Pre-commit and pre-push validation for fabio contributions. Run this skill before committing or creating a PR to ensure code quality, formatting, test coverage, and documentation are all correct. Invoke when: ready to commit, preparing a PR, reviewing changes before push.
Installation
Install with Codex or Claude Copy this prompt, paste it into Codex, Claude, or another assistant, and let it review the skill page and install it for you.
Pre-commit and pre-push validation for fabio contributions. Run this skill before committing or creating a PR to ensure code quality, formatting, test coverage, and documentation are all correct. Invoke when: ready to commit, preparing a PR, reviewing changes before push.
PR Checklist for Fabio
Run this checklist before every commit. Each step must pass before proceeding to the next.
Note: The project uses prek pre-commit hooks (prek.toml). When installed (cargo install prek && prek install), Steps 1-2 run automatically on git commit. Tests (Step 3) are NOT in hooks — always run them manually. Do NOT bypass hooks with --no-verify.
Step 1: Format
cargo fmt -- --check
If it fails, fix with cargo fmt and re-check.
Step 2: Lint
cargo clippy --tests -- -D warnings
Fix ALL warnings. Common issues:
Unused imports — remove them, don't leave for later
case_sensitive_file_extension_comparisons — use Path::extension() instead of ends_with(".json")
too_many_lines — split the function or add #[allow(clippy::too_many_lines)]
doc_markdown — wrap identifiers in backticks in doc comments
Step 3: Test
cargo test
All tests must pass. If you added new code, verify it has tests.
Only needed if you added, modified, or removed commands/flags:
cargo test generate_agent_schema -- --ignored
cargo test agent_schema_covers
Skipping this leaves commands.json stale: the new subcommand won't be exposed as an MCP tool and can't be referenced by --enable-commands / --disable-commands policy patterns. The agent_schema_covers_* drift tests fail CI if you forget.
Step 5: Self-review
Run git diff --staged (or git diff if not yet staged) and review every hunk:
Best-practice or workflow added if applicable (just drop a .json file in src/commands/context/data/best_practices/ or workflows/)
Output examples added for non-obvious response shapes (src/commands/context/data/examples/)
README.md updated if user-facing behavior changed
Step 7: Check irreversible operation safety
If your change adds or modifies a destructive operation (deletes data, overwrites
without backup, replaces a definition, kills a session/job, or is otherwise
irreversible), verify the FULL guardrail stack (see AGENTS.md → "Standard
guardrail stack for a NEW destructive command"):
--dry-run guard via output::dry_run_guard(cli, "<group> <cmd>", &preview), returning early before any mutating call
--readonly enforced (mutation routes through a client post/put/patch/delete helper that calls guard_readonly)
"destructive": true (and "mutates": true) confirmed in commands.json after generate_agent_schema — set manually for non-delete*-named ops (reset, kill, prune, update-definition, --hard-delete, --force*)
Blast-radius input guard for catastrophic inputs (empty/root path, match-all glob, missing filter) — a pure validate_* fn that fails before any network call, with a unit test
FabioError::with_hint() used when suggesting safety-bypass flags
New safety-bypass flags added to DANGEROUS_FLAGS in src/agent.rs
"destructive": true/false included in batch/plan output if applicable
Protected types added to PROTECTED_DELETE_TYPES if new data-bearing item type
e2e test for the --dry-run output AND for the blast-radius guard error
Removal verb is delete (never remove)
Step 8: Commit
git add <files>
git status # verify only intended files staged
git commit -m "<type>: <description>"
Commit message format: imperative mood, concise subject (50 chars), body if needed.
Types: feat, fix, docs, refactor, test, chore.
Include Assisted-by: trailer for AI attribution.
Step 9: Pre-push validation
Before pushing, run the cross-compilation check:
./scripts/cross-check.sh
This catches Windows/macOS/ARM64 compilation issues that local tests miss.
Iterate faster with: ./scripts/cross-check.sh --target windows-x64
