verify
Run this repository's gates on the work about to be committed — the backtest suite, and the quality score for every staged `.qmd`, `.tex` or `.R` file — and stop the commit if one fails. Claude Code tells Claude to run a project skill named `verify` right before committing (not f
Install
npx skills add https://github.com/pedrohcgs/claude-code-my-workflow/tree/main/.claude/skills/verify
claude plugin marketplace add https://llmmart.ai/marketplace.json && claude plugin install pedrohcgs-claude-code-my-workflow@llmmart
git clone https://github.com/pedrohcgs/claude-code-my-workflow.git
The skills CLI installs just this skill, for any of its supported agents. Claude Code installs the whole pedrohcgs/claude-code-my-workflow collection as a plugin from our marketplace. Git is the plain clone.
Skill manifest
/verify — the gates, before a commit
Since Claude Code 2.1.286, Claude is told to run a project skill named verify right before it commits, except for docs-only and tests-only commits. This one runs the same gates /commit runs. It is a reminder, not enforcement: Claude can skip it, and an exempt commit never triggers it, even one that changes a .qmd or .tex file. The enforcing gates are the git pre-commit hook (./scripts/install-hooks.sh) and /commit. It does not commit, push, stage, or edit anything.
Steps
Is the git hook already doing this? Only when all three checks from
/commitStep 0b hold:git config --get core.hooksPathprints.githooks;test -x "$(git rev-parse --show-toplevel)/.githooks/pre-commit" && echo executableprintsexecutable(git skips a hook file that is missing or not executable, and still commits); andgit status --porcelain -- ':/.githooks'prints nothing (an edited hook is itself what needs testing, and it could exit 0 without running anything). Then the pre-commit hook will run the backtest and the quality gate on this commit anyway. Say so in one line and stop, because running everything twice wastes a minute and adds nothing. Otherwise go on to step 2.Run the backtest:
./scripts/backtest.sh. Capture the exit code and the last lines. A non-zero exit fails/verify; report the failing gate's own message, not a summary of it.Score staged prose and code: for each staged file ending in
.qmd,.texor.R(git diff --cached --name-only --diff-filter=ACMR), runpython3 scripts/quality_score.py <file>. A score below 80 fails/verify.Report. One line per check (
PASS/FAIL, with the score or the gate's message), then the verdict.- FAIL: do not commit. Say what failed and what fixes it.
- PASS: say the commit can go ahead.
If the user has explicitly said "commit anyway" or "skip quality gate", pass that on:
/commitrecords such an override asSKIP_QUALITY_GATE=1, which skips only the quality score. Never suggestgit commit --no-verify.
Not this skill's job
Fixing what failed (do that, then run /verify again), writing the commit message (/commit), and deciding whether a change is wanted (the user). A gate that cannot run, such as a missing interpreter or a script error, is a FAIL, not a pass.
Files (claude-code-my-workflow)
-
SKILL.md 2.9 KB
--- name: verify description: Run this repository's gates on the work about to be committed — the backtest suite, and the quality score for every staged `.qmd`, `.tex` or `.R` file — and stop the commit if one fails. Claude Code tells Claude to run a project skill named `verify` right before committing (not for docs-only or tests-only commits); also use it when the user says "verify", "run the gates", "check before I commit", or "is this ready to commit?". It checks and reports; it never commits, pushes or fixes anything itself. allowed-tools: ["Bash", "Read"] effort: low --- # /verify — the gates, before a commit Since Claude Code 2.1.286, Claude is *told* to run a project skill named `verify` right before it commits, except for docs-only and tests-only commits. This one runs the same gates `/commit` runs. It is a reminder, not enforcement: Claude can skip it, and an exempt commit never triggers it, even one that changes a `.qmd` or `.tex` file. The enforcing gates are the git pre-commit hook (`./scripts/install-hooks.sh`) and `/commit`. It **does not** commit, push, stage, or edit anything. ## Steps 1. **Is the git hook already doing this?** Only when all three checks from `/commit` Step 0b hold: `git config --get core.hooksPath` prints `.githooks`; `test -x "$(git rev-parse --show-toplevel)/.githooks/pre-commit" && echo executable` prints `executable` (git skips a hook file that is missing or not executable, and still commits); and `git status --porcelain -- ':/.githooks'` prints nothing (an edited hook is itself what needs testing, and it could exit 0 without running anything). Then the pre-commit hook will run the backtest and the quality gate on this commit anyway. Say so in one line and stop, because running everything twice wastes a minute and adds nothing. Otherwise go on to step 2. 2. **Run the backtest:** `./scripts/backtest.sh`. Capture the exit code and the last lines. A non-zero exit fails `/verify`; report the failing gate's own message, not a summary of it. 3. **Score staged prose and code:** for each staged file ending in `.qmd`, `.tex` or `.R` (`git diff --cached --name-only --diff-filter=ACMR`), run `python3 scripts/quality_score.py <file>`. A score below 80 fails `/verify`. 4. **Report.** One line per check (`PASS` / `FAIL`, with the score or the gate's message), then the verdict. - **FAIL:** do not commit. Say what failed and what fixes it. - **PASS:** say the commit can go ahead. If the user has explicitly said "commit anyway" or "skip quality gate", pass that on: `/commit` records such an override as `SKIP_QUALITY_GATE=1`, which skips only the quality score. Never suggest `git commit --no-verify`. ## Not this skill's job Fixing what failed (do that, then run `/verify` again), writing the commit message (`/commit`), and deciding whether a change is wanted (the user). A gate that cannot run, such as a missing interpreter or a script error, is a FAIL, not a pass.
Comments (0)
Sign in to join the conversation.
Reviews (0)
No reviews yet.
No comments yet.