools, false)?;
if let Some((id, name, input)) = pending_tools.remove(&index) {
events.push(AssistantEvent::ToolUse { id, name, input });
}
}
@@ -2792,6 +2836,7 @@ mod tests {
use api::{MessageResponse, OutputContentBlock, Usage};
use runtime::{AssistantEvent, ContentBlock, ConversationMessage, MessageRole, PermissionMode};
use serde_json::json;
use std::collections::BTreeMap;
use std::path::PathBuf;
#[test]
@@ -2871,6 +2916,9 @@ mod tests {
assert_eq!(resolve_model_alias("opus"), "claude-opus-4-6");
assert_eq!(resolve_model_alias("sonnet"), "claude-sonnet-4-6");
assert_eq!(resolve_model_alias("haiku"), "claude-haiku-4-5-20251213");
assert_eq!(resolve_model_alias("grok"), "grok-3");
assert_eq!(resolve_model_alias("grok-mini"), "grok-3-mini");
assert_eq!(resolve_model_alias("grok-2"), "grok-2");
assert_eq!(resolve_model_alias("claude-opus"), "claude-opus");
}
@@ -3338,15 +3386,16 @@ mod tests {
fn push_output_block_renders_markdown_text() {
let mut out = Vec::new();
let mut events = Vec::new();
let mut pending_tool = None;
let mut pending_tools = BTreeMap::new();
push_output_block(
OutputContentBlock::Text {
text: "# Heading".to_string(),
},
0,
&mut out,
&mut events,
&mut pending_tool,
&mut pending_tools,
false,
)
.expect("text block should render");
@@ -3360,7 +3409,7 @@ mod tests {
fn push_output_block_skips_empty_object_prefix_for_tool_streams() {
let mut out = Vec::new();
let mut events = Vec::new();
let mut pending_tool = None;
let mut pending_tools = BTreeMap::new();
push_output_block(
OutputContentBlock::ToolUse {
@@ -3368,20 +3417,83 @@ mod tests {
name: "read_file".to_string(),
input: json!({}),
},
1,
&mut out,
&mut events,
&mut pending_tool,
&mut pending_tools,
true,
)
.expect("tool block should accumulate");
assert!(events.is_empty());
assert_eq!(
pending_tool,
pending_tools.remove(&1),
Some(("tool-1".to_string(), "read_file".to_string(), String::new(),))
);
}
#[test]
fn pending_tools_preserve_multiple_streaming_tool_calls_by_index() {
let mut out = Vec::new();
let mut events = Vec::new();
let mut pending_tools = BTreeMap::new();
push_output_block(
OutputContentBlock::ToolUse {
id: "tool-1".to_string(),
name: "read_file".to_string(),
input: json!({}),
},
1,
&mut out,
&mut events,
&mut pending_tools,
true,
)
.expect("first tool should accumulate");
push_output_block(
OutputContentBlock::ToolUse {
id: "tool-2".to_string(),
name: "grep_search".to_string(),
input: json!({}),
},
2,
&mut out,
&mut events,
&mut pending_tools,
true,
)
.expect("second tool should accumulate");
pending_tools
.get_mut(&1)
.expect("first tool pending")
.2
.push_str("{\"path\":\"src/main.rs\"}");
pending_tools
.get_mut(&2)
.expect("second tool pending")
.2
.push_str("{\"pattern\":\"TODO\"}");
assert_eq!(
pending_tools.remove(&1),
Some((
"tool-1".to_string(),
"read_file".to_string(),
"{\"path\":\"src/main.rs\"}".to_string(),
))
);
assert_eq!(
pending_tools.remove(&2),
Some((
"tool-2".to_string(),
"grep_search".to_string(),
"{\"pattern\":\"TODO\"}".to_string(),
))
);
}
#[test]
fn response_to_events_preserves_empty_object_json_input_outside_streaming() {
let mut out = Vec::new();

View File

@@ -6,10 +6,12 @@ license.workspace = true
publish.workspace = true
[dependencies]
api = { path = "../api" }
runtime = { path = "../runtime" }
reqwest = { version = "0.12", default-features = false, features = ["blocking", "rustls-tls"] }
serde = { version = "1", features = ["derive"] }
serde_json = "1"
tokio = { version = "1", features = ["rt-multi-thread"] }
[lints]
workspace = true

View File

@@ -3,10 +3,17 @@ use std::path::{Path, PathBuf};
use std::process::Command;
use std::time::{Duration, Instant};
