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.
Instalação
Instalar com Codex ou Claude Copie este prompt, cole no Codex, Claude ou outro assistente e deixe que ele revise a página da skill e instale para você.
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
