ticket-delivery（门票递送）

Ticket Workflow

End-to-end workflow that takes a ticket ID and drives it through analysis, codebase investigation, planning, TDD implementation, committing, and PR creation. Adapts to your stack via an interactive supplement created on first run.

Phase 0: Environment Supplement

Before anything else, check for a supplement file stored alongside this skill (e.g., .supplement.md in the skill directory).

If supplement exists

Read the file and load all $VARIABLE values for use throughout the workflow.

If supplement is missing

Run an interactive setup. Present an interactive prompt (or the platform's equivalent) for each question, wait for the answer, then proceed to the next. Save all answers to a supplement file stored alongside this skill (e.g., .supplement.md in the skill directory).

Questions to ask (see supplement reference for full details):

1. Ticket system -- GitHub Issues / GitLab Issues / Jira / Linear / Other? Integration tool name for fetching tickets?
2. Version control -- Bitbucket / GitHub / GitLab? CLI tool (gh, glab)? Main branch name?
3. Error tracking -- Bugsnag / Datadog / Sentry / none? Integration tool name?
4. Logging -- CloudWatch / Datadog / ELK / Grafana / none? Tool access or manual paste?
5. Database -- ORM (ActiveRecord / Doctrine / Eloquent / Prisma / TypeORM / none)? Read-only query command?
6. CI/CD hooks -- Pre-commit hooks to skip? (e.g., SKIP=lint,test git commit)
7. Testing framework -- Go test / Jest / PHPUnit / pytest / RSpec / other? TDD style preference (London / Chicago)?
8. Architecture -- Monolith / microservices? Admin panel? API layer (REST / GraphQL / both)?
9. PR conventions -- Branch naming pattern? Commit message format?

After saving, confirm: "Supplement saved. You can edit it anytime at the supplement file path or delete it to re-run setup."

Delegation Strategy

This workflow can use a multi-agent team when available. The coordinating agent delegates to specialized sub-agents for parallel investigation.

Approach

Delegate to specialized sub-agents if available. Use delegation only when the ticket scope justifies parallel investigation. For small bug fixes, a single agent can handle everything alone.

Phase 1: Ticket Intelligence Gathering

Step 1.1 -- Fetch Ticket Details

If $TICKET_MCP_TOOL is configured:

Use $TICKET_MCP_TOOL to fetch ticket $TICKET_ID

If no integration tool is configured: Ask the developer: "Please paste the ticket title, description, and acceptance criteria."

Step 1.2 -- Parse Ticket Information

Extract and organize:

• Title and description
• Acceptance criteria (numbered list)
• Ticket type (bug / feature / improvement / refactor)
• Priority and linked tickets
• Attachments or screenshots (note them for later)

Step 1.3 -- Fetch Error Context (bugs only)

If $ERROR_TRACKING_MCP_TOOL is configured and ticket is a bug:

Use $ERROR_TRACKING_MCP_TOOL to fetch error details linked in the ticket

Extract: stack trace, affected file paths, frequency, first/last seen.

If no integration tool is configured: Ask developer to paste relevant error details or stack traces.

Step 1.4 -- Fetch Log Context (if applicable)

If $LOGGING_TOOL is configured: Fetch relevant log entries around the error timeframe.

If manual: Ask developer to paste relevant log snippets if available.

Step 1.5 -- Summarize Findings

Present a structured summary:

TICKET: $TICKET_ID
TYPE: [bug|feature|improvement|refactor]
SUMMARY: [one-line summary]
ACCEPTANCE CRITERIA: [numbered list]
ERROR CONTEXT: [if bug -- stack trace summary, affected files]
INITIAL HYPOTHESES: [2-3 possible approaches]

Ask the developer: "Does this look correct? Any additional context before I investigate the codebase?"

Phase 2: Branch Setup

git fetch origin
git checkout $MAIN_BRANCH
git pull origin $MAIN_BRANCH

Do NOT create a feature branch yet -- that happens in Phase 5 after planning.

Phase 3: Codebase Investigation

Investigate the codebase to understand the area affected by the ticket. Adapt the investigation to your project's architecture ($ARCHITECTURE).

Investigation Areas

Run these in parallel when using sub-agents, or sequentially as a single agent:

3a -- Entity / Model Layer

• Find domain models or entities related to the ticket
• Check properties, relationships, validation rules
• If using an ORM ($ORM): check mappings, migrations, schema

3b -- Service / Business Logic Layer

• Find services, handlers, use cases related to the ticket
• Trace the execution flow from entry point to persistence
• Identify dependencies and collaborators

3c -- API Layer (if applicable)

• REST: Find controllers, routes, request/response DTOs
• GraphQL: Find types, resolvers, mutations related to the feature
• Check authentication/authorization on relevant endpoints

3d -- Admin Panel (if configured)

• If the ticket involves admin functionality, investigate admin controllers or configuration
• Check how admin interfaces relate to the underlying models

3e -- Repository / Data Access Layer

• Find repositories, query builders, or data access objects
• Understand existing query patterns for the affected area
• If read-only queries are available ($DB_QUERY_COMMAND): verify current data state if relevant

3f -- Test Layer

• Find existing tests for the affected area
• Identify test patterns used (unit, integration, functional)
• Note any test fixtures, factories, or helpers

Investigation Summary

After all areas are explored, compile:

AFFECTED FILES: [list of files that will likely need changes]
DEPENDENCIES: [services, classes, interfaces involved]
EXISTING TESTS: [test files covering the affected area]
DATABASE IMPACT: [schema changes needed, if any]
API IMPACT: [endpoint changes, if any]
RISK AREAS: [parts of the codebase that could break]

Phase 4: Implementation Planning

Create a detailed plan before writing any code.

