spec_system

Grill an implementation. Audits code for correctness, security, error handling, testing quality, observability, and overcomplexity. When a spec and/or task list are available, also verifies spec compliance and task completeness. Works on any code — from a full feature with spec and tasks, to a bare git diff with no context. Spawns a separate read-only agent. Use after implementation, before merging, or when the developer agent claims work is done. Triggers on "grill code", "grill my code", "grill the code", "grill implementation", "review code", "review implementation", "audit code", "check implementation", "verify work", "is this actually done".

Grill a specification or design document. Critically analyses specs for ambiguity, missing edge cases, security gaps, overcomplexity, operability blind spots, and traceability failures. Spawns a separate agent in read-only mode. Works on any specification, design document, RFC, or ADR — with enhanced structural checks when reviewing plan-spec format specs. Triggers on "grill spec", "grill my spec", "grill the spec", "review spec", "red team spec", "audit spec", "critique spec".

Creates detailed feature specifications and implementation plans with TDD requirements, BDD scenarios in Given-When-Then format, and comprehensive test datasets for boundary conditions, edge cases, and error scenarios. Use when planning a feature, designing a specification, writing a spec, preparing a plan, or when asked to design, specify, or plan functionality.

Reads an existing spec and current implementation, identifies where the code has diverged from the spec, and produces an updated spec as a new versioned file. Use after implementation when the spec needs updating to reflect what was actually built. For review-driven spec updates, use /plan-spec --revise.

Validate your JSON output against a workflow schema template and write it to disk. Use this whenever a prompt instructs you to produce structured JSON — run outvalid with the specified schema, read the errors, fix your output, and retry until validation passes.

Audit a project's data model maturity across five levels: Existence, Completeness, Stability, Correspondence, and Sufficiency. Based on the principle that good data structures make good code inevitable. Examines schema files, migrations, OpenAPI specs, protobuf definitions, ORM models, and any data model definitions. Uses git history to measure schema-first development and churn rates. Produces a structured report with per-level scores and actionable recommendations. Use this skill whenever the user asks to "audit data model", "check schema quality", "data model review", "schema maturity", "evaluate data structures", "is my schema good", "review my database design", "check data model", or any request to assess whether a project's data model is well-designed or follows schema-first principles. Also trigger when someone asks about the relationship between their schema and their code, or whether their data model is complete enough to drive development.

Creates detailed feature specifications and implementation plans with TDD requirements, BDD scenarios in Given-When-Then format, and comprehensive test datasets for boundary conditions, edge cases, and error scenarios. Use when planning a feature, designing a specification, writing a spec, preparing a plan, or when asked to design, specify, or plan functionality.

Decompose a spec, plan, or text description into structured task graph JSON, validate it against the Structured Task Template Spec, and record the tasks as Beads issues. Use when the user wants to break down work into trackable, validated tasks.

Investigate workflow issues using the system's log files. Produces a structured diagnostic report with root cause analysis, timeline reconstruction, and relevant log excerpts. Fire this skill when the user asks "why did it escalate", "what happened to workflow X", "check the logs", "why did it fail", "what went wrong", "trace the error", "investigate the issue", "debug the workflow", