Send: Review, Commit, Push, and Merge/Pull Request

A workflow that takes local changes from working directory to merge/pull request. Each phase gates the next -- if something fails, stop and report rather than pushing broken code upstream.

IMPORTANT: All git commands in this skill MUST use --no-pager to prevent interactive paging from hanging the shell.

Argument Handling

Parse $ARGUMENTS (appended at the bottom of this skill) to extract the user's intent.

• No arguments: Proceed normally -- auto-generate everything.
• Arguments provided (e.g., feat: add repo argument support to prime command): Store as INTENT and use it throughout:
  • Branch name (Phase 0d): derive a slug from the intent
  • Commit message (Phase 3): use as the commit subject if it follows conventional commit format (type: description or type(scope): description), otherwise use it as input to the commit tool
  • PR/MR title (Phase 5): use as the title

Deriving a branch slug from intent: Strip conventional commit prefix if present (feat: , fix(scope): , etc.), lowercase, replace spaces and special characters with hyphens, collapse consecutive hyphens, truncate to 50 characters, strip leading/trailing hyphens.

Example: feat: add repo argument support to prime command → add-repo-argument-support-to-prime

Prerequisites

• git -- version control
• Remote CLI (auto-detected from remote URL):
  • GitLab remotes (opengroup.org, gitlab): requires glab
  • GitHub remotes (github.com): requires gh
• Optional (graceful degradation if absent):
  • wt (worktrunk) -- worktree-based branching and commits
  • aipr -- AI-powered commit message generation

Sandbox note: In bare repo worktree layouts (.bare/ directory), branch creation writes to .bare/refs/heads/. This may require sandbox write access to the repo's .bare/ path.

Phase 0: Preflight

0a. Check for changes (bail early if nothing to send)

git --no-pager status --short

If clean, also check for unpushed commits:

git --no-pager log --oneline @{upstream}..HEAD 2>/dev/null

If both are clean, STOP -- nothing to send. If there are unpushed commits but no local changes, skip to Phase 4 (Push).

0b. Detect environment

Run these checks. The results gate all later phases.

Remote platform:

REMOTE_URL=$(git remote get-url origin 2>/dev/null)
if echo "$REMOTE_URL" | grep -q "github\.com"; then
  PLATFORM=github
elif echo "$REMOTE_URL" | grep -qE "opengroup\.org|gitlab"; then
  PLATFORM=gitlab
else
  PLATFORM=unknown
fi

Bare repo / worktree layout:

IS_BARE_WORKTREE=false
if [ -f .git ]; then
  IS_BARE_WORKTREE=true
fi

This matters for branch naming -- bare repos cannot use slashes in branch names because they create subdirectories under .bare/refs/heads/ which may conflict or fail.

Worktree tool:

if command -v wt >/dev/null 2>&1 || command -v git-wt >/dev/null 2>&1; then
  HAS_WT=true
else
  HAS_WT=false
fi

Commit tool (prefer wt step commit > aipr > manual):

if [ "$HAS_WT" = true ]; then
  COMMIT_TOOL=wt
elif command -v aipr >/dev/null 2>&1; then
  COMMIT_TOOL=aipr
else
  COMMIT_TOOL=manual
fi

0c. Get the current branch

git branch --show-current

0d. Branch safety

Branch naming rules (gated by IS_BARE_WORKTREE):

• IS_BARE_WORKTREE=true: use flat names with hyphens (e.g., add-repo-argument-support)
• IS_BARE_WORKTREE=false: use slashed names (e.g., feature/add-repo-argument-support)

On main or master: These are release branches -- changes must go on a feature branch.

If INTENT is available (from ARGUMENTS), auto-derive a branch name from the slug and create the branch without asking the user:

# Example for bare worktree with INTENT slug "add-repo-argument-support"
git checkout -b add-repo-argument-support
# Example for standard clone
git checkout -b feature/add-repo-argument-support

If no INTENT, ask the user for a feature name, then create the branch.

On dev: The user is on the integration branch -- cannot merge from dev to dev. Same logic as above: auto-derive from INTENT if available, otherwise ask.

When using wt for branch creation:

# wt always uses flat names for worktree directory names
wt switch --create <slug> --base dev

This changes the working directory to a new worktree path. All subsequent commands run from the new directory.

On feature/* or any other name: proceed normally.

0e. Contribution check -- am I on someone else's branch?

GitLab:

glab api "projects/:id/merge_requests?source_branch=$(git branch --show-current)&state=opened" \
  --hostname community.opengroup.org 2>/dev/null

GitHub:

gh pr list --head "$(git branch --show-current)" --state open --json number,author 2>/dev/null

If an open MR/PR exists and the author is not the current user, ask: "You're on the <branch> branch from MR/PR #X by <author>. Do you want to contribute these changes to that MR/PR, or create a separate one?" If they want to contribute, hand off to the contribute skill.

Phase 1: Lite Code Review

A quick sanity check -- catch obvious problems before they become review comments.

