ask-ai-catapult
Ask which skill or flow fits your situation. Use when you know what you want to achieve but not which skills to combine, or which order to run them in.
- License 许可
- Apache-2.0
- 所属仓库
- r3dlex/skills
- 最近更新
- 2026-10-29
- Stars
- ★ 0
- Forks
- 0
01 前言 Front Matter
- name
- ask-ai-catapult
- description
- Ask which skill or flow fits your situation. Use when you know what you want to achieve but not which skills to combine, or which order to run them in.
02 资源文件 2
- 参考文档CATALOG.md7.0 KB
- 参考文档PHASE-BOUNDARIES.md2.4 KB
03 SKILL.md 正文
Ask ai-catapult
Recommend a combination of skills for the situation described — not a single answer, and not a fixed script. Enumerate what is actually installed, map it to the use case, then propose a sequence with reasons.
How to route
- Detect the harness. Claude Code and Codex expose different ecosystems. Establish which one you are running under before recommending anything from tier 2.
- Enumerate what is actually available. Query the catalog for tier 1 (CATALOG.md is the generated inventory); for tiers 2 and 3, look at what is genuinely installed in this environment. Never name a skill you have not confirmed exists — a confident pointer to a missing skill is worse than no pointer.
- Map the use case to a phase, then to skills within it.
- Propose a sequence, with one line of reasoning per step and the decision points called out. Say what you would skip, and why.
The three tiers
Tier 1 — the ai-catapult catalog. The SDLC spine: discover → govern/plan → configure/generate → validate/handoff. Always present. Prefer these for the phase work itself.
Tier 2 — the host ecosystem, resolved at runtime. Under Claude Code that is oh-my-claudecode; under Codex, oh-my-codex. These are orchestration and execution accelerators layered on top of the spine, never replacements for it. Both are optional — ai-catapult ships to users who have neither, so their absence must degrade gracefully rather than produce dead references.
Route by capability, not by name, since these ecosystems version independently of this catalog and any roster written down here would rot:
- Need parallel execution across independent tasks → look for an orchestration skill.
- Need persistence across a long multi-session effort → look for a durable-loop skill.
- Need an independent review or verification lane → look for a review or verify skill.
- Need deep multi-file search → look for a codebase-search skill.
Tier 3 — anything else installed. Use opportunistically when it fits better than tier 1 or 2.
Two flows, deliberately
This catalog has two intake→ship paths. Say which one applies rather than presenting both as equally canonical:
northstar→autobahnis the governed spine. Use it when the work needs a tracked A→B handoff, consensus planning, sliced goals, and a merge gate — multi-PR efforts, anything with review and CI requirements. It is the default for substantial work.to-spec→to-tickets→implementis the lighter path. Use it when the decisions are already made and the work just needs specifying and building, without the handoff and merge-authority machinery.
Both end in code-review. Choosing between them is the single most useful thing this skill
does — ask how much governance the effort actually needs.
On-ramps
Route by where the user actually is, not where a tidy process would start them:
- Something is broken, slow, or throwing →
diagnosing-bugs. - A pile of untriaged issues →
triage. - Work too big to hold in one session, and the route is not yet visible →
wayfinder. - Repo not yet initialized for this workflow →
ai-catapult-init.
Vocabulary layer
domain-modeling and codebase-design are references consulted during other work, not
sessions to run on their own. Recommend them alongside a flow — "use codebase-design's
vocabulary while shaping that interface" — rather than as a step.
Answering
Give a sequence, not a menu. Name the first concrete action, and be explicit about what you are leaving out. If the situation genuinely does not need a flow — a one-line fix, a question — say so instead of manufacturing process.
If nothing fits, say that plainly rather than stretching a skill to cover it.
References
- CATALOG.md — generated tier-1 inventory, grouped by phase.
- PHASE-BOUNDARIES.md — what belongs in each phase, and the handoffs.
04 同仓库更多 Skills
skill
Fixture.
stable-skill
Fixture stable-skill.
experimental-skill
Fixture experimental-skill.
deprecated-skill
Fixture deprecated-skill.
karpathy-guidelines
Apply think-before-coding, simplicity, surgical-change, and goal-driven engineering rules. Use when planning, implementing, or reviewing code.
karpathy-guidelines
Apply think-before-coding, simplicity, surgical-change, and goal-driven engineering rules. Use when planning, implementing, or reviewing code.