Coding Flow

Problem: Unstructured coding leads to scope drift, missing validation, autonomous runaway, and misaligned deliverables. Solution: Sequential workflow with reviewer gates, HITL gates, subagent delegation, and skill-driven execution scaled per Request size classification. Validation: Each phase produces verifiable outputs; reviewer catches issues before user; HITL gates prevent autonomous runaway; final validation confirms implementation matches approved intent.

1. All Rosetta prep steps MUST be FULLY completed
2. MUST USE SKILL load-project-context (required: all), orchestration (all except trivial), hitl (all, unless No HITL or Fully Autonomous)
3. MUST ALWAYS use todo tasks ledger, ASAP. Phases are sequential. Independent tasks can run in parallel.
4. No rush, take your time, MUST FOLLOW WORKFLOW ENTIRELY, no skipping, if in doubt - select the safest / longest path, no deviation from the workflow is allowed
5. When debugging is needed, INVOKE SUBAGENT engineer with debugging skill to save LLM context
6. INVOKE SUBAGENT executor for building, running tests, installing packages, and similar mechanical actions.
7. MUST just-in-time load each phase's skills
8. If workflow is for REQUIREMENTS, MUST USE SKILL requirements-use and LOAD all affected requirements. Use refs to requirements for subagents.
9. If /goal is set repeat phases 7-12 postponing user_review_impl and final_validation until goal is met.
10. If migrate/modernize: implementation phase MUST use tiny batches ONLY (1-3 files), never bulk-read (other phases may); specs/plan enforce; FS-copy RECOMMENDED; no behavior change/new code; mirror source; subagents same; REQUIRED TO log started/completed; Use impl subagents like MAP-REDUCE;
11. Run architect subagent with required model in the background and consult with it if already supported or prefer advisor if already available
12. Coding workflow state MUST be saved to agents/TEMP/<FEATURE>/coding-flow-state.md file.
13. If context already contains request for TDD => implement properly red and green adjusting this workflow, including by moving execution of phases tests and review_tests to be after user_review_plan and before implementation.

1. Gather project context, affected areas, dependencies, constraints, requirements. SMALL: orchestrator handles inline.
2. Input: user request + CONTEXT.md + ARCHITECTURE.md + IMPLEMENTATION.md. Output: discovery-notes.md in FEATURE PLAN folder.
3. Required skills: load-project-context
4. Recommended skills: codemap (structural project discovery)
5. If REQUIREMENTS in use: requirements-use skill is required.
6. Additionally request to discover existing libraries, packages, search web for similar problems/tasks (if this make sense)
7. Update coding-flow-state.md
8. Do not stop until 100% clear

1. Step 1: discover affected and related code to design architecture requirements to address user request fully.
2. Step 2: design 3 best architecture solutions on high level with pro/cons analysis.
3. Step 3: define the best solution, but concise, phrase-terse, compressed, etc.
4. Input: user request + CONTEXT.md + ARCHITECTURE.md + IMPLEMENTATION.md. Output: concise architecture-notes.md in FEATURE PLAN folder.
5. Required skills: reasoning
6. Recommended skills: questioning
7. Update coding-flow-state.md

1. Present main solution first and then alternatives, do not assume user is in context, give him full information with TLDR.
2. Present specs, plan, and review findings. User MUST approve: "Yes, I reviewed the design" or "Approve, the design was reviewed".
3. Strict approval; anything else = review feedback, iterate.
4. SMALL: combine with Phase 6 into single checkpoint.

1. MUST USE SKILL tech-specs and planning together. Split: specs own WHAT, plan owns HOW. Target: 100% clarity.
2. Input: discovery notes, user request, ARCHITECTURE.md. Output: plans/<FEATURE>/<FEATURE>-SPECS.md + plans/<FEATURE>/<FEATURE>-PLAN.md.
3. SMALL: output as message, no files. MEDIUM: concise. LARGE: full.
4. Required skills: tech-specs, planning
5. If medium/large reasoning skill is required
6. If REQUIREMENTS in use: requirements-use skill is required. Plan/Specs must have pointers to requirements identifiers.
7. Recommended skills: questioning
8. Update coding-flow-state.md

1. Review specs and plan against user request and discovery notes, do not assume user is in context, give him full information with TLDR.
2. Input: specs, plan, user request. Output: review findings and recommendations.
3. Update coding-flow-state.md

1. Present specs, plan, and review findings. User MUST approve: "Yes, I reviewed the plan" or "Approve, the plan and specs were reviewed".
2. Strict approval; anything else = review feedback, iterate.

1. Implement approved plan. Build MUST succeed. Tests excluded.
2. Input: approved specs + plan. Demand subagent to read and execute it fully. Do not repeat contents => reference instead. Output: working code, build passing, update relevant documentation briefly (CONTEXT.md, ARCHITECTURE.md, etc).
3. MUST follow approved scope. MUST stop and escalate if blocked.
4. Required skills: coding
5. Recommended skills: debugging, sensitive-data, testing, dangerous-actions
6. If requirements are used code must contain comments refs to requirements identifiers
7. Spawn multiple implementation agents on independent tasks without dependencies and files intersection if reasonable
8. Update coding-flow-state.md

1. Review code changes against approved specs and plan.
2. Input: implementation diff, specs, plan, check if documentation is updated, brief, and matches the file intent. Output: review findings and recommendations.
3. Required skills: coding
4. Recommended skills: reasoning, debugging, sensitive-data, testing, dangerous-actions
5. Update coding-flow-state.md
6. If SMALL must also validate by running locally and check implementation actually works, once code review is done and there are no major issues

1. Validate implementation against specs: git changes, spec coverage, gaps, perform search and MCP fact-checking.
2. Then it must run locally and check it actually works if there are no major issues
3. Input: implementation diff, specs, plan, review findings. Demand subagent to read and verify specs/plan fully. Do not repeat contents => reference instead. Output: validation findings.
4. SMALL: orchestrator performs quick inline check.
5. Recommended skills: reverse-engineering, debugging, sensitive-data, testing, dangerous-actions
6. Update coding-flow-state.md

1. Present implementation, review findings, and validation findings. User MUST approve: "Yes, I approve the implementation".
2. Strict approval required before proceeding to tests.
3. SMALL: combined with Phase 13 final validation.

1. Write and execute tests. All MUST succeed, isolated, idempotent.
2. Input: implementation, specs. Demand subagent to read specs fully. Do not repeat contents => reference instead. Output: passing tests with coverage.
3. Required skills: testing, coding
4. Recommended skills: debugging, sensitive-data, dangerous-actions
5. Update coding-flow-state.md

1. Review tests against specs: coverage, scenarios, edge cases, mocking correctness.
2. Input: tests, specs, implementation. Output: review findings and recommendations.
3. Required skills: testing, coding
4. Recommended skills: debugging, sensitive-data, dangerous-actions
5. Update coding-flow-state.md

1. Systematic by-dependency validation: databases, APIs, web, mobile. Check logs, clean up.
2. Additionally systematic "manual QA" by yourself.
3. Input: full delivery (code + tests + specs + review findings). Demand subagent to read specs fully. Do not repeat contents => reference instead. Output: final validation report.
4. SMALL: orchestrator confirms build + tests pass.
5. Recommended skills: coding, debugging, sensitive-data, testing, dangerous-actions
6. Update coding-flow-state.md

MCPs:

• DeepWiki, Context7 — external documentation and library knowledge
• Playwright, Chrome-DevTools — web app testing
• Appium — mobile app testing
• GitNexus — codebase knowledge graph
• Serena — semantic code retrieval at symbol level