work_skill
- Shell
212
GitHub Stars
1
Bundled Files
2 months ago
Catalog Refreshed
4 months ago
First Indexed
Readme & install
Copy the install command, review bundled files from the catalogue, and read any extended description pulled from the listing source.
Installation
Preview and clipboard use veilstrat where the catalogue uses aiagentskills.
npx veilstrat add skill chachamaru127/claude-code-harness --skill work- SKILL.md9.2 KB
Overview
This skill automates implementing tasks listed in Plans.md based on a user-specified scope. It asks for scope when invoked without arguments, then selects an execution strategy automatically and runs implementations, reviews, and optional commits. It supports single-task direct coding, parallel workers for multiple tasks, and iterative loops for large batches.
How this skill works
When triggered it inspects Plans.md to identify target tasks and chooses a strategy by task count and independence. For one task it implements directly; for a few tasks it spawns parallel task-workers; for many or "all" it runs parallel workers plus auto-iteration until tasks pass build/test/review or reach iteration limits. After implementation it runs review cycles and can auto-commit unless suppressed.
When to use it
- User types /work or variants (e.g., /work 3, /work all, /work 3-6).
- You want code implemented from Plans.md without manual orchestration.
- You need single-task quick fixes, multi-task parallel work, or full-batch iterative completion.
- When you want integrated review loops and optional auto-commit.
- Invoke with --codex to delegate implementation to Codex MCP.
Best practices
- Specify a clear scope when possible (task number, range, or all) to avoid interactive prompts.
- Use --no-commit in experimental sessions to prevent automatic commits.
- For independent changes, allow parallel workers to speed completion; limit parallelism for edits to the same file.
- Set --max-iterations for very large batches to bound automatic retries.
- Use --codex only when you want external engine execution and stricter guard behavior.
Example use cases
- /work to implement the next pending task and commit the change.
- /work 3-5 to implement a small group of related tasks in parallel and run a review loop.
- /work all to iterate until all Plans.md tasks are cc:done, tests pass, and reviews approve.
- /work --codex all to delegate implementation to Codex MCP while the harness manages review and quality gates.
- /work --no-commit during a dry run or exploratory coding session.
FAQ
The skill retries failed tasks according to the strategy: single tasks will be fixed and re-reviewed; large batches use auto-iteration to learn from failures and reattempt until success or iteration limit.
How do I stop automatic commits?
Include --no-commit to suppress auto-commit. You can also configure defaults in the project harness config to disable auto-commit.
When should I use --codex?
Use --codex to offload implementation to Codex MCP. The harness still handles orchestration, review, and quality gates, but edit permissions and guard behavior differ under Codex.