1. View the full diff:

   git --no-pager diff --stat
   git --no-pager diff
   

2. Scan the changes for:
   • Hardcoded secrets, credentials, API keys, .env files
   • Files that should not be committed: binaries, .tfstate, .env, credential files
   • Obvious bugs or logic errors
3. Present a brief review summary listing changed files and any concerns.
4. If there are blocking concerns (secrets, dangerous files), STOP and ask the user to fix them before continuing.

Phase 2: Quality Checks

Run checks based on which file types changed -- skip checks that don't apply. Only run a check if the relevant tool is available.

• Terraform (.tf files changed):

  terraform fmt -check -recursive ./infra 2>/dev/null
  terraform fmt -check -recursive ./platform 2>/dev/null
  

• YAML (.yaml or .yml files changed):

  git --no-pager diff --name-only --diff-filter=ACM -- '*.yaml' '*.yml' | xargs -I{} python3 -c "import yaml, sys; yaml.safe_load(open(sys.argv[1]))" {}
  

• JSON (.json files changed):

  git --no-pager diff --name-only --diff-filter=ACM -- '*.json' | xargs -I{} python3 -c "import json, sys; json.load(open(sys.argv[1]))" {}
  

• Python (.py files changed) -- syntax check only:

  git --no-pager diff --name-only --diff-filter=ACM -- '*.py' | xargs -I{} python3 -m py_compile {}
  

• Java (pom.xml exists and .java files changed) -- compile check only:

  mvn compile -q 2>/dev/null
  

If any check fails, STOP and report. Do not proceed to commit.

Phase 3: Commit

Stage changes and create a conventional commit. The method depends on which tools are available (detected in Phase 0b). If INTENT is available and already follows conventional commit format, prefer using it as the commit message rather than auto-generating -- the user told you what this change is.

Option A: wt step commit (if HAS_WT=true)

worktrunk handles staging, diff analysis, and message generation:

wt step commit

Review the generated message -- if it doesn't follow the commit rules below, amend.

If wt step commit fails (non-zero exit), fall through to Option B or C. Don't stop the workflow because of a tool failure when alternatives exist.

Option B: aipr commit -s (if aipr is available)

git add -A
git commit -m "$(aipr commit -s)"

Option C: Manual commit (fallback)

git add -A

If INTENT is available and matches conventional commit format (type: ... or type(scope): ...), use it directly as the commit message. Otherwise, generate the message from the staged diff using the Commit Prompt Reference.

git commit -m "<message>"

Commit rules -- hard requirements (apply to ALL options)

• One-line summary under 72 characters: type(scope): description
• Types: feat fix docs refactor chore ci style test build perf
• Use imperative mood (add, implement, fix -- not adds, added, adding)
• Add 1-2 detail lines only for large changes (15+ files). Max 3 lines total.
• NEVER add Co-Authored-By trailers -- not for any AI or agent
• NEVER add "Generated with", "Built by", or any agent/AI attribution
• NEVER add Signed-off-by unless the user explicitly requests DCO sign-off

Phase 4: Push

Push the branch to the remote:

git push -u origin $(git branch --show-current)

Phase 5: Merge / Pull Request

5a. Check for an existing MR/PR

GitLab:

glab mr list --source-branch="$(git branch --show-current)"

GitHub:

gh pr list --head "$(git branch --show-current)" --state open

If one already exists, report its URL and skip creation.

5b. Determine the target branch

Default target: dev (OSDU convention). If no dev branch exists on the remote, fall back to main or master:

git --no-pager ls-remote --heads origin dev main master 2>/dev/null

Use the first branch that exists, in order: dev, main, master.

5c. Determine the title

If INTENT is available and follows conventional commit format, use it as the title. Otherwise, derive from the most recent commit (or summarize if multiple commits):

TITLE=$(git --no-pager log -1 --format='%s')

5d. Generate the description

Analyze the commit log and diff stats to produce a description:

DIFF_STATS=$(git --no-pager diff --stat origin/$TARGET_BRANCH..HEAD)
COMMITS=$(git --no-pager log origin/$TARGET_BRANCH..HEAD --format='%s%n%b')

Use the MR Description Prompt Reference as a guide for the description structure. When INTENT is available, incorporate the user's stated purpose into the Summary section as the "why" -- don't just describe the diff mechanically.

5e. Create the MR/PR

GitLab:

ASSIGNEE=$(glab auth status 2>&1 | grep 'Logged in' | sed 's/.* as \([^ ]*\).*/\1/')
glab mr create \
  --title "$TITLE" \
  --description "$BODY" \
  --target-branch "$TARGET_BRANCH" \
  --assignee "$ASSIGNEE" \
  --remove-source-branch

GitHub:

gh pr create \
  --title "$TITLE" \
  --body "$BODY" \
  --base "$TARGET_BRANCH"

5f. Report the MR/PR URL to the user.

Final Summary

After all phases complete, present a compact summary:

Review:  <clean or list of concerns addressed>
Commit:  <short-hash> <commit message>
Branch:  <branch-name> -> pushed to origin
MR/PR:   <URL>

Source: danielscholl/claude-osdu — distributed by TomeVault.