Pravidhi Commit Protocol Skill

​> Developed and maintained by Pravidhi

Overview

This skill drives a complete, repeatable, and quality-enforced coding workflow inside any Git repository. Every major stage can pause and ask the user for input. The goal is clean atomic commits, zero leaked secrets, minimal diff noise, and a passing CI pipeline before a single line is pushed.

Formatting threshold (hardcoded default): If more than 10 files would be reformatted, stop and ask the user before applying formatting. Adjust per project if the user instructs you to.

Quick Reference

Stage 1 — Repository Hygiene

1.1 Branch Safety Check

git branch --show-current

Rules:

• If branch is main, master, dev, or production: STOP. Ask user: “You are on a protected branch. Should I create a new feature branch named feature/<​task-slug​>?”
• If branch is a feature branch: Proceed.

1.2 Sync with Remote

git fetch --all --prune
git pull --ff-only

If --ff-only fails, stop and ask the user whether to rebase or merge.

1.3 Check for Uncommitted Changes

git status --porcelain

• Empty output → repository is clean, continue.
• Non-empty output → inspect each entry:

Present a clean summary to the user and ask how to proceed before continuing.

1.4 .gitignore Triage

Scan untracked paths (?? entries) and immediately add known suspects to .gitignore without asking:

# Build artifacts
dist/         build/        out/           target/       .next/
__pycache__/ *.pyc         *.pyo          *.class       *.o

# Logs & temp
*.log         *.tmp         .DS_Store      Thumbs.db

# Generated files
coverage/     .nyc_output/ *.lcov        htmlcov/

# Secrets & environment (NEVER commit these)
.env          .env.* *.pem         *.key         *.p12
secrets/      credentials/ .aws/         .gcloud/

# IDE / editor
.idea/        .vscode/      *.swp         *.swo

For ambiguous untracked paths (not obviously an artifact), ask the user before adding.

Stage and commit .gitignore changes as a standalone commit:

chore: update .gitignore — exclude build artifacts and env files

Stage 2 — Dependency & Security Audit

Run the appropriate audit for the detected ecosystem(s). Multiple ecosystems may be present in a monorepo.

Detection

# Detect ecosystems
ls package.json package-lock.json yarn.lock pnpm-lock.yaml   # Node
ls requirements.txt Pipfile pyproject.toml setup.py            # Python
ls Cargo.toml                                                  # Rust
ls go.sum                                                      # Go
ls Gemfile.lock                                                # Ruby
ls composer.lock                                               # PHP

Audit Commands by Ecosystem

# Node
npm audit --audit-level=moderate          # or: yarn audit / pnpm audit
npx audit-ci --moderate                   # CI-friendly exit code

# Python
pip-audit                                  # pip install pip-audit
safety check --full-report                 # pip install safety

# Rust
cargo audit                                # cargo install cargo-audit

# Go
govulncheck ./...                          # go install golang.org/x/vuln/cmd/govulncheck

# Ruby
bundle audit check --update                # gem install bundler-audit

# PHP
composer audit

Decision Logic

If the user chooses to patch:

1. Apply fixes (npm audit fix, pip install --upgrade <​pkg​>, etc.)
2. Run audit again to confirm clean
3. Commit as a standalone commit:

chore(deps): fix [critical|high] security vulnerabilities — <​brief description​>

Stage 3 — Pre-coding Quality Checks

3.1 Check Memory for User Preferences

Before running any tools, check if there is a stored preference from previous sessions:

• skip_pre_coding_format: true → skip formatter dry-run
• skip_pre_coding_lint: true → skip linter check-only pass
• format_commit_threshold: <​N​> → override the default threshold of 10

If no memory exists, proceed with defaults and store the user’s decisions at the end of this stage (see Stage 6).

3.2 Formatter Dry-Run

Detect the formatter from config files:

# Prettier (JS/TS/CSS/HTML/JSON/YAML)
npx prettier --list-different "**/*.{js,ts,jsx,tsx,css,json,yaml,md}"

# Black (Python)
black --check --diff .

# Rustfmt
cargo fmt -- --check

# gofmt
gofmt -l ./...

# PHP CS Fixer
php-cs-fixer fix --dry-run --diff

# rubocop (Ruby)
rubocop --auto-correct-all --dry-run

Count the number of files that would be reformatted.

If count ​> threshold (default: 10):

​> 🛑 Formatting check found N files that would be reformatted. ​> Creating a dedicated formatting commit before implementing your task has these benefits: ​> Clean diffs — reviewers see only your feature changes, not whitespace noise ​> Clearer Git history — formatting is a separate, revertable concern ​> Easier code review — no mixing of stylistic and semantic changes ​> Reduced merge conflicts — formatting is settled before feature branching ​> ​> ​> Would you like to: ​> 1. Apply formatting now in a dedicated commit, then proceed with your task ​> 2. Skip formatting for now and proceed directly with the task ​> 3. Apply formatting only to files touched by the task (post-coding) ​> ​>

Store the user’s choice for future sessions.

3.3 Linter Check-Only

Detect and run:

# ESLint
npx eslint . --format=compact 2​>&1 | tail -1    # count violations

# Pylint / Flake8 / Ruff
ruff check . --statistics
flake8 . --statistics --count

# Clippy (Rust)
cargo clippy -- -D warnings 2​>&1 | grep -c "^error"

# golint / staticcheck (Go)
staticcheck ./...

# rubocop (Ruby)
rubocop --format progress | tail -3

Classify output:

If user chooses to fix lint:

1. Run auto-fix where safe: npx eslint . --fix, ruff check . --fix, cargo clippy --fix
2. Manually address remaining issues
3. Commit:

chore(lint): resolve linting violations before feature work

Stage 4 — CI/CD Pre-validation

4.1 Detect CI Configuration

ls .github/workflows/     # GitHub Actions
ls .gitlab-ci.yml         # GitLab CI
ls Jenkinsfile            # Jenkins
ls .circleci/config.yml   # CircleCI
ls .travis.yml            # Travis CI
ls azure-pipelines.yml    # Azure DevOps
ls bitbucket-pipelines.yml
ls .drone.yml

4.2 Extract and Mirror CI Steps Locally

Parse the CI config and identify steps that can be run locally. Common patterns to extract:

# From GitHub Actions: find "run:" steps
grep -A1 "run:" .github/workflows/*.yml

# Run them in the same order as CI

Example mapping (GitHub Actions → local equivalents):

# CI step:            # Local equivalent:
npm ci               →   npm ci
npm run build        →   npm run build
npm test             →   npm test
npm run lint         →   npx eslint .
npm run type-check   →   npx tsc --noEmit

For each detected CI step, run it locally and capture exit codes. If any step fails:

​> ⚠️ CI pre-validation failure: `<​step name​>​> The following CI step failed locally:<​command​>​> Exit code:<​N​>​> Output:<​last 20 lines​>` ​> This will likely cause CI to fail after push. ​> Would you like to: ​> 1. Fix the issue now before implementing the task ​> 2. Note it and proceed (I’ll flag this again at post-coding validation) ​> 3. Abort ​> ​>

4.3 Pre-commit / Pre-push Hook Detection

ls .git/hooks/pre-commit .git/hooks/pre-push
ls .husky/pre-commit .husky/pre-push            # Husky
ls .lefthook.yml lefthook.yml                   # Lefthook
ls .pre-commit-config.yaml                      # pre-commit framework

If hooks exist, run them now as a pre-flight check:

# Native hooks
bash .git/hooks/pre-commit
bash .git/hooks/pre-push

# Husky
npx husky run pre-commit

# pre-commit framework
pre-commit run --all-files

# Lefthook
lefthook run pre-commit

If hooks fail, stop and report. Hooks represent the project’s own quality gate — do not bypass them.

Stage 5 — Refactor Safety Net (Major Refactors Only)

Trigger this stage if the task involves:

• Renaming or moving multiple files/modules
• Changing a public API or interface
• Modifying database schemas or migrations
• Restructuring directory layout
• Changing build configuration or dependency graph

5.1 Document Before Touching Code

Before writing a single line, output and store:

## Refactor Safety Net

**Task:** <​description of refactor​>
**Date:** <​ISO timestamp​>

### Revert Command
 ```bash
git stash         # if not yet committed
# OR
git revert <​SHA​>  # if committed
# OR
git reset --hard <​SHA-before-changes​>

Known Risks

• <​Risk 1: e.g., “API consumers outside this repo may break”​>
• <​Risk 2: e.g., “Database migration is irreversible without a down migration”​>
• <​Risk 3: e.g., “Cache keys will be invalidated — plan for cold cache on deploy”​>

Rollback Verification Steps

1. Run: <​command to verify system is working post-rollback​>
2. Check: <​what to look for in logs or UI​>
3. Confirm: <​specific assertion that proves rollback succeeded​>

Present this to the user and ask for confirmation before proceeding.

### 5.2 Create a Pre-refactor Tag

```bash
git tag pre-refactor/<​task-slug​>-$(date +%Y%m%d)
git push origin pre-refactor/<​task-slug​>-$(date +%Y%m%d)

This gives a named rollback point that survives branch resets.

Stage 6 — Implementation

6.1 Store User Preferences from Prior Stages

At this point, persist the following to memory for future workflow runs:

skip_pre_coding_format: <​true|false​>
skip_pre_coding_lint: <​true|false​>
format_commit_threshold: <​N​>
audit_severity_threshold: <​critical|high|moderate​>

6.2 Define Scope

Before writing code, state clearly:

Task:   <​one-line description of what is being implemented​>
Files:  <​exhaustive list of files to modify​>
Scope:  <​what is explicitly OUT of scope for this change​>

6.3 Implementation Rules

• Follow existing project architecture, patterns, and naming conventions
• Do not introduce formatting changes in files unrelated to the task
• Do not refactor code outside the defined scope
• Do not modify test fixtures, mocks, or test utilities unless the task requires it
• Preserve existing error handling patterns
• Keep new public APIs consistent with existing ones
• Do not stage any changes yet — post-coding validation comes first

Stage 7 — Post-coding Validation

Run every applicable tool against only the files that were changed. Fix issues that are safe and mechanical (e.g., trailing whitespace, import ordering). For anything requiring judgment, stop and ask the user.

​> Context Guard: When running linters, tests, or builds, DO NOT output the full log if it exceeds 50 lines. Use | tail -n 50 or pipe to a temporary file and grep for “Error” or “Fail” to avoid crashing the agent context window.

7.1 Formatter (Write Mode)

# Only on changed files
git diff --name-only | xargs npx prettier --write
git diff --name-only | xargs black

7.2 Linter

git diff --name-only | xargs npx eslint --fix
git diff --name-only | xargs ruff check --fix

If unfixable violations remain after auto-fix, stop and report each one with file, line, and rule name. Ask: fix manually, suppress with justification, or abort?

7.3 Type Checker

npx tsc --noEmit                    # TypeScript
mypy <​changed_files​>                # Python
cargo check                         # Rust
go vet ./...                        # Go

Type errors are blocking — do not proceed past this step with type errors.

7.4 Tests & Coverage

Detect Testing Framework

# Node
ls jest.config.* vitest.config.* .mocharc.*
grep -E '"test":|"jest":|"vitest"' package.json

# Python
ls pytest.ini setup.cfg pyproject.toml | xargs grep -l '\[tool.pytest'
ls tests/ test_*.py *_test.py

# Rust
grep '\[dev-dependencies\]' Cargo.toml

# Go — built-in
ls *_test.go

# Ruby
ls spec/  test/

Run Tests

# Run full suite first (with context guard)
npm test 2​>&1 | tail -n 50
pytest | tail -n 50
cargo test
go test ./...
bundle exec rspec

# If full suite passes, run coverage for changed files
npx jest --coverage --collectCoverageFrom='<​changed_files​>'
pytest --cov=<​module​> --cov-report=term-missing <​changed_files​>

Coverage Rules

If coverage drops below the project’s configured threshold (e.g., coverageThreshold in jest.config), stop and ask whether to write additional tests or document the coverage gap.

Test Failure Protocol

• First failure: Attempt to fix
• Second consecutive failure on same test: Stop, show full output, ask user how to proceed

7.5 Build

npm run build
cargo build --release
go build ./...
python -m build
mvn package -q

Build failures are blocking.

7.6 Pre-commit and Pre-push Hooks (Final Run)

Run all detected hooks one final time against the actual changed files:

pre-commit run --files $(git diff --name-only)
bash .git/hooks/pre-commit

If hooks fail here, the commit will fail anyway — fix issues before staging.

Stage 8 — Secret Detection

Always run before staging. Never skip.

# gitleaks (recommended — install: brew install gitleaks / apt install gitleaks)
gitleaks detect --source . --no-git

# truffleHog
trufflehog filesystem .

# detect-secrets (Python)
detect-secrets scan --baseline .secrets.baseline

# git-secrets (AWS focused)
git secrets --scan

# Fallback: manual pattern grep if no tool is available
grep -rE \
  '(AKIA[0-9A-Z]{16}|sk-[a-zA-Z0-9]{32,}|ghp_[a-zA-Z0-9]{36}|-----BEGIN (RSA|EC|OPENSSH) PRIVATE KEY|password\s*=\s*["\x27][^"\x27]{4,}["\x27])' \
  $(git diff --name-only)

Secret Found Protocol

​> 🚨 Potential secret detected in <​filename​> at line `<​N​>​> Pattern matched:<​redacted — show only type, not value​>​> **DO NOT COMMIT.** Options: ​> 1. Replace with environment variable reference and add to.env.example​> 2. Move to a secrets manager (Vault, AWS Secrets Manager, etc.) ​> 3. Mark as false positive and add to.secrets.baseline` ​> ​>

Do not proceed to staging until all findings are resolved.

Stage 9 — CHANGELOG Check

9.1 Detect CHANGELOG

ls CHANGELOG.md CHANGELOG.rst CHANGELOG.txt CHANGES.md HISTORY.md RELEASES.md 2​>/dev/null

9.2 Evaluation Rules

9.3 CHANGELOG Entry Format

Use the project’s existing format. If none exists, default to Keep a Changelog:

## [Unreleased]

### Added
- <​feature description​> — <​brief rationale​>

### Fixed
- <​bug description​> — fixes #<​issue​> if applicable

### Changed
- <​what changed and why​>

If user confirms an entry is needed, add it and stage CHANGELOG.md alongside the feature files in the same commit (not a separate commit — changelog belongs with the change).

Stage 10 — Version Control & Commit

10.1 Inspect Final Diff

git status
git diff --stat

Classify every modified file:

10.2 Final Secret Scan on Staged Files Only

git stash       # stash unstaged
git diff --cached | gitleaks detect --no-git --pipe
git stash pop   # restore unstaged

10.3 Stage Files

git add <​file1​> <​file2​> ...    # Always stage explicitly, never `git add .`
git diff --cached --stat       # Confirm staged set looks correct

10.4 Commit Message

Follow Conventional Commits format. Craft a commit that answers:

• What changed (summary line, imperative mood, ≤72 chars)
• Why it was changed (body)
• What problem it solves (body)
• Architectural/behavioral impact if any (body or footer)
• References — issue/PR numbers, related commits

<​type​>(<​scope​>): <​short imperative summary​>

<​Paragraph explaining WHY the change was made and WHAT problem it solves.
Wrap at 72 characters. Use present tense. Be specific — avoid "fix bug".​>

<​Optional paragraph for architectural impact, migration steps, or
behavioral changes users/consumers should be aware of.​>

Refs: #<​issue​>, #<​pr​>
Breaking-Change: <​description if applicable​>

Type vocabulary:

10.5 Push

git push origin <​branch​>

If the push is rejected (non-fast-forward), stop and ask the user before force-pushing. Never force-push to main, master, or production branches.

Stage 11 — Pull Request (Optional)

If gh (GitHub) or glab (GitLab) CLI tools are installed and authenticated:

GitHub

gh pr create \
  --title "<​commit summary​>" \
  --body "<​commit body​>" \
  --base main \
  --head <​current-branch​> \
  --draft  # Ask user: Draft or Ready?

GitLab

glab mr create \
  --title "<​commit summary​>" \
  --description "<​commit body​>" \
  --source-branch <​current-branch​>

If no CLI tool is found, generate the HTTPS link for the user to click to create the PR manually (e.g., https://github.com/org/repo/compare/main...feature-branch).

Error Recovery Playbook

Commit Sequence Summary

A full workflow may produce multiple commits in this order:

1. chore: update .gitignore                      (if needed)
2. chore(deps): fix security vulnerabilities     (if audit findings patched)
3. style: apply formatter to codebase            (if user approved pre-coding format)
4. chore(lint): resolve linting violations       (if user approved pre-coding lint)
5. feat/fix/refactor(<​scope​>): <​main task​>       (the actual work + changelog + tests)