Writing & Documentation Skills

Use when /api-test is invoked after /executing-plans has implemented APIs. Drives the 6-step pipeline — API inventory from -implementation-plan.md and code → SQL guides for test data acquisition (user pastes results) → pytest scenario code generation (with conftest.py auth fixture auto-selected via detectauth.py) → execution → result recording in -implementation-plan.md change-history → fail-handling proposal.

auto-flow 진입점 — Socratic clarifying Q (1~5개 적응) + AI 자동 approach 선택 + 자동 section 작성 + change-history 자동 + auto-tech-design 자동 invoke. 사용자 입력은 clarifying Q 답변에만. Visual Companion 호출 X. 추가 AskUserQuestion 호출 X (사용자 입력은 clarifying Q 답변에만, 그 질문 자체는 AskUserQuestion 도구로). generating-html 은 Step 4.5 fire-and-forget dispatch (v2.3.2+ — v1.1.17 '호출 부재' 룰이 반전됨).

auto-flow 4단계 (마지막) — implementation-plan.md 읽기 + Entry Guard (preflight.subagenttaskentrycheck) + DAG 자동 build + 무조건 wave-parallel subagent 강제 (Gate 14 override) + failure isolation 그대로 + End-of-run consolidator 자동 + finishing-a-development-branch 자동 호출. 추가 AskUserQuestion 호출 X (사용자 입력은 clarifying Q 답변에만, 그 질문 자체는 AskUserQuestion 도구로). generating-html 은 Step 4.5 fire-and-forget dispatch (v2.3.2+ — v1.1.17 '호출 부재' 룰이 반전됨).

auto-flow 2단계 — requirements.md 읽기 + adaptive 7-topic 자동 판정 + design decision 자동 alternatives 비교 → recommendation 자동 선택 + verifying-spec 4축 보고서 transition 직전 노출 + auto-writing-plans 자동 invoke. 추가 AskUserQuestion 호출 X (사용자 입력은 clarifying Q 답변에만, 그 질문 자체는 AskUserQuestion 도구로). generating-html 은 Step 4.5 fire-and-forget dispatch (v2.3.2+ — v1.1.17 '호출 부재' 룰이 반전됨).

auto-flow 3단계 — requirements + tech-design 읽기 + AI 자동 task 분해 (TDD bite-sized + Model hint 자동) + RISK 코드 지점 §2 자동 + verifying-spec 자동 + code-pretty 호출 X (D-T12 와 일관) + change-history 자동 + auto-executing-plans 자동 invoke. 추가 AskUserQuestion 호출 X (사용자 입력은 clarifying Q 답변에만, 그 질문 자체는 AskUserQuestion 도구로). generating-html 은 Step 4.5 fire-and-forget dispatch (v2.3.2+ — v1.1.17 '호출 부재' 룰이 반전됨).

Use when ANY MD or code is added/modified/deleted in a feature folder under docs/features/ - /, before commits — appends a structured change-log entry to the relevant MD's "변경이력" footer with auto-generated CH-id, reason, scope, before/after code (for code edits), and impact links.

Use whenever the user expresses a change to existing 요구사항/개발방향/구현계획서/code in a feature folder (natural language like "X 바꿔/추가해/빼" or explicit edit-skill invocation). Identifies the change level, applies the impact matrix to surface affected downstream artifacts, gates user approval, then performs cascading updates and writes change-history entries with cross-linked CH-ids.

Use AFTER verifying-spec passes and BEFORE generating-html during the initial-creation iteration loop of -implementation-plan.md ONLY. Dispatches a Sonnet subagent that performs a strict format + 자명한 정리 + 중복 통합 pass on every "수정 후" code block in the plan. NEVER touches "원본" blocks, prose, or tables. Stops firing once the first change-history entry is logged. Idempotent on already-clean blocks (no-op rule).

Use during the initial-creation iteration loop of -requirements.md / -tech-design.md / -implementation-plan.md. Fires before user review on every draft (v1.1.15+ unified timing — pre-review per-draft). Re-fires on each user-fix iteration. STOPS firing once the first change-history entry is logged — that boundary marks "live doc". Dispatches a single Sonnet subagent (B) in fire-and-forget mode (v2.2.1+) — generates a sibling .html companion (사람 전용 시각화 사본) with semantic 1:1 preservation. Main does NOT wait for the result. RAW .md is shown to user as-is (v2.2.1 removed the v2.2.0 A format-only pass — B alone now handles human readability). NEVER invoked on change-history appends, change-propagation cascades (use /sync-html instead), or partial revisions.

Upstream-original (superpowers 5.0.7) brainstorming, exposed under intent-locked-workflow. Use when invoked via /og-brainstorm — free-form Socratic dialogue producing docs/superpowers/specs/YYYY-MM-DD- -design.md, terminal state is og-writing-plans. Identical behavior to upstream brainstorming except for the og- naming and downstream pointer. Use this when the intent-locked-workflow extended workflow (PRD + 변경이력 + 위험 주석) feels heavy and you want the original superpowers experience.

Upstream-original (superpowers 5.0.7) executing-plans, exposed under intent-locked-workflow. Use when invoked via /og-execute-plan after /og-write-plan. Identical behavior to upstream executing-plans — load plan, review critically, execute task-by-task, follow steps exactly. No 변경이력 / 위험주석 / git-fast mode / batched logging — those are intent-locked-workflow extensions, intentionally absent here.

Upstream-original (superpowers 5.0.7) writing-plans, exposed under intent-locked-workflow. Use when invoked via /og-write-plan after /og-brainstorm. Identical behavior to upstream writing-plans except for the og- naming and downstream pointer (handoff goes to og-executing-plans / subagent-driven-development). Use this when the intent-locked-workflow extended workflow (변경이력 / 위험주석 / verifying-spec / docs-pretty) feels heavy and you want the original superpowers experience.

Use during /executing-plans whenever code is created or modified, and during verifying-spec when surveying existing code. Applies a 3-item self-checklist to detect side-effect/breaking/race risks, then attaches standardized " ⚠️ RISK(category): reason — by context" comments to risky lines and records the placement in change-history.

Use immediately after writing -tech-design.md (via tech-design) or -implementation-plan.md (via writing-plans), before final user handoff. Performs main-agent self-verification — A) consistency cross-check between target MD and all upstream MDs, plus C) code impact analysis (file existence, callers, side-effect candidates) — and produces a structured 4-axis report for user decision.