souschef
Server Configuration
Describes the environment variables required to run the server.
| Name | Required | Description | Default |
|---|---|---|---|
| SOUSCHEF_AGY | No | Path to the Antigravity CLI (`agy`) binary. If not set, souschef finds it on PATH or in the installer's default location. |
Instructions
Guidance the server publishes about itself, which clients place ahead of the tool catalog so the model reads it before choosing anything.
This server publishes no instructions, or was last inspected before Glama recorded them.
Capabilities
Features and capabilities supported by this server
Protocol revision2025-11-25
| Capability | Details |
|---|---|
| tools | {
"listChanged": true
} |
Tools
Functions exposed to the LLM to take actions
| Name | Description |
|---|---|
| delegate_taskA | Starts a task (currently taskType "tests") in an isolated workspace and returns a taskId immediately. SousChef detects the test command, lets the agent write tests, runs build/tests, checks that only test files changed, and applies verified test files to the checkout. Follow with wait_for_task, then get_task_result. The answer carries |
| get_task_statusB | Current status, phase and one-line progress of a task, plus its latest steps in |
| wait_for_taskA | Blocks until the task finishes or timeoutSeconds passes, then returns its status and |
| get_task_resultC | Compact JSON report (status, build, test counts, security check, applied files, summary). Logs are included only when asked for via include. |
| get_task_diffA | Unified diff of the files the task changed (including out-of-scope ones). Large diffs return a file list first; then request one path at a time and page with offset. |
| cancel_taskA | Stops a queued or running task and the processes SousChef started for it. Nothing is applied. |
| doctorB | Checks the agent binary, the project config and the detected test command. ping=true also sends the agent a one-word prompt. |
Prompts
Interactive templates invoked by user choice
| Name | Description |
|---|---|
No prompts | |
Resources
Contextual data attached and managed by the client
| Name | Description |
|---|---|
No resources | |
TDQS
Scored across 7 tools
The tools have distinct primary roles: delegate_task starts a task, wait_for_task blocks, get_task_status gives a snapshot, get_task_result returns a final report, get_task_diff shows changes, cancel_task stops, and doctor checks setup. However, wait_for_task, get_task_status, and get_task_result all surface status or progress details, so an agent could occasionally choose the wrong one when checking on a task.
Most names follow a clear verb_noun snake_case pattern (delegate_task, get_task_result, etc.), but 'doctor' breaks the pattern as a bare noun with no verb. Still readable and mostly consistent.
Seven tools is well within the sweet spot for a task orchestrator; each covers a distinct lifecycle operation (start, monitor, result, diff, cancel, environment check) without bloat.
The surface covers the core lifecycle: delegation, blocking wait, status polling, result retrieval, diff inspection, cancellation, and environment diagnosis. Minor gaps exist—e.g., no tool to list tasks or retrieve a task by anything other than ID—but agents can work around these by tracking task IDs.