Mk ask me
Skill ngocsangyem/MeowKit/packages/mewkit/src/migrate/modules/codex/root/.agents/skills/mk-ask-me
Answer factual project questions (codebase, architecture) with cited evidence. NOT for brainstorming, trade-offs (mk:party), debugging (mk:investigate), review (mk:review), or plans (mk:plan-creator).From its SKILL.md
npx -y skills add ngocsangyem/MeowKit --skill mk-ask-meAssembled from the repository path, not quoted from the project. Check it against their README if it does not work.
One thing to look at
- 15 stars15 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.
SKILL.md
4.6 KB, ~1.0k tokens by cl100k_base, as published. Nobody here has run it
mk:ask-me
Overview
Answer questions about the current project — codebase, architecture, conventions,
workflows, rationale, constraints — with minimal cited file:line evidence.
Gives the agent deep, fast project understanding instead of long grep/search
sessions. Read-only by default; the answer is the deliverable.
Route first
Classify the request before gathering evidence. If a specialist owns it, name that skill and stop; do not partially answer. The complete routing table is references/specialist-routes.md.
This skill answers what is true, how it works here, why it exists, and what constraints apply — about the current project only.
When to Use
Auto-activate on:
- "how does X work here / in this project / in this repo"
- "what does X mean" (project file, config, convention)
- "why is X structured this way"
- "is it true that ... in this repo" (claim-check)
- "what constraints apply before touching X"
- "explain X with sources / citations"
Explicit: the ask-me skill [question] [--depth quick|standard|deep] [--save-to <path>]
Do NOT invoke when: the request matches any Redirect First row, asks to change code, or is a pure general-knowledge question (answer those directly, no scouting).
Process
- Route check. Match against Redirect First. Owned elsewhere → name the skill, stop.
- Parse. Extract the question. Infer kind: factual, explanation, rationale,
constraint-summary, or claim-check. Depth:
standardunless flagged. - Minimal evidence plan. Prefer user-provided paths; else grep targeted keywords. Docs/project context first, then targeted code/config. Apply the forbidden-path list below.
- Bounded gather. Read only relevant slices within the depth budget. Stop
as soon as the answer is sufficiently supported. Track
path:linefor every source-backed claim. - Synthesize. Direct answer first; separate verified claims from inference; state confidence (high/medium/low — mandatory); list limitations; mention a related specialist skill only if the user's next likely action leaves scope.
- Optional save. Only via
--save-toor explicit request, per the save-path policy below. On unsafe path or conflict, refuse the write and return the answer inline.
Depth budgets and output/failure contracts: references/answer-contract.md.
Forbidden Paths (never read)
.env*, secrets/keys/credentials/tokens of any kind, .meowkit/memory/,
session-state/, dependency dirs (node_modules/, .venv/), build outputs,
browser profiles. No WebSearch — current-docs lookup belongs to mk:docs-finder.
Save-Path Policy
Write only when asked, and only under: an explicit user path inside the project
root, tasks/reports/, or ${PLUGIN_DATA}/ask-me/. Path escapes a safe
root → refuse and answer inline. Target exists → ask before overwrite or use a
timestamped name.
Answer contract
Return a direct answer, confidence, cited evidence, and only needed inference/limitations. Include one related specialist only when the likely next action leaves this scope. Load references/answer-contract.md for the exact template and failure handling.
Workflow Integration
On-demand — no single phase anchor; usable at any point from Phase 0 (orient) through Phase 6 (reflect). Does not gate or feed any phase artifact.
Handoff Protocol
The answer is terminal by default — no agent receives it. If the user's next action leaves this skill's scope, point to the single owning specialist skill (one pointer, never auto-invoke it).
Gotchas
- Don't turn an answer request into a plan — the answer IS the deliverable.
- Don't over-scout: stop at sufficiency, not at budget exhaustion.
- Don't answer option-selection questions; route them (
mk:party/mk:brainstorming). - Never cite project facts without
path:linesources; unsourced points go under Inference. - Missing docs lower confidence but don't block — fall back to targeted code/config reads.
- "How does [library] work" is
mk:docs-finderterritory; this skill only answers "how does X work in this project".
What ships with it: 2 files
1.9 KB alongside SKILL.md
references/
- answer-contract.md770 B
- specialist-routes.md1.2 KB
Gives 0 of the 12 instructions most architecture codebase skills give in ~1.0k tokens
Counted across 811 of the 1,134 authors here whose files we hold, read 2026-08-07
- Ask the user which candidate to explorein 45 of 811, across 15 files
- Apply the deletion test to suspected shallow modulesin 43 of 811, across 15 files
- Read any relevant architecture decision records firstin 31 of 811, across 8 files
- Use exact glossary terms in every suggestionin 30 of 811, across 10 files
- Accept dependencies instead of creating themin 24 of 811, across 5 files
- Include before and after visualisations for each candidatein 24 of 811, across 5 files
- Read the domain glossary before exploringin 24 of 811, across 6 files
- Return results instead of producing side effectsin 23 of 811, across 4 files
- Explore the codebase for shallow modules and frictionin 23 of 811, across 3 files
- Introduce seams only where things varyin 22 of 811, across 3 files
- Reduce the number of methodsin 21 of 811, across 2 files
- Design deep modules with small interfacesin 21 of 811, across 3 files
Said here and by no other author read
- classify the request before gathering evidence
- extract the question and infer depth
- read only relevant slices within depth budget
- track path and line for every source-backed claim
- separate verified claims from inference
- write only when explicitly asked
Grouped from the skills themselves: near-identical wordings counted once, and counted by distinct author, so one author publishing three of these counts once. Length counted with cl100k_base; the agent that loads this file may tokenize it differently.