Deployer Workflow Skill
Global Context
- User request: $ARGUMENTS
- Agents directory:
agents/
- Agent definition files: all markdown files in
agents/
- Design doc:
design-doc.md (project root)
- Progress file:
todo.md (project root)
- Mistakes file:
mistakes.md (project root)
- Review folder:
review/{yyyy-MM-dd}/
- Changelog:
changelog.md (project root)
- README:
README.md (project root)
Agents
There are 5 agents total:
- Backend Developer (java-backend-developer-agent.md / go-backend-developer-agent.md / rust-backend-developer-agent.md)
- React Developer (react-developer-agent.md)
- Relational DBA (relational-dba-agent.md)
- Testing Agent (testing-agent.md) - handles unit tests, integration tests, UI tests (Playwright), stress tests (K6)
- Reviewer Agent (reviewer-agent.md) - handles code review, security review, design doc sync, feature docs, changes summary
Rules
- If $ARGUMENTS is empty, ask: "What do you want to build? Describe the feature or application."
- Ask backend language using AskUserQuestion with options:
- Java: Spring Boot 4.x, Java 25
- Go: Go 1.25+, Gin Gonic
- Rust: Rust 1.93+, Axum/Actix-web
- Read all agent definitions from
agents/ and pass them as context to subagents.
- Use Task subagents with
subagent_type: "general-purpose".
- Include user request, relevant files, and the full agent definition in each subagent prompt.
- Phase order: Build -> Test -> Review -> Changelog -> README.
- Phase dependencies: Test depends on Build. Review depends on Build + Test. Changelog depends on all previous phases.
- After each phase, update
todo.md by marking completed items with [x].
Mistakes Tracking
mistakes.md in the project root tracks all build failures, test failures, and issues across all phases.
- Every agent MUST read
mistakes.md before starting work to avoid repeating past mistakes.
- Every agent MUST append new mistakes/issues they encounter to
mistakes.md with the phase name and a short description.
- At the start of the workflow, create
mistakes.md if it does not exist with a header # Mistakes Log.
- Format:
- [Phase N: Name] description of the mistake or issue and how it was fixed
Build and Test Enforcement
- The build MUST compile and pass before moving to Phase 2 (Test).
- ALL tests (unit, integration, UI, stress) MUST pass before moving to Phase 3 (Review).
- If the build fails, fix it immediately. Do not proceed until the build is green.
- If any test fails, debug and fix the root cause. Re-run all tests until they all pass.
- After fixing build or test failures, record what went wrong and how it was fixed in
mistakes.md.
- At the end of Phase 1, run the full build and verify it succeeds.
- At the end of Phase 2, run all tests and verify they all pass.
Step 1: Review Plan
Use AskUserQuestion with checkboxes (all checked by default) and allow unchecking. Store selections and skip unchecked items:
- Phase 1: Build
- Build Components (Backend, Frontend, Database)
- Verify Build (compile, run, connectivity)
- Phase 2: Test
- All Tests (Unit, Integration, UI Playwright, K6 Stress)
- Phase 3: Review
- Full Review (Code, Security, Design Doc Sync, Feature Docs, Changes Summary)
- Changelog & README Update
Initialize todo.md with current date (yyyy-MM-dd) and all selected items as [ ].
Step 2: Design Doc
Create design-doc.md with:
- Architecture overview
- Backend API endpoints and responsibilities
- Frontend components and interactions
- Database schema design
- Integration points between frontend, backend, database
Phase 1: Build
Build Components (parallel)
Spawn 3 subagents in parallel:
- Backend Developer: use chosen language agent (java/go/rust backend agent). Implement backend per
design-doc.md. Agent MUST read mistakes.md first.
- React Developer: use react-developer-agent.md. Implement frontend per
design-doc.md. Agent MUST read mistakes.md first.
- Relational DBA: use relational-dba-agent.md. Design and create DB schema per
design-doc.md. Agent MUST read mistakes.md first.
Verify Build
After all 3 subagents complete:
- Verify DB schema/migrations apply successfully.
- Verify backend compiles, builds, and runs with DB connection.
- Verify frontend builds and can be served and connects to backend.
- If ANY build step fails: fix the issue, record it in
mistakes.md, and re-verify.
- Do NOT proceed to Phase 2 until the full build is green.
Phase 2: Test
Spawn 1 subagent:
- Testing Agent: use testing-agent.md. Run all tests (unit, integration, UI Playwright, K6 stress) for all code from Phase 1. Agent MUST read
mistakes.md first.
After the testing agent completes:
- Verify ALL tests pass by running the full test suite.
- If ANY test fails: fix the issue, record it in
mistakes.md, re-run all tests.
- Do NOT proceed to Phase 3 until all tests are green.
Phase 3: Review
Create review/{current-date}/.
Spawn 1 subagent:
- Reviewer Agent: use reviewer-agent.md. Perform code review, security review, sync design doc, write feature docs, and summarize changes. Agent MUST read
mistakes.md first. Outputs:
review/{current-date}/code-review.md
review/{current-date}/sec-review.md
review/{current-date}/features.md
review/{current-date}/summary.md
- Updates
design-doc.md
If critical issues are found, fix them and record in mistakes.md.
Changelog
Create changelog.md using git info:
- Use
git status, git diff, git log.
- Include current date, what was built, files created/modified, test coverage summary, review findings and fixes, docs generated, remaining issues or recommendations.
README
Update/create README.md with:
- Overview
- Links to:
design-doc.md, review/{current-date}/code-review.md, review/{current-date}/sec-review.md, review/{current-date}/features.md, review/{current-date}/summary.md, changelog.md
- Summary highlights from
review/{current-date}/summary.md
- Quick start (backend, frontend, database)