use api::{
max_tokens_for_model, resolve_model_alias, ContentBlockDelta, InputContentBlock, InputMessage,
MessageRequest, MessageResponse, OutputContentBlock, ProviderClient,
StreamEvent as ApiStreamEvent, ToolChoice, ToolDefinition, ToolResultContentBlock,
};
use reqwest::blocking::Client;
use runtime::{
edit_file, execute_bash, glob_search, grep_search, read_file, write_file, BashCommandInput,
GrepSearchInput, PermissionMode,
edit_file, execute_bash, glob_search, grep_search, load_system_prompt, read_file, write_file,
ApiClient, ApiRequest, AssistantEvent, BashCommandInput, ContentBlock, ConversationMessage,
ConversationRuntime, GrepSearchInput, MessageRole, PermissionMode, PermissionPolicy,
RuntimeError, Session, TokenUsage, ToolError, ToolExecutor,
};
use serde::{Deserialize, Serialize};
use serde_json::{json, Value};
@@ -702,7 +709,7 @@ struct SkillOutput {
prompt: String,
}
#[derive(Debug, Serialize, Deserialize)]
#[derive(Debug, Clone, Serialize, Deserialize)]
struct AgentOutput {
#[serde(rename = "agentId")]
agent_id: String,
@@ -718,6 +725,20 @@ struct AgentOutput {
manifest_file: String,
#[serde(rename = "createdAt")]
created_at: String,
#[serde(rename = "startedAt", skip_serializing_if = "Option::is_none")]
started_at: Option<String>,
#[serde(rename = "completedAt", skip_serializing_if = "Option::is_none")]
completed_at: Option<String>,
#[serde(skip_serializing_if = "Option::is_none")]
error: Option<String>,
}
#[derive(Debug, Clone)]
struct AgentJob {
manifest: AgentOutput,
prompt: String,
system_prompt: Vec<String>,
allowed_tools: BTreeSet<String>,
}
#[derive(Debug, Serialize)]
@@ -1315,7 +1336,18 @@ fn resolve_skill_path(skill: &str) -> Result<std::path::PathBuf, String> {
Err(format!("unknown skill: {requested}"))
}
const DEFAULT_AGENT_MODEL: &str = "claude-opus-4-6";
const DEFAULT_AGENT_SYSTEM_DATE: &str = "2026-03-31";
const DEFAULT_AGENT_MAX_ITERATIONS: usize = 32;
fn execute_agent(input: AgentInput) -> Result<AgentOutput, String> {
execute_agent_with_spawn(input, spawn_agent_job)
}
fn execute_agent_with_spawn<F>(input: AgentInput, spawn_fn: F) -> Result<AgentOutput, String>
where
F: FnOnce(AgentJob) -> Result<(), String>,
{
if input.description.trim().is_empty() {
return Err(String::from("description must not be empty"));
}
@@ -1329,6 +1361,7 @@ fn execute_agent(input: AgentInput) -> Result<AgentOutput, String> {
let output_file = output_dir.join(format!("{agent_id}.md"));
let manifest_file = output_dir.join(format!("{agent_id}.json"));
let normalized_subagent_type = normalize_subagent_type(input.subagent_type.as_deref());
let model = resolve_agent_model(input.model.as_deref());
let agent_name = input
.name
.as_deref()
@@ -1336,6 +1369,8 @@ fn execute_agent(input: AgentInput) -> Result<AgentOutput, String> {
.filter(|name| !name.is_empty())
.unwrap_or_else(|| slugify_agent_name(&input.description));
let created_at = iso8601_now();
let system_prompt = build_agent_system_prompt(&normalized_subagent_type)?;
let allowed_tools = allowed_tools_for_subagent(&normalized_subagent_type);
let output_contents = format!(
"# Agent Task
@@ -1359,21 +1394,516 @@ fn execute_agent(input: AgentInput) -> Result<AgentOutput, String> {
name: agent_name,
description: input.description,
subagent_type: Some(normalized_subagent_type),
model: input.model,
status: String::from("queued"),
model: Some(model),
status: String::from("running"),
output_file: output_file.display().to_string(),
manifest_file: manifest_file.display().to_string(),
created_at,
created_at: created_at.clone(),
started_at: Some(created_at),
completed_at: None,
error: None,
};
std::fs::write(
&manifest_file,