Step 4.1 -- Plan Structure

## Implementation Plan for $TICKET_ID

### Changes Required
1. [File path] -- [What changes and why]
2. [File path] -- [What changes and why]
...

### New Files
1. [File path] -- [Purpose]
...

### Test Plan
1. [Test file] -- [What to test]
2. [Test file] -- [What to test]
...

### Migration / Schema Changes
[If applicable]

### Order of Operations
1. [First step -- typically write failing tests]
2. [Second step]
...

Step 4.2 -- Design Consultation

If the implementation involves design decisions, consider consulting specialized skills:

Step 4.3 -- Plan Approval

Present the plan to the developer and wait for approval before proceeding. "Here is my implementation plan. Shall I proceed, or would you like changes?"

Phase 5: Branch Creation

After plan approval, create the feature branch.

Using $BRANCH_PATTERN from supplement:

git checkout -b $BRANCH_PATTERN

Example patterns:

• feat/PROJ-123-short-description
• fix/PROJ-123-short-description
• PROJ-123/short-description

Phase 6: TDD Implementation

Follow a strict TDD cycle. Use $TEST_COMMAND from supplement.

Step 6.1 -- Write Failing Tests First

For each item in the test plan:

1. Write the test
2. Run it to confirm it fails: $TEST_COMMAND path/to/TestFile
3. Confirm the failure is for the RIGHT reason (not a syntax error)

Step 6.2 -- Implement to Make Tests Pass

Write the minimum code to make each test pass:

1. Implement the change
2. Run the specific test: $TEST_COMMAND path/to/TestFile
3. Confirm it passes

Step 6.3 -- Refactor

After tests pass:

1. Look for duplication, unclear naming, or structural improvements
2. Refactor while keeping tests green
3. Re-run tests after each refactor step

Step 6.4 -- Run Full Test Suite

Run the broader test suite to catch regressions:

$TEST_COMMAND [relevant test directory or suite]

Fix any failures before proceeding.

Step 6.5 -- Commit Changes

Commit using the project's conventions ($COMMIT_FORMAT):

$SKIP_HOOKS git add [specific files]
git commit -m "$COMMIT_FORMAT"

For large changes, make multiple focused commits rather than one monolithic commit.

Phase 7: Implementation Summary

After all changes are committed, generate a summary:

## Implementation Summary for $TICKET_ID

### Changes Made
- [File]: [What changed and why]
- [File]: [What changed and why]

### Tests Added/Modified
- [Test file]: [What is tested]

### Commits
- [hash]: [message]

### Verification
- All new tests pass: YES/NO
- Existing tests pass: YES/NO
- Schema changes applied: YES/NO/N/A

### Notes
- [Any caveats, follow-up work, or decisions made during implementation]

Phase 8: PR Creation

Use the pr-message-writer skill if available, or create the PR directly.

Using pr-message-writer skill

Invoke the skill with the implementation summary and ticket context to generate a well-structured PR description.

Direct PR creation

Use $VCS_CLI (e.g., gh, glab) to create the PR:

git push -u origin HEAD
$VCS_CLI pr create --title "$TICKET_ID: [short description]" --body "[PR body]"

The PR body should include:

• Link to the ticket
• Summary of changes
• Test plan / how to verify
• Screenshots (if UI changes)
• Any deployment notes

Phase 9: Report Generation

If the report-writer skill is available, invoke it to create a work summary report covering:

• Ticket details and acceptance criteria
• Investigation findings
• Implementation decisions
• Test coverage
• PR link

This is optional but recommended for complex tickets.

Critical Rules

1. Never skip Phase 0. The supplement ensures the workflow adapts to your stack. Without it, the skill cannot function correctly.
2. Always get plan approval before coding. Phase 4 must end with explicit developer approval.
3. TDD is not optional. Write tests first. If the project has no test infrastructure, flag this to the developer and discuss before proceeding.
4. Commit granularity matters. One logical change per commit. Do not bundle unrelated changes.
5. Never force-push to the main branch. Always work on feature branches.
6. Ask when uncertain. If the ticket is ambiguous, ask the developer rather than guessing. Wrong assumptions waste more time than a quick question.
7. Respect existing patterns. Match the project's coding style, naming conventions, and architectural patterns -- even if you would do it differently.
8. Keep the developer informed. At the end of each phase, provide a brief status update. Do not silently proceed through multiple phases.
9. Handle failures gracefully. If tests fail unexpectedly, if the codebase does not match expectations, or if the ticket requirements are unclear -- stop and communicate rather than pushing through.
10. Security first. Never commit secrets, credentials, or sensitive data. If a ticket requires configuration changes, use environment variables or configuration files that are in .gitignore.

Reference Files

Integration with Other Skills