# Feature Delivery

> Use this skill ALWAYS whenever adding new functionality, new endpoint, new component, new agent, or extending existing behavior in a non-trivial way. Trigger this skill at the start of ANY feature work even if the user just describes "add X" or "implement Y" without explicitly mentioning a process.

- Skill: `engsimsoft/feature-delivery` (Agent Skill)
- Install (CLI): `npx skillmds@latest add engsimsoft/feature-delivery`
- Raw SKILL.md: https://api.skillmd.com/api/skills/engsimsoft/feature-delivery/raw
- Safety review: pending
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: Web & Frontend
- Author: engsimsoft (https://skillmd.com/u/engsimsoft)
- Updated: 2026-09-17
- Page: https://skillmd.com/skills/engsimsoft/feature-delivery

---


# Feature delivery workflow

For ANY new functionality — endpoint, component, AI agent, feature flag, pipeline, etc.

## 1. SPEC.md first (NO code yet)
- Run `/start-tz <name>` slash command — creates `specs/<SERIES>/<TZ_N>/SPEC.md` from template
- Fill: goal (≤3 sentences), explicit IN scope, explicit OUT scope, definition of done, deadline
- Owner reviews and approves SPEC before code

## 2. Architecture check
- If feature touches architecture (new module, new external dep, new pattern) → ADR REQUIRED before code
- Use `/adr` slash command to create one-line ADR fast
- Read related ADRs first (UserPromptSubmit hook auto-injects matching ADRs)

## 3. ROADMAP.md — staged plan
- Run `/start-tz` already created it. Fill stages with task checklists.
- Each stage independently validatable: code → mypy/ruff per task (R12) → pytest per stage (R13)
- Each stage ends with manual smoke test from owner before next stage starts

## 4. External tech check (R2)
- Pydantic AI? FastAPI? React component lib? — Context7 MCP `resolve-library-id` → `get-library-docs`
- Quote the relevant API clause + URL in SPEC.md
- Do NOT write code from training memory

## 5. Implement stage by stage
- Stage 1: write code → mypy clean per task → write tests → pytest passes → request manual smoke
- WAIT for owner OK
- Stage 2: only after stage 1 OK
- ...

## 6. NO drive-by improvements (R3)
- See a bug while building? → log to `docs/agent/06_known_issues.md` (FINDINGS.md inside TZ dir)
- See a refactor opportunity? → log to backlog
- Stop hook will block scope creep at end of turn — heed it

## 7. One commit per ТЗ (R5)
- Stage validation does NOT mean commit. Code accumulates.
- Single commit at /close-tz: `feat(tz-XX): <one-sentence summary>`
- Commit body 3-5 lines, no exhaustive file list

## 8. Close ТЗ
- Run `/close-tz` slash command:
  - SQL/data sanity check if DB touched
  - Manual smoke OK from owner
  - FINDINGS.md → backlog (significant findings become follow-up TZ)
  - docs/agent/04_changelog.md updated
  - docs/agent/00_current_state.md updated (only durable facts)
  - ADR for architectural decisions
  - Single commit
  - Folder moved to `_archive/` when series closes

## Anti-patterns
- ✗ Code before SPEC
- ✗ Code before reading docs (R2)
- ✗ Skip ADR for architectural decision
- ✗ Combine 3 ТЗ into one commit
- ✗ Mark ТЗ done without owner smoke test