Agent Skills: Post-Implementation Procedure

Post-Implementation Procedure

Execute quality review, refactoring, and finalization after code implementation is complete.

Prerequisites

• Code implementation complete (files created/modified)
• Working directory in project root
• Optional: List of files to review (if not provided, detect from git status)

Phase 1: Quality Review

Goal: Find real issues in implemented code with confidence-scored findings.

Step 1: Collect modified files

Obtain list of all created/modified files:

git status --short
git diff --name-only HEAD

Or use file list provided by plan/user.

Step 2: Launch parallel reviewers

Launch 3 code-reviewer agents simultaneously with different review focuses:

Reviewer 1 - Simplicity / DRY / Elegance

Task(
  subagent_type: "smart-plan:code-reviewer",
  prompt: "Review the following files for code quality:
[list of files]

Focus: simplicity, DRY principle, code elegance, unnecessary complexity.
Only report findings with confidence >= 80%.
Include concrete fix suggestions for each finding."
)

Reviewer 2 - Bugs / Functional Correctness

Task(
  subagent_type: "smart-plan:code-reviewer",
  prompt: "Review the following files for bugs and logic errors:
[list of files]

Focus: bugs, logic errors, edge cases, error handling, race conditions, null safety.
Only report findings with confidence >= 80%.
Include concrete fix suggestions for each finding."
)

Reviewer 3 - Conventions / Abstractions

Task(
  subagent_type: "smart-plan:code-reviewer",
  prompt: "Review the following files for convention adherence:
[list of files]

Project conventions:
[key conventions from project CLAUDE.md, package.json, or codebase patterns]

Focus: naming conventions, architectural patterns, import organization, abstraction quality.
Only report findings with confidence >= 80%.
Include concrete fix suggestions for each finding."
)

Step 3: Consolidate findings

1. Collect all findings from 3 reviewers
2. Deduplicate overlapping issues
3. Filter: keep only findings with confidence >= 80%
4. Present consolidated review to user

Step 4: Display results

Present consolidated findings to user with:

• File path and line number
• Issue description
• Confidence score
• Concrete fix suggestion

Phase 2: Refactoring

Goal: Auto-fix review findings with high confidence.

Step 1: Evaluate findings

If findings with confidence >= 80% exist from Phase 1:

Launch code-refactorer agent:

Task(
  subagent_type: "smart-plan:code-refactorer",
  prompt: "Apply the following corrections to the codebase:

[list of findings with confidence >= 80%, including file paths, line numbers, and suggested fixes]

After applying ALL corrections:
1. Run build/compile check
2. Run tests (if test suite exists)
3. Run linter (if configured)

Report all changes made and validation results."
)

Step 2: Review refactorer report

Review refactorer's report:

• If validation passed: proceed to finalization
• If validation failed: present failures to user for decision

If no findings >= 80%:

• Skip this phase
• Inform user code passed review without critical issues

Phase 3: Finalization

Goal: Document, summarize, and optionally commit.

Step 1: Write feature summary

Derive kebab-case name from feature description (e.g., "add-user-authentication").

If feature was planned with smart-plan, use plan name. Otherwise, generate descriptive name.

Write summary to .claude/features/<feature-name>.md:

# <Feature Name>

## What Was Built

- [Feature description and scope]

## Files Created

- [List with brief description of each]

## Files Modified

- [List with brief description of changes]

## Review Results

- [Summary of review findings and fixes applied]
- [Number of issues found and auto-fixed]
- [Manual fixes needed (if any)]

## Dependencies Added

- [New packages installed, if any]

## Next Steps

- [Suggested follow-up actions]
- [Tests to add or expand]
- [Documentation to write or update]

Step 2: Present summary to user

Show feature summary file path and brief summary of accomplishments.

Step 3: Ask about committing changes

Use AskUserQuestion to determine if user wants to commit:

If yes:

1. Follow user's git conventions (conventional commits)
2. Stage specific files (avoid git add . or git add -A)
3. Single-line commit message (max 96 chars)
4. Execute git status to verify commit success

If no:

• Inform user changes are ready for manual review

Step 4: Suggest next steps

Provide potential follow-up actions:

• Tests to add or expand
• Documentation to write or update
• Related features to consider
• Performance optimizations
• Security hardening

Orchestration Rules

Progress tracking:

• If tasks created for phases exist, update status (in_progress/completed) at phase start/end
• Consolidate agent outputs before presenting to user

User interaction:

• Be transparent about current phase and actions
• Present consolidated findings, not raw agent outputs
• Request approval at key decision points (commit in Phase 3)

Failure handling:

• If agent fails, inform user and offer retry or adjustment
• If build/test validation fails, present failures and request guidance
• Never proceed silently when errors occur

Common Issues

Issue: Cannot find modified files

Solution: Execute git status to detect changes, or ask user for file list

Issue: Review produces too many low-confidence findings

Solution: Filter strictly for confidence >= 80%; present only actionable issues

Issue: Refactorer breaks tests

Solution: Refactorer should run tests; if tests fail, present to user for manual fix

Issue: No detectable project conventions

Solution: Search for CLAUDE.md, README.md, or analyze codebase patterns; ask user if needed

Summary

This skill executes a structured post-implementation workflow:

1. Quality Review: Review code with 3 parallel reviewers (confidence >= 80%)
2. Refactoring: Auto-fix findings with code-refactorer agent
3. Finalization: Document feature, optionally commit, suggest next steps

Can be used after any implementation, not just smart-plan. Be transparent with user. Fail gracefully when issues occur.