Implement

Implement Workflow

This skill handles the full lifecycle: plan -> implement -> verify -> commit -> draft PR.

Phase 1: Plan (requires approval)

1. Understand the request — Read the issue description, linked context, or user instructions.
2. Explore the codebase — Find relevant files, understand existing patterns, read tests.
3. Create an implementation plan:
   • What files will be modified/created
   • What the changes will do
   • What edge cases to consider
   • What tests to add or update
4. Present the plan and STOP — Wait for explicit user approval before proceeding.

Do NOT start coding until the plan is approved. If requirements are ambiguous, ask.

Phase 2: Implement

Execute the approved plan:

• Follow existing code patterns and conventions in the repo
• Keep changes minimal and focused — don't refactor unrelated code
• Kotlin warnings are treated as errors (allWarningsAsErrors = true) — write clean code
• Use org.wordpress.aztec package conventions
• Unit tests: Add or update tests covering new/changed logic. Follow existing test patterns in the repo. If writing meaningful tests would require disproportionate effort (e.g., complex setup, heavy mocking of framework internals), skip but notify the developer explaining why.

Phase 3: Verify

Run verification checks and fix any failures. Iterate until all checks pass.

3a: Compile

./gradlew :aztec:assembleRelease :app:assembleDebug

If other modules were changed, compile those too:

./gradlew assembleRelease

3b: Lint

./gradlew ktlint

• Fix violations — prefer real fixes over suppression
• Only suppress when it's a false positive and document why

3c: Unit Tests

./gradlew :aztec:testRelease

Or run specific tests related to the changes:

./gradlew :aztec:testReleaseUnitTest --tests "org.wordpress.aztec.SpecificTest" --info

Test rules:

• NEVER weaken or remove assertions to make tests pass
• Don't make unrelated or incorrect changes to production code solely to appease a test. If a test reveals a real issue, fix it properly.
• Tests that pass only by not crashing are invalid — every test needs meaningful assertions
• If a test won't pass after reasonable attempts: stop and ask

3d: Fix Errors

If any step fails:

1. Analyze the error
2. Fix the issue
3. Re-run the failing check
4. Repeat until green

Phase 4: Present Changes (requires approval)

Show a summary of all changes made:

• Files modified/created
• Key behavioral changes
• Test coverage

STOP and wait for user approval before committing.

Phase 5: Commit and Draft PR

Only proceed after explicit approval.

5a: Run Final Lint

./gradlew ktlint

Fix any remaining issues.

5b: Inspect Changes

git status
git diff --stat
git diff

5c: Plan Commits

Review the changes and determine if they should be split into multiple commits:

• Independent logical units = separate commits
• Bug fix + feature = separate commits
• Formatting + logic = separate commits

For each commit, prepare:

1. Commit message: Imperative summary + brief body
2. Files: Paths to stage
3. Summary: What and why

Commit message format — use direct multi-line strings:

git commit -m "Imperative summary

- Detail one
- Detail two
"

Rules:

• NO co-author lines — NEVER add "Co-Authored-By" or AI attribution
• Each commit should be cohesive and buildable
• Use git add -p to split mixed concerns if needed

5d: Stage and Commit

git add <specific files>
git commit -m "message"

5e: Push and Create Draft PR

# Get the correct remote owner/repo
git remote get-url origin

# Push
git push -u origin HEAD

# Create draft PR
PAGER=cat gh pr create --draft --title "PR title" --body "$(cat <<'EOF'
### Fix
<description>

### Test
1. Step 1
2. Step 2
EOF
)"

PR rules:

• Title: short, under 70 characters
• Body: follows the repo's PR template format (### Fix, ### Test, ### Review)
• Always create as draft
• Return the PR URL when done

Handling Edge Cases

Working on an issue with a branch name

If the user mentions an issue with a known branch name:

git checkout trunk && git pull && git checkout -b <branch-name>

Determining diff base for existing PRs

PRs can be stacked — check the actual base:

PAGER=cat gh pr view <NUMBER> --json baseRefName

Posting review comments

When reviewing or addressing PR feedback:

# Create review JSON
printf '%s\n' '{
  "event": "COMMENT",
  "body": "Review comment",
  "comments": [
    {"path": "file.kt", "line": 42, "body": "Inline comment"}
  ]
}' > /tmp/pr_review.json

PAGER=cat gh api repos/{owner}/{repo}/pulls/{number}/reviews --method POST --input /tmp/pr_review.json