serde_json::to_string_pretty(&manifest).map_err(|error| error.to_string())?,
)
.map_err(|error| error.to_string())?;
write_agent_manifest(&manifest)?;
let manifest_for_spawn = manifest.clone();
let job = AgentJob {
manifest: manifest_for_spawn,
prompt: input.prompt,
system_prompt,
allowed_tools,
};
if let Err(error) = spawn_fn(job) {
let error = format!("failed to spawn sub-agent: {error}");
persist_agent_terminal_state(&manifest, "failed", None, Some(error.clone()))?;
return Err(error);
}
Ok(manifest)
}
fn spawn_agent_job(job: AgentJob) -> Result<(), String> {
let thread_name = format!("clawd-agent-{}", job.manifest.agent_id);
std::thread::Builder::new()
.name(thread_name)
.spawn(move || {
let result =
std::panic::catch_unwind(std::panic::AssertUnwindSafe(|| run_agent_job(&job)));
match result {
Ok(Ok(())) => {}
Ok(Err(error)) => {
let _ =
persist_agent_terminal_state(&job.manifest, "failed", None, Some(error));
}
Err(_) => {
let _ = persist_agent_terminal_state(
&job.manifest,
"failed",
None,
Some(String::from("sub-agent thread panicked")),
);
}
}
})
.map(|_| ())
.map_err(|error| error.to_string())
}
fn run_agent_job(job: &AgentJob) -> Result<(), String> {
let mut runtime = build_agent_runtime(job)?.with_max_iterations(DEFAULT_AGENT_MAX_ITERATIONS);
let summary = runtime
.run_turn(job.prompt.clone(), None)
.map_err(|error| error.to_string())?;
let final_text = final_assistant_text(&summary);
persist_agent_terminal_state(&job.manifest, "completed", Some(final_text.as_str()), None)
}
fn build_agent_runtime(
job: &AgentJob,
) -> Result<ConversationRuntime<ProviderRuntimeClient, SubagentToolExecutor>, String> {
let model = job
.manifest
.model
.clone()
.unwrap_or_else(|| DEFAULT_AGENT_MODEL.to_string());
let allowed_tools = job.allowed_tools.clone();
let api_client = ProviderRuntimeClient::new(model, allowed_tools.clone())?;
let tool_executor = SubagentToolExecutor::new(allowed_tools);
Ok(ConversationRuntime::new(
Session::new(),
api_client,
tool_executor,
agent_permission_policy(),
job.system_prompt.clone(),
))
}
fn build_agent_system_prompt(subagent_type: &str) -> Result<Vec<String>, String> {
let cwd = std::env::current_dir().map_err(|error| error.to_string())?;
let mut prompt = load_system_prompt(
cwd,
DEFAULT_AGENT_SYSTEM_DATE.to_string(),
std::env::consts::OS,
"unknown",
)
.map_err(|error| error.to_string())?;
prompt.push(format!(
"You are a background sub-agent of type `{subagent_type}`. Work only on the delegated task, use only the tools available to you, do not ask the user questions, and finish with a concise result."
));
Ok(prompt)
}
fn resolve_agent_model(model: Option<&str>) -> String {
model
.map(str::trim)
.filter(|model| !model.is_empty())
.unwrap_or(DEFAULT_AGENT_MODEL)
.to_string()
}
fn allowed_tools_for_subagent(subagent_type: &str) -> BTreeSet<String> {
let tools = match subagent_type {
"Explore" => vec![
"read_file",
"glob_search",
"grep_search",
"WebFetch",
"WebSearch",
"ToolSearch",
"Skill",
"StructuredOutput",
],
"Plan" => vec![
"read_file",
"glob_search",
"grep_search",
"WebFetch",
"WebSearch",
"ToolSearch",
"Skill",
"TodoWrite",
"StructuredOutput",
"SendUserMessage",
],
"Verification" => vec![
"bash",
"read_file",
"glob_search",
"grep_search",
"WebFetch",
"WebSearch",
"ToolSearch",
"TodoWrite",
"StructuredOutput",
"SendUserMessage",
"PowerShell",
],
"claude-code-guide" => vec![
"read_file",
"glob_search",
"grep_search",
"WebFetch",
"WebSearch",
"ToolSearch",
"Skill",
"StructuredOutput",
"SendUserMessage",
],
"statusline-setup" => vec![
"bash",
"read_file",
"write_file",
"edit_file",
"glob_search",
"grep_search",
"ToolSearch",
],
_ => vec![
"bash",
"read_file",
"write_file",
"edit_file",
"glob_search",
"grep_search",
"WebFetch",
"WebSearch",
"TodoWrite",
"Skill",
"ToolSearch",
"NotebookEdit",
"Sleep",
"SendUserMessage",
"Config",
"StructuredOutput",
"REPL",
"PowerShell",
],
};
tools.into_iter().map(str::to_string).collect()
}
fn agent_permission_policy() -> PermissionPolicy {
mvp_tool_specs().into_iter().fold(
PermissionPolicy::new(PermissionMode::DangerFullAccess),
|policy, spec| policy.with_tool_requirement(spec.name, spec.required_permission),
)
}
fn write_agent_manifest(manifest: &AgentOutput) -> Result<(), String> {
std::fs::write(
&manifest.manifest_file,
serde_json::to_string_pretty(manifest).map_err(|error| error.to_string())?,
)
.map_err(|error| error.to_string())
}
fn persist_agent_terminal_state(
manifest: &AgentOutput,
status: &str,
result: Option<&str>,
error: Option<String>,
) -> Result<(), String> {
append_agent_output(
&manifest.output_file,
&format_agent_terminal_output(status, result, error.as_deref()),
)?;
let mut next_manifest = manifest.clone();
next_manifest.status = status.to_string();
next_manifest.completed_at = Some(iso8601_now());
next_manifest.error = error;
write_agent_manifest(&next_manifest)
}
fn append_agent_output(path: &str, suffix: &str) -> Result<(), String> {
use std::io::Write as _;
let mut file = std::fs::OpenOptions::new()
.append(true)
.open(path)
.map_err(|error| error.to_string())?;
file.write_all(suffix.as_bytes())
.map_err(|error| error.to_string())
}
fn format_agent_terminal_output(status: &str, result: Option<&str>, error: Option<&str>) -> String {
let mut sections = vec![format!("\n## Result\n\n- status: {status}\n")];
if let Some(result) = result.filter(|value| !value.trim().is_empty()) {
sections.push(format!("\n### Final response\n\n{}\n", result.trim()));
}
if let Some(error) = error.filter(|value| !value.trim().is_empty()) {
sections.push(format!("\n### Error\n\n{}\n", error.trim()));
}
sections.join("")
}
struct ProviderRuntimeClient {
runtime: tokio::runtime::Runtime,
client: ProviderClient,
model: String,
allowed_tools: BTreeSet<String>,
}
impl ProviderRuntimeClient {
fn new(model: String, allowed_tools: BTreeSet<String>) -> Result<Self, String> {
let model = resolve_model_alias(&model).to_string();
let client = ProviderClient::from_model(&model).map_err(|error| error.to_string())?;
Ok(Self {
runtime: tokio::runtime::Runtime::new().map_err(|error| error.to_string())?,
client,
model,
allowed_tools,
})
}
}
impl ApiClient for ProviderRuntimeClient {
fn stream(&mut self, request: ApiRequest) -> Result<Vec<AssistantEvent>, RuntimeError> {
let tools = tool_specs_for_allowed_tools(Some(&self.allowed_tools))
.into_iter()
.map(|spec| ToolDefinition {
name: spec.name.to_string(),
description: Some(spec.description.to_string()),
input_schema: spec.input_schema,
})
.collect::<Vec<_>>();
let message_request = MessageRequest {
model: self.model.clone(),
max_tokens: max_tokens_for_model(&self.model),
messages: convert_messages(&request.messages),
system: (!request.system_prompt.is_empty()).then(|| request.system_prompt.join("\n\n")),
tools: (!tools.is_empty()).then_some(tools),
tool_choice: (!self.allowed_tools.is_empty()).then_some(ToolChoice::Auto),
stream: true,
};
self.runtime.block_on(async {
let mut stream = self
.client
.stream_message(&message_request)
.await
.map_err(|error| RuntimeError::new(error.to_string()))?;
let mut events = Vec::new();
let mut pending_tools: BTreeMap<u32, (String, String, String)> = BTreeMap::new();
let mut saw_stop = false;
while let Some(event) = stream
.next_event()
.await
.map_err(|error| RuntimeError::new(error.to_string()))?
{
match event {
ApiStreamEvent::MessageStart(start) => {
for block in start.message.content {
push_output_block(block, 0, &mut events, &mut pending_tools, true);
}
}
ApiStreamEvent::ContentBlockStart(start) => {
push_output_block(
start.content_block,
start.index,
&mut events,
&mut pending_tools,
true,
);
}
ApiStreamEvent::ContentBlockDelta(delta) => match delta.delta {
ContentBlockDelta::TextDelta { text } => {
if !text.is_empty() {
events.push(AssistantEvent::TextDelta(text));
}
}
ContentBlockDelta::InputJsonDelta { partial_json } => {
if let Some((_, _, input)) = pending_tools.get_mut(&delta.index) {
input.push_str(&partial_json);
}
}
},
ApiStreamEvent::ContentBlockStop(stop) => {
if let Some((id, name, input)) = pending_tools.remove(&stop.index) {
events.push(AssistantEvent::ToolUse { id, name, input });
}
}
ApiStreamEvent::MessageDelta(delta) => {
events.push(AssistantEvent::Usage(TokenUsage {
input_tokens: delta.usage.input_tokens,
output_tokens: delta.usage.output_tokens,
cache_creation_input_tokens: 0,
cache_read_input_tokens: 0,
}));
}
ApiStreamEvent::MessageStop(_) => {
saw_stop = true;
events.push(AssistantEvent::MessageStop);
}
}
}
if !saw_stop
&& events.iter().any(|event| {
matches!(event, AssistantEvent::TextDelta(text) if !text.is_empty())
|| matches!(event, AssistantEvent::ToolUse { .. })
})
{
events.push(AssistantEvent::MessageStop);
}
if events
.iter()
.any(|event| matches!(event, AssistantEvent::MessageStop))
{
return Ok(events);
}
let response = self
.client
.send_message(&MessageRequest {
stream: false,
..message_request.clone()
})
.await
.map_err(|error| RuntimeError::new(error.to_string()))?;
Ok(response_to_events(response))
})
}
}
struct SubagentToolExecutor {
allowed_tools: BTreeSet<String>,
}
impl SubagentToolExecutor {
fn new(allowed_tools: BTreeSet<String>) -> Self {
Self { allowed_tools }
}
}
impl ToolExecutor for SubagentToolExecutor {
fn execute(&mut self, tool_name: &str, input: &str) -> Result<String, ToolError> {
if !self.allowed_tools.contains(tool_name) {
return Err(ToolError::new(format!(
"tool `{tool_name}` is not enabled for this sub-agent"
)));
}
let value = serde_json::from_str(input)
.map_err(|error| ToolError::new(format!("invalid tool input JSON: {error}")))?;
execute_tool(tool_name, &value).map_err(ToolError::new)
}
}
fn tool_specs_for_allowed_tools(allowed_tools: Option<&BTreeSet<String>>) -> Vec<ToolSpec> {
mvp_tool_specs()
.into_iter()
.filter(|spec| allowed_tools.is_none_or(|allowed| allowed.contains(spec.name)))
.collect()
}
fn convert_messages(messages: &[ConversationMessage]) -> Vec<InputMessage> {
messages
.iter()
.filter_map(|message| {
let role = match message.role {
MessageRole::System | MessageRole::User | MessageRole::Tool => "user",
MessageRole::Assistant => "assistant",
};
let content = message
.blocks
.iter()
.map(|block| match block {
ContentBlock::Text { text } => InputContentBlock::Text { text: text.clone() },
ContentBlock::ToolUse { id, name, input } => InputContentBlock::ToolUse {
id: id.clone(),
name: name.clone(),
input: serde_json::from_str(input)
.unwrap_or_else(|_| serde_json::json!({ "raw": input })),
},
ContentBlock::ToolResult {
tool_use_id,
output,
is_error,
..
} => InputContentBlock::ToolResult {
tool_use_id: tool_use_id.clone(),
content: vec![ToolResultContentBlock::Text {
text: output.clone(),
}],
is_error: *is_error,
},
})
.collect::<Vec<_>>();
(!content.is_empty()).then(|| InputMessage {
role: role.to_string(),
content,
})
})
.collect()
}
fn push_output_block(
block: OutputContentBlock,
block_index: u32,
events: &mut Vec<AssistantEvent>,
pending_tools: &mut BTreeMap<u32, (String, String, String)>,
streaming_tool_input: bool,
) {
match block {
OutputContentBlock::Text { text } => {
if !text.is_empty() {
events.push(AssistantEvent::TextDelta(text));
}
}
OutputContentBlock::ToolUse { id, name, input } => {
let initial_input = if streaming_tool_input
&& input.is_object()
&& input.as_object().is_some_and(serde_json::Map::is_empty)
{
String::new()
} else {
input.to_string()
};
pending_tools.insert(block_index, (id, name, initial_input));
}
}
}
fn response_to_events(response: MessageResponse) -> Vec<AssistantEvent> {
let mut events = Vec::new();
let mut pending_tools = BTreeMap::new();
for (index, block) in response.content.into_iter().enumerate() {
let index = u32::try_from(index).expect("response block index overflow");
push_output_block(block, index, &mut events, &mut pending_tools, false);
if let Some((id, name, input)) = pending_tools.remove(&index) {
events.push(AssistantEvent::ToolUse { id, name, input });
}
}
events.push(AssistantEvent::Usage(TokenUsage {
input_tokens: response.usage.input_tokens,
output_tokens: response.usage.output_tokens,
cache_creation_input_tokens: response.usage.cache_creation_input_tokens,
cache_read_input_tokens: response.usage.cache_read_input_tokens,
}));
events.push(AssistantEvent::MessageStop);
events
}
fn final_assistant_text(summary: &runtime::TurnSummary) -> String {
summary
.assistant_messages
.last()
.map(|message| {
message
.blocks
.iter()
.filter_map(|block| match block {
ContentBlock::Text { text } => Some(text.as_str()),
_ => None,
})
.collect::<Vec<_>>()
.join("")
})
.unwrap_or_default()
}
#[allow(clippy::needless_pass_by_value)]
fn execute_tool_search(input: ToolSearchInput) -> ToolSearchOutput {
let deferred = deferred_tool_specs();
@@ -2207,7 +2737,7 @@ fn execute_shell_command(
persisted_output_path: None,
persisted_output_size: None,
sandbox_status: None,
});
});
}
let mut process = std::process::Command::new(shell);
@@ -2276,7 +2806,7 @@ Command exceeded timeout of {timeout_ms} ms",
persisted_output_path: None,
persisted_output_size: None,
sandbox_status: None,
});
});
}
std::thread::sleep(Duration::from_millis(10));
}
@@ -2365,6 +2895,8 @@ fn parse_skill_description(contents: &str) -> Option<String> {
#[cfg(test)]
mod tests {
use std::collections::BTreeMap;
use std::collections::BTreeSet;
use std::fs;
use std::io::{Read, Write};
use std::net::{SocketAddr, TcpListener};
@@ -2373,7 +2905,13 @@ mod tests {
use std::thread;
use std::time::Duration;
use super::{execute_tool, mvp_tool_specs};
use super::{
agent_permission_policy, allowed_tools_for_subagent, execute_agent_with_spawn,
execute_tool, final_assistant_text, mvp_tool_specs, persist_agent_terminal_state,
push_output_block, AgentInput, AgentJob, SubagentToolExecutor,
};
use api::OutputContentBlock;
use runtime::{ApiRequest, AssistantEvent, ConversationRuntime, RuntimeError, Session};
use serde_json::json;
fn env_lock() -> &'static Mutex<()> {
@@ -2587,6 +3125,63 @@ mod tests {
assert!(error.contains("relative URL without a base") || error.contains("empty host"));
}
#[test]
fn pending_tools_preserve_multiple_streaming_tool_calls_by_index() {
let mut events = Vec::new();
let mut pending_tools = BTreeMap::new();
push_output_block(
OutputContentBlock::ToolUse {
id: "tool-1".to_string(),
name: "read_file".to_string(),
input: json!({}),
},
1,
&mut events,
&mut pending_tools,
true,
);
push_output_block(
OutputContentBlock::ToolUse {
id: "tool-2".to_string(),
name: "grep_search".to_string(),
input: json!({}),
},
2,
&mut events,
&mut pending_tools,
true,
);
pending_tools
.get_mut(&1)
.expect("first tool pending")
.2
.push_str("{\"path\":\"src/main.rs\"}");
pending_tools
.get_mut(&2)
.expect("second tool pending")
.2
.push_str("{\"pattern\":\"TODO\"}");
assert_eq!(
pending_tools.remove(&1),
Some((
"tool-1".to_string(),
"read_file".to_string(),
"{\"path\":\"src/main.rs\"}".to_string(),
))
);
assert_eq!(
pending_tools.remove(&2),
Some((
"tool-2".to_string(),
"grep_search".to_string(),
"{\"pattern\":\"TODO\"}".to_string(),
))
);
}
#[test]
fn todo_write_persists_and_returns_previous_state() {
let _guard = env_lock()
@@ -2765,32 +3360,48 @@ mod tests {
.unwrap_or_else(std::sync::PoisonError::into_inner);
let dir = temp_path("agent-store");
std::env::set_var("CLAWD_AGENT_STORE", &dir);
let captured = Arc::new(Mutex::new(None::<AgentJob>));
let captured_for_spawn = Arc::clone(&captured);
let result = execute_tool(
"Agent",
&json!({
"description": "Audit the branch",
"prompt": "Check tests and outstanding work.",
"subagent_type": "Explore",
"name": "ship-audit"
}),
let manifest = execute_agent_with_spawn(
AgentInput {
description: "Audit the branch".to_string(),
prompt: "Check tests and outstanding work.".to_string(),
subagent_type: Some("Explore".to_string()),
name: Some("ship-audit".to_string()),
model: None,
},
move |job| {
*captured_for_spawn
.lock()
.unwrap_or_else(std::sync::PoisonError::into_inner) = Some(job);
Ok(())
},
)
.expect("Agent should succeed");
std::env::remove_var("CLAWD_AGENT_STORE");
let output: serde_json::Value = serde_json::from_str(&result).expect("valid json");
assert_eq!(output["name"], "ship-audit");
assert_eq!(output["subagentType"], "Explore");
assert_eq!(output["status"], "queued");
assert!(output["createdAt"].as_str().is_some());
let manifest_file = output["manifestFile"].as_str().expect("manifest file");
let output_file = output["outputFile"].as_str().expect("output file");
let contents = std::fs::read_to_string(output_file).expect("agent file exists");
assert_eq!(manifest.name, "ship-audit");
assert_eq!(manifest.subagent_type.as_deref(), Some("Explore"));
assert_eq!(manifest.status, "running");
assert!(!manifest.created_at.is_empty());
assert!(manifest.started_at.is_some());
assert!(manifest.completed_at.is_none());
let contents = std::fs::read_to_string(&manifest.output_file).expect("agent file exists");
let manifest_contents =
std::fs::read_to_string(manifest_file).expect("manifest file exists");
std::fs::read_to_string(&manifest.manifest_file).expect("manifest file exists");
assert!(contents.contains("Audit the branch"));
assert!(contents.contains("Check tests and outstanding work."));
assert!(manifest_contents.contains("\"subagentType\": \"Explore\""));
assert!(manifest_contents.contains("\"status\": \"running\""));
let captured_job = captured
.lock()
.unwrap_or_else(std::sync::PoisonError::into_inner)
.clone()
.expect("spawn job should be captured");
assert_eq!(captured_job.prompt, "Check tests and outstanding work.");
assert!(captured_job.allowed_tools.contains("read_file"));
assert!(!captured_job.allowed_tools.contains("Agent"));
let normalized = execute_tool(
"Agent",
@@ -2819,6 +3430,195 @@ mod tests {
let _ = std::fs::remove_dir_all(dir);
}
#[test]
fn agent_fake_runner_can_persist_completion_and_failure() {
let _guard = env_lock()
.lock()
.unwrap_or_else(std::sync::PoisonError::into_inner);
let dir = temp_path("agent-runner");
std::env::set_var("CLAWD_AGENT_STORE", &dir);
let completed = execute_agent_with_spawn(
AgentInput {
description: "Complete the task".to_string(),
prompt: "Do the work".to_string(),
subagent_type: Some("Explore".to_string()),
name: Some("complete-task".to_string()),
model: Some("claude-sonnet-4-6".to_string()),
},
|job| {
persist_agent_terminal_state(
&job.manifest,
"completed",
Some("Finished successfully"),
None,
)
},
)
.expect("completed agent should succeed");
let completed_manifest = std::fs::read_to_string(&completed.manifest_file)
.expect("completed manifest should exist");
let completed_output =
std::fs::read_to_string(&completed.output_file).expect("completed output should exist");
assert!(completed_manifest.contains("\"status\": \"completed\""));
assert!(completed_output.contains("Finished successfully"));
let failed = execute_agent_with_spawn(
AgentInput {
description: "Fail the task".to_string(),
prompt: "Do the failing work".to_string(),
subagent_type: Some("Verification".to_string()),
name: Some("fail-task".to_string()),
model: None,
},
|job| {
persist_agent_terminal_state(
&job.manifest,
"failed",
None,
Some(String::from("simulated failure")),
)
},
)
.expect("failed agent should still spawn");
let failed_manifest =
std::fs::read_to_string(&failed.manifest_file).expect("failed manifest should exist");
let failed_output =
std::fs::read_to_string(&failed.output_file).expect("failed output should exist");
assert!(failed_manifest.contains("\"status\": \"failed\""));
assert!(failed_manifest.contains("simulated failure"));
assert!(failed_output.contains("simulated failure"));
let spawn_error = execute_agent_with_spawn(
AgentInput {
description: "Spawn error task".to_string(),
prompt: "Never starts".to_string(),
subagent_type: None,
name: Some("spawn-error".to_string()),
model: None,
},
|_| Err(String::from("thread creation failed")),
)
.expect_err("spawn errors should surface");
assert!(spawn_error.contains("failed to spawn sub-agent"));
let spawn_error_manifest = std::fs::read_dir(&dir)
.expect("agent dir should exist")
.filter_map(Result::ok)
.map(|entry| entry.path())
.filter(|path| path.extension().and_then(|ext| ext.to_str()) == Some("json"))
.find_map(|path| {
let contents = std::fs::read_to_string(&path).ok()?;
contents
.contains("\"name\": \"spawn-error\"")
.then_some(contents)
})
.expect("failed manifest should still be written");
assert!(spawn_error_manifest.contains("\"status\": \"failed\""));
assert!(spawn_error_manifest.contains("thread creation failed"));
std::env::remove_var("CLAWD_AGENT_STORE");
let _ = std::fs::remove_dir_all(dir);
}
#[test]
fn agent_tool_subset_mapping_is_expected() {
let general = allowed_tools_for_subagent("general-purpose");
assert!(general.contains("bash"));
assert!(general.contains("write_file"));
assert!(!general.contains("Agent"));
let explore = allowed_tools_for_subagent("Explore");
assert!(explore.contains("read_file"));
assert!(explore.contains("grep_search"));
assert!(!explore.contains("bash"));
let plan = allowed_tools_for_subagent("Plan");
assert!(plan.contains("TodoWrite"));
assert!(plan.contains("StructuredOutput"));
assert!(!plan.contains("Agent"));
let verification = allowed_tools_for_subagent("Verification");
assert!(verification.contains("bash"));
assert!(verification.contains("PowerShell"));
assert!(!verification.contains("write_file"));
}
#[derive(Debug)]
struct MockSubagentApiClient {
calls: usize,
input_path: String,
}
impl runtime::ApiClient for MockSubagentApiClient {
fn stream(&mut self, request: ApiRequest) -> Result<Vec<AssistantEvent>, RuntimeError> {
self.calls += 1;
match self.calls {
1 => {
assert_eq!(request.messages.len(), 1);
Ok(vec![
AssistantEvent::ToolUse {
id: "tool-1".to_string(),
name: "read_file".to_string(),
input: json!({ "path": self.input_path }).to_string(),
},
AssistantEvent::MessageStop,
])
}
2 => {
assert!(request.messages.len() >= 3);
Ok(vec![
AssistantEvent::TextDelta("Scope: completed mock review".to_string()),
AssistantEvent::MessageStop,
])
}
_ => panic!("unexpected mock stream call"),
}
}
}
#[test]
fn subagent_runtime_executes_tool_loop_with_isolated_session() {
let _guard = env_lock()
.lock()
.unwrap_or_else(std::sync::PoisonError::into_inner);
let path = temp_path("subagent-input.txt");
std::fs::write(&path, "hello from child").expect("write input file");
let mut runtime = ConversationRuntime::new(
Session::new(),
MockSubagentApiClient {
calls: 0,
input_path: path.display().to_string(),
},
SubagentToolExecutor::new(BTreeSet::from([String::from("read_file")])),
agent_permission_policy(),
vec![String::from("system prompt")],
);
let summary = runtime
.run_turn("Inspect the delegated file", None)
.expect("subagent loop should succeed");
assert_eq!(
final_assistant_text(&summary),
"Scope: completed mock review"
);
assert!(runtime
.session()
.messages
.iter()
.flat_map(|message| message.blocks.iter())
.any(|block| matches!(
block,
runtime::ContentBlock::ToolResult { output, .. }
if output.contains("hello from child")
)));
let _ = std::fs::remove_file(path);
}
#[test]
fn agent_rejects_blank_required_fields() {
let missing_description = execute_tool(