Skills

Every skill is a method the workflow runs by name; each row cites the file that defines it.

  • api-design

    Designs or changes API contracts.

    Source Skill name

  • concurrency-proof

    Concurrency review of a change (working tree, branch, pull request or proofline run) that reports only proven problems.

    Source Skill name

  • contract-proof

    Proves backward and forward compatibility of changed API, event and data contracts (OpenAPI/Swagger, AsyncAPI, JSON Schema, GraphQL SDL, protobuf, whichever the repository has).

    Source Skill name

  • data-design

    Designs the data side of a change for the project's actual database: data model, Mermaid ERD, invariants, a migration plan with rollback per step, and a data-loss assessment.

    Source Skill name

  • debug-proof

    Debugs a failure with proof instead of guesses.

    Source Skill name

  • design-sync

    Checks durable design documents against the code and reports drift as doc:line -> missing code fact.

    Source Skill name

  • docs-writer

    Writes and maintains project documentation that matches the code and follows a writing profile.

    Source Skill name

  • evidence-map

    Bounded codebase investigation that returns compact, verified evidence.

    Source Skill name

  • explain-html

    Explains a topic, a task, a review or a completion report as one self-contained HTML page with inline SVG diagrams, inline CSS and JS, a light and a dark theme, and no network request.

    Source Skill name

  • exploit-proof

    Defensive security review of a change in the user's own repository (working tree, branch or pull request) before merge.

    Source Skill name

  • intent-grill

    Interviews the user about a vague or high-stakes request before a task is written.

    Source Skill name

  • llm-wiki

    Keeps a persistent, Git-backed Markdown wiki under docs/wiki that grows with the project instead of rediscovering facts on every question.

    Source Skill name

  • performance-proof

    Proves or refutes a performance regression by measurement, never by reading code.

    Source Skill name

  • plan-contract

    Writes an executable, scoped PLAN.md for one task and validates it mechanically.

    Source Skill name

  • product-assess

    Tests the assumptions behind a product plan: it lists them with risk type, importance and evidence level, locks probe thresholds before the probes run, runs a premortem, records one decision (build, clarify or stop) and later records the outcome against a measured baseline.

    Source Skill name

  • product-discovery

    Discovers a product problem before a PRD exists: it frames the problem, researches the market from sources, synthesises interviews, compares competitors with a source and date per cell and lists opportunities with their evidence.

    Source Skill name

  • product-spec

    Writes a product requirements document with the user: problem, goals, non-goals, user stories, numbered requirements, measurable success metrics and open questions, with every claim marked as fact, assumption or decision.

    Source Skill name

  • proof-review

    Reviews a code change (uncommitted work, a branch, or a PR) and reports only what it can prove.

    Source Skill name

  • proofline

    The user-facing /proofline entry point.

    Source Skill name

  • proofline-lead

    The proofline lead's lifecycle method for the main session (AGENT_ROLE unset = lead).

    Source Skill name

  • qa-plan

    Independent QA in two phases.

    Source Skill name

  • system-design

    Designs the smallest valid architecture for a task, or reverse-engineers docs/design/ARCHITECTURE.md from an existing codebase.

    Source Skill name

  • task-author

    Turns a one-sentence request or an external issue into the canonical TASK.md: goal, context, numbered observable acceptance criteria, out of scope, constraints, risks and a suggested class.

    Source Skill name

  • tdd-proof

    Implements a change test-first with recorded proof.

    Source Skill name

  • test-trust

    Audits how far a test suite can be trusted, using runnable evidence instead of opinion.

    Source Skill name

  • ui-design

    Designs UI by visitor mode for each surface.

    Source Skill name

  • verify-proof

    Mechanical proof that a frozen plan was implemented: every acceptance criterion has a successful test or recorded command evidence, changed paths stay inside scope, and build, test and lint commands pass.

    Source Skill name

  • visual-proof

    Measures UI states with existing browser tools, including an optional design audit and source lint, and records judged concerns separately.

    Source Skill name