mirror of
https://github.com/alirezarezvani/claude-skills.git
synced 2026-10-09 03:17:54 +00:00
abd9c9d8de
generate-docs.py learns the agent-launcher domain (5 hardcoded maps extended); regenerated docs tree: 343 skill pages / 96 agent pages / 122 command pages (561 total). mkdocs.yml nav gains the Agent Launcher skill section (7 pages), 4 cs-agent-* agent entries, and 8 /cs:* command entries; all nav targets verified to exist. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_012FwXG6TqCXKZQvF4iD69cv
40 lines
2.4 KiB
Markdown
40 lines
2.4 KiB
Markdown
---
|
|
title: "/cs-harness — Slash Command for AI Coding Agents"
|
|
description: "Compile a goal into a verified agent-harness loop for a domain and drive it to close — /cs:harness <domain> <goal>. Slash command for Claude Code, Codex CLI, Gemini CLI."
|
|
---
|
|
|
|
# /cs-harness
|
|
|
|
<div class="page-meta" markdown>
|
|
<span class="meta-badge">:material-console: Slash Command</span>
|
|
<span class="meta-badge">:material-github: <a href="https://github.com/alirezarezvani/2-claude-skills/tree/main/engineering/agent-harness/commands/cs-harness.md">Source</a></span>
|
|
</div>
|
|
|
|
|
|
Parse `$ARGUMENTS`: the first token is the domain (one of the 18 manifest names under
|
|
`engineering/agent-harness/skills/agent-harness/assets/harnesses/`); the rest is the goal.
|
|
If the domain token doesn't match a manifest file, list the available manifests and ask.
|
|
|
|
## Sequence (gates are blocking — never skip forward)
|
|
|
|
1. **Compile** —
|
|
`python3 engineering/agent-harness/skills/agent-harness/scripts/goal_compiler.py --goal "<goal>" --manifest engineering/agent-harness/skills/agent-harness/assets/harnesses/<domain>.json --out .agent-harness/plan.json`
|
|
- Exit 3: relay the forcing questions to the user one at a time (recommended answer
|
|
first), then recompile with the enriched goal. Do not proceed on a vague goal.
|
|
- Exit 4: show `nearest_candidates`, ask whether to switch domain or refine the goal.
|
|
2. **Review the plan with the user** — show tasks, verifications, and caps. Confirm before
|
|
initializing: this is the only approval gate in the loop.
|
|
3. **Init** — `python3 .../scripts/loop_controller.py init --plan .agent-harness/plan.json --state .agent-harness/state.json`
|
|
4. **Drive** — repeat: `next` → execute the task per its skill's SKILL.md → `record` →
|
|
`verify`. For long goals, spawn the `harness-runner` agent per task instead of executing
|
|
inline, one at a time (writes stay serialized).
|
|
5. **On exit 2 or 5** — stop, show `status` and the failing evidence; the user decides:
|
|
fix and continue, waive with a reason, or abandon.
|
|
6. **Close** — `close --state .agent-harness/state.json`; paste the handoff block
|
|
(tasks, statuses, evidence, waivers) as the deliverable summary.
|
|
|
|
## Rules
|
|
|
|
- Never edit checks, manifests, or the plan mid-loop to make verification pass.
|
|
- Never report an exhausted budget as success.
|
|
- `.agent-harness/` is git-ignorable working state; the handoff block is the record.
|