メニュー
This skill should be used when explaining why Bee follows spec-first TDD workflows, or when teaching mode is on. Contains the reasoning behind spec → plan → test → code ordering.
Codex または Claude でインストール この Prompt をコピーして Codex、Claude、または他のアシスタントに貼り付けると、Skill ページを確認してインストールできます。
SOC 職業分類に基づく
This skill should be used when an orchestrator needs to choose or resolve a model for a delegated agent — i.e. by /bee:planner when authoring a plan (assigning a capability tier per agent per slice) and by /bee:plan-implementer when executing one (resolving a tier to a concrete subagent model value). Contains the single source of truth mapping capability tiers (best / excellent / fast) to Claude Code subagent model values, with per-tier guidance for picking the right one. Load this skill instead of hardcoding any model name.
Engine for the /ccaf:practice command. Asks which specific CCAF domains to focus on and how many questions (10/20/30), assembles a proportionally-weighted partial exam anchored to the reference bank and independently verified, administers it 4 questions per screen with resumable progress, and scores it with a per-domain performance chart. Attempt state lives in ~/.claude/ccaf-practice.local.md — separate from /ccaf:mock-exam. Use when running or resuming /ccaf:practice.
Engine for the /ccaf:mock-exam mock exam. Assembles a 60-question CCAF mock (all questions generated fresh per attempt, anchored to a self-authored reference bank and independently verified), administers it 4 questions per screen with resumable progress, and scores it on the real 100–1000 band with a 720 pass line. Use when running or resuming /ccaf:mock-exam.
Engine for the /ccaf:prepare conversational tutor. Teaches the Claude Certified Architect – Foundations (CCAF) syllabus turn by turn — one concept at a time, with a knowledge check every turn — adapting pace and difficulty to the learner. Opens with a multi-select topic menu over the blueprint's 5 domains and 30 task statements; the learner picks what to learn. Spawns the ccaf-check-author agent for scenario checks; hands off to /ccaf:mock-exam when a domain looks solid. Use when running /ccaf:prepare.
Generates a full incident postmortem markdown document from a cached timeline (built by incident-timeline-builder) and the source materials gathered during investigation. Uses a Google-SRE-style blameless template by default. Produces the postmortem as plain markdown text for the caller to render or save; the skill itself never writes to disk or posts anywhere. Use when the incident-postmortem agent reaches Phase 4, or when a user has a structured timeline and wants the full document.
Restructures a poorly written tracker issue into a clear, self-contained document a stranger can open cold and act on. Works on any archetype (Bug, Incident, Story, Feature, Task, Spike) and any supported tracker (Azure DevOps, Jira). Returns the refined title and markdown body; the caller owns the write (issuekit:tracker-adapter projects markdown to the vendor's format).
| name | ai-workflow |
| description | This skill should be used when explaining why Bee follows spec-first TDD workflows, or when teaching mode is on. Contains the reasoning behind spec → plan → test → code ordering. |
The #1 reason AI-generated code misses the mark is ambiguous requirements. "Add authentication" forces the AI to guess 100 decisions — session vs JWT, cookie vs header, redirect vs 401, password rules, lockout policy. Each guess compounds.
A spec with clear acceptance criteria eliminates guessing. The AI gets unambiguous targets and nails them on the first pass.
The tradeoff: 10 minutes writing a spec saves hours of rework. Without a spec, you'll spend that time anyway — just spread across debugging, re-prompting, and rewriting.
AI doesn't know when it's finished. It will keep generating code until you stop it. Tests give it a concrete finish line.
A failing test is the clearest possible prompt: "Make this pass." The AI doesn't need to interpret requirements — the test IS the requirement in executable form.
Red-green-refactor with AI:
AI works best with focused, bounded tasks. A vertical slice (UI + backend + data for one capability) gives the AI everything it needs to deliver something complete and testable.
Horizontal slicing (all database tables, then all APIs, then all UI) forces the AI to build things it can't test in isolation. Each layer depends on layers that don't exist yet.
Vertical slice benefits:
Low-risk code (internal tool, easy to revert) doesn't need the same rigor as high-risk code (payment flow, auth, data migration). Applying the same process to both wastes time on low-risk tasks and under-serves high-risk ones.
Risk-aware workflow means:
Not every task needs a spec. Not every change needs TDD. The workflow should match the task:
The right amount of process is the minimum needed to produce correct, maintainable code. More process than needed is waste. Less process than needed is risk.
