Prompt file imported from zahidprinters/ai-lan-agent (
.github/prompts/new-tool.prompt.md). Copyright stays with the author.
Create or update a repository-native tool or action workflow for this workspace.
Requirements:
- Follow the repo's layered boundaries from docs/PROJECT_STRUCTURE.md and docs/ARCHITECTURE.md.
- Use the flow from docs/AI_GUIDELINES.md: schema first, then policy/confirmation behavior, then a safe stub or implementation, then registration, then tests.
- Preserve Windows-first, CPU-friendly behavior and keep temporary artifacts in
temp/. - Prefer minimal changes and preserve compatibility with both the newer layered scaffold and any active legacy path.
- Keep policy behavior explicit: state whether the new action is allowed, denied, or confirmation-required.
Expected workflow:
- Inspect the existing router, schema, policy, and tool registration surfaces relevant to the requested capability.
- Identify the smallest safe design that satisfies the request.
- Implement the change end-to-end where feasible, including tests and any necessary docs updates.
- Run the smallest relevant validation commands, starting with unit tests.
- Summarize the user-facing behavior, safety posture, and files changed.
Implementation checklist:
- Schema:
- Action name and argument shape are explicit and validated.
- Policy:
- Confirmation or denial behavior is intentional and testable.
- Dispatch:
- Registration path is clear and compatibility with legacy path is preserved.
- Tool path:
- Initial implementation is safe-by-default with deterministic failure output.
- Validation:
- Added or updated tests cover successful and blocked/denied cases where relevant.
- Docs:
- Config, workflow, or API docs updated when behavior changed.
Output expectations:
- State where the capability was wired.
- State whether it is read-only, confirmation-required, or blocked by policy.
- Mention which validation commands were run and anything not verified.
- Include a short risk note for any unverified integration path.
