mast

Author or restructure a single .mspec via /mast:spec, or apply typed AST mutations via /mast:spec

Answer corpus questions via mast list/graph/describe and /mast:spec

Adversarially review a target spec, report gaps

Analyze a candidate spec through OOP boundary lens, propose refactors

Unified verification across four modes. Pre-push (default) runs local CI checks. CI-fix diagnoses GitHub Actions failures and fixes in one commit. Audit scores architecture, spec corpus, and code alignment. Pre-flight verifies referenced files exist before implementation. Use when asked to "check", "am I ready to push", "pre-push", "CI failed", "fix CI", "why is CI red", "audit", "health check", "what's stale", "what needs attention", "check layer separation", "grade this repo", "pre-flight", or "verify files exist".

The continuity companion — use when the user asks "what's next", "where am I", "what should I work on", "what's ready to ship", "what's blocked", "give me a worklist", "continue", "what's the state of the corpus", or returns to a project and needs to re-orient on what to do next. This skill is subagent-driven (it does NOT run the corpus sweep in the main context) so the main thread stays uncluttered: it dispatches a sub-agent to query the read-only projections, compute each spec's standing, and rank candidates, then returns a terse ranked worklist. It SURFACES and RANKS; it never decides — selecting what to actually do is a value judgment the human owns. For a guided tour of an unfamiliar corpus use `/mast:orient`; to implement a chosen spec use `/mast:dag-plan`; to audit health use `/mast:check`.

Use when the user asks to extract, mine, harvest, scaffold, or seed a mast corpus from an existing codebase — typically right after `mast spec init` on a new repo, or when an existing repo needs a candidate `.march` / `.mtypes` / `.mspec` set drafted from the code as it actually is. This skill orchestrates sub-agents (it does NOT read source files in the main context) so the main thread stays uncluttered and tokens don't decay. The output is a proposal manifest with confidence levels and explicit uncertainty flags — never direct writes. Triggers on phrases like "mine this codebase for specs", "extract conventions from this repo", "scaffold mast for X", "what features does this project have", "draft .march files from the current code", "seed the corpus from source", "I just installed mast and need an initial corpus", "infer the edge-type vocabulary from the transports we use", "discover the domains in this project". For authoring a single file from a finished proposal, hand off to `/mast:spec`. For the conceptual grounding (why we mine descriptively and not prescriptively), see `/mast:orient` Mode E.

Unified orientation skill: narrative corpus walkthrough, spec deep-dive, task routing, code-to-spec reverse lookup, and conceptual Q&A about the mast model. Use whenever a user asks to "tour", "walk through", "explain", "give me the lay of the land", "what is this codebase", "where do I start", "show me around", "help me understand X", "explain the layered model", "what's the relationship between spec and arch", "what goes at each layer", "how does attachment work", "what does DDD say about this", "which spec governs this file", "what spec covers this code", "I'm looking at file X", or asks open-ended orientation or model-level questions about a .mspec corpus. Five modes: corpus overview, spec deep-dive, task routing, file-to-spec reverse lookup, conceptual Q&A.

Read, create, rewrite, or patch .mspec/.march/.mtypes files. Routes internally to read, write, or patch mode based on intent. Triggers on "read/show/inspect spec X", "what depends on/cites/extends/blocks X", "is X ready", "list domains/components/edge-types", "what's attached to X", "create/rewrite spec", "new domain/march/mtypes", "add/update/remove rule", "graduate/amend/retire R<n>", "set rule status", "add/remove boundary", "edit title/header", "declare edge type", "wire feature to component", "attach spec to component".

3 hooks across 2 events