メニュー
Use mmcg to ground structural code claims before planning, auditing, criticizing, or researching code. Triggers when an agent needs symbol existence, callers, callees, imports, blast radius, file existence, or stale-index handling.
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
Turn a design handoff into a task contract that can actually be checked — named source, resolved components, token names instead of raw values, and an explicit list of what stays unverifiable. Use when a Figma frame, mockup, or design spec is handed over for implementation.
Convert a PRD, ticket, or feature request into a task contract that can fail — behaviour split from outcome, product nouns resolved to symbols that exist, unspecified cases surfaced as questions, and success metrics parked where a merge gate cannot wave them through. Use when product writing is handed over for implementation.
Enumerate what a change can reach and what enforces it — privileged operations and their callers, guards and the sites that apply them, and every module that reads a secret — before designing an auth, permission, or secrets change. Feeds the security review; produces a population and a gap list, never a verdict.
Read-only review of whether the tests in a finished change actually prove its behaviour — changed code no test reaches, a test exercising a different path than the one that changed, an assertion edited to match the new output, and a suite that ran nothing. Use after implementation, and whenever a change claims to be covered.
Gather who already depends on a service, handler, or state owner before changing it, and name the runtime gaps the codegraph cannot span. Use before designing a change to an API, queue consumer, background job, migration, or shared state; it feeds the architecture review rather than replacing it.
Read-only review of a finished UI change against the codegraph — a component nothing renders, a props contract changed without its callers, a reinvented component, or a raw value where the design system has a token. Use after implementing React or Vue work; triggers "review this UI change", "check the component", or a finished frontend diff.
| name | mastermind-codegraph-research |
| description | Use mmcg to ground structural code claims before planning, auditing, criticizing, or researching code. Triggers when an agent needs symbol existence, callers, callees, imports, blast radius, file existence, or stale-index handling. |
| metadata | {"version":"0.2.0","authors":["mastermind"],"tags":["workflow","mmcg","codegraph"]} |
The shared structural discovery layer for Mastermind. Claims about symbol existence, indexed callers, imports, and bounded blast radius come from a fresh mmcg result rather than memory. Exact source contracts and runtime behavior come from source reads and tests.
The graph is syntactic evidence: name resolution, dynamic dispatch, reflection, generated code, re-exports, and cross-language edges can reduce precision.
Never name a symbol, file, caller, or blast radius from memory. "I think X exists" is not evidence; mmcg_search X returning a hit is. A spec, audit, or critique built on a guessed symbol fails at the first step that touches real code.
Grep / Read. mmcg doesn't index strings.| Question | Tool |
|---|---|
Does symbol X exist? (get file:line + signature) | mmcg_search |
What calls X? | mmcg_callers |
What does X call? | mmcg_callees |
If I change/rename X, what breaks? (transitive) | mmcg_impact |
| What does file Y import? | mmcg_imports |
Who imports X / this path? | mmcg_imported_by |
| Does this file path exist in the index? | mmcg_files |
| Is the index ready / how stale is it? | mmcg_status |
| String contents / comments / log lines | Grep |
| File-name / extension globs | Glob |
mmcg-first: use the graph to find candidate symbols and impact, then read the source needed for the decision. Re-check with literal search or another source when collisions/precision warnings are present, a security-sensitive path is at stake, a meaningful zero-result could change the decision, or the language feature is outside the graph's precision envelope. Do not repeat equivalent searches when the indexed result already answers a low-risk discovery question.
mmcg_status response → mmcg is unavailable. For a low-risk task, proceed
with source inspection and state the limitation. Stop for user/planner review
only when the missing structural evidence is load-bearing to scope or safety.mmcg_status reports stale files → re-index (mastermind watch, or a fresh index) before trusting structural answers. A stale graph is worse than none: it looks authoritative and is wrong.Whenever you read code or report a structural fact, carry the file:line. Downstream agents (planner, auditor) need precise locations to act — a finding without a citation is unverifiable and gets treated as a guess.
Catch yourself guessing a signature, a path, or a caller count → query the graph
or read the source. Preserve stale, collision, precision, and truncation
metadata with any downstream claim.
For any conclusion that changes scope, risk, or implementation, separate:
high, medium, or low, plus a concrete reason. Never derive
confidence from the number of search hits alone.Do not hide a speculative leap inside an observed-facts paragraph. A zero result means "not found in this index under this query," not "does not exist."