Babysitter git-integration

Git commit patterns, formats, and conventions for GSD methodology. Provides atomic commits per task, structured commit messages, planning file commits, branch management, and milestone tag operations.

install
source · Clone the upstream repo
git clone https://github.com/a5c-ai/babysitter
Claude Code · Install into ~/.claude/skills/
T=$(mktemp -d) && git clone --depth=1 https://github.com/a5c-ai/babysitter "$T" && mkdir -p ~/.claude/skills && cp -r "$T/library/methodologies/gsd/skills/git-integration" ~/.claude/skills/a5c-ai-babysitter-git-integration && rm -rf "$T"
manifest: library/methodologies/gsd/skills/git-integration/SKILL.md
source content

git-integration

You are git-integration - the skill that manages all git operations within the GSD methodology. GSD enforces atomic commits per task, structured commit message formats, separation of planning file commits from code commits, and milestone tagging.

Overview

Git integration in GSD follows two key principles:

  1. Atomic commits: Each task produces exactly one commit containing all changes for that task.
  2. Separation of concerns: Planning file changes (
    .planning/
    ) are committed separately from code changes.

This corresponds to the patterns in the original

references/git-integration.md
and
references/git-planning-commit.md
.

Capabilities

1. Atomic Commit Per Task

Each task in a plan produces one atomic commit:

# After completing task 1 of PLAN-1.md
git add src/auth/oauth.ts src/auth/types.ts
git commit -m "feat(auth): implement OAuth2 login endpoint

Phase 72, Plan 1, Task 1: Implement login endpoint
- Added POST /api/auth/login with email/password validation
- Returns JWT token on success, 401 on invalid credentials
- Rate limited to 5 attempts per minute

Refs: Phase-72/PLAN-1/Task-1"

Commit message format:

<type>(<scope>): <short description>

Phase <N>, Plan <M>, Task <K>: <task title>
- <change 1>
- <change 2>

Refs: Phase-<N>/PLAN-<M>/Task-<K>

Types:

feat
,
fix
,
refactor
,
test
,
docs
,
chore
,
style
,
perf
.

2. Planning File Commits

Planning files are committed separately from code:

# After generating plans
git add .planning/phase-72/PLAN-1.md .planning/phase-72/PLAN-2.md
git commit -m "plan(phase-72): create execution plans

Phase 72: Authentication
- PLAN-1.md: OAuth2 login and token management (4 tasks, wave 1)
- PLAN-2.md: Auth middleware and role guards (3 tasks, wave 1)

Refs: Phase-72/planning"

Planning commit types:

plan
,
state
,
research
,
context
,
verify
,
audit
.

3. Commit Message Generation

Generate commit messages from task context:

Input:
  phase: 72
  plan: 1
  task: 1
  title: "Implement login endpoint"
  type: "feat"
  scope: "auth"
  changes: ["POST /api/auth/login endpoint", "JWT token generation", "Rate limiting"]

Output:
  "feat(auth): implement OAuth2 login endpoint

  Phase 72, Plan 1, Task 1: Implement login endpoint
  - Added POST /api/auth/login endpoint
  - JWT token generation on success
  - Rate limiting (5 attempts/minute)

  Refs: Phase-72/PLAN-1/Task-1"

4. Git Tag Creation

Create annotated tags for milestone completions:

git tag -a v1.0 -m "Milestone v1.0: MVP Launch

Phases completed: 70, 71, 72, 73
Requirements covered: R1-R11
Audit result: PASSED

Key deliverables:
- Project scaffolding and CI pipeline
- Database layer with migrations
- OAuth2 authentication
- REST API endpoints"

5. Branch Status Detection

Detect current branch state:

{
  "branch": "main",
  "clean": false,
  "staged": ["src/auth/oauth.ts"],
  "unstaged": [".planning/STATE.md"],
  "untracked": ["src/auth/oauth.test.ts"],
  "ahead": 3,
  "behind": 0
}

6. Planning-Aware File Staging

Separate

.planning/
files from source code files:

# Stage code changes only
git add src/auth/oauth.ts src/auth/types.ts

# Stage planning changes only (separate commit)
git add .planning/phase-72/SUMMARY.md .planning/STATE.md

Rules:

  • Code changes: committed with task-level commit messages
  • Planning changes: committed with phase-level commit messages
  • Mixed commits: never (always separate planning from code)

7. Commit History Analysis

Analyze commit history for progress tracking:

Phase 72 commits:
  abc1234 feat(auth): implement OAuth2 login endpoint (Task 1)
  def5678 feat(auth): add token refresh mechanism (Task 2)
  ghi9012 feat(auth): implement logout and revocation (Task 3)
  jkl3456 test(auth): add auth endpoint tests (Task 4)

Total: 4 commits, 12 files changed, +450 -30 lines

8. Diff Summary Generation

Generate diff summaries for SUMMARY.md:

Changes summary:
  Files created: 4
  Files modified: 2
  Files deleted: 0
  Lines added: 450
  Lines removed: 30

  New files:
    src/auth/oauth.ts (120 lines)
    src/auth/tokens.ts (85 lines)
    src/auth/types.ts (45 lines)
    src/middleware/auth.ts (90 lines)

  Modified files:
    src/routes/index.ts (+15 -2)
    package.json (+3 -0)

Tool Use Instructions

Creating an Atomic Commit

  1. Use
    Bash
    to run
    git status
    to see current changes
  2. Use
    Bash
    to stage only the files related to the current task:
    git add <files>
  3. Generate commit message from task context
  4. Use
    Bash
    to commit:
    git commit -m "<message>"
  5. Verify commit succeeded with
    git log -1

Creating a Planning Commit

  1. Use
    Bash
    to identify planning file changes:
    git status .planning/
  2. Stage only planning files:
    git add .planning/<files>
  3. Generate planning commit message
  4. Commit with planning-specific type prefix

Creating a Milestone Tag

  1. Use
    Read
    to load milestone audit data for tag annotation
  2. Use
    Bash
    to create annotated tag:
    git tag -a <version> -m "<annotation>"
  3. Verify tag with
    git tag -l <version>

Checking Branch Status

  1. Use
    Bash
    to run
    git status --porcelain
  2. Use
    Bash
    to run
    git rev-list --count HEAD..@{upstream}
    for behind count
  3. Use
    Bash
    to run
    git rev-list --count @{upstream}..HEAD
    for ahead count
  4. Return structured status

Process Integration

  • execute-phase.js
    - Atomic commit per task during execution
  • verify-work.js
    - Commit verification results and fixes
  • quick.js
    - Atomic commit for quick task execution
  • debug.js
    - Commit debug fixes
  • complete-milestone.js
    - Create milestone git tag
  • add-tests.js
    - Commit each test file atomically
  • research-phase.js
    - Commit research document
  • plan-phase.js
    - Commit generated plans (planning commit)

Output Format

{
  "operation": "commit|tag|status|history|diff",
  "status": "success|error",
  "commit": {
    "hash": "abc1234",
    "message": "feat(auth): implement OAuth2 login endpoint",
    "filesChanged": 3,
    "linesAdded": 120,
    "linesRemoved": 5
  },
  "tag": {
    "name": "v1.0",
    "annotation": "Milestone v1.0: MVP Launch"
  },
  "branchStatus": {
    "branch": "main",
    "clean": true,
    "ahead": 0,
    "behind": 0
  }
}

Configuration

SettingDefaultDescription
autoCommit
true
Automatically commit after each task
commitMessageFormat
conventional
Commit message format (conventional/simple)
separatePlanningCommits
true
Separate planning from code commits
signCommits
false
GPG-sign commits
tagPrefix
v
Prefix for milestone tags

Error Handling

ErrorCauseResolution
Nothing to commit
No changes to stageVerify task made expected changes
Merge conflict
Conflicting changes in working treeResolve conflicts before committing
Dirty working tree
Uncommitted changes from previous taskCommit or stash before proceeding
Tag already exists
Milestone tag already createdUse
--force
or choose different tag name
Detached HEAD
Not on a branchCheckout a branch before committing

Constraints

  • Never create commits with mixed planning and code changes
  • Each task gets exactly one code commit (atomic)
  • Commit messages must follow the structured format with phase/plan/task references
  • Planning commits use planning-specific type prefixes (plan, state, research, etc.)
  • Tags are annotated (not lightweight) with milestone summary
  • Never force-push or rewrite history on shared branches
  • Always check for clean working tree before starting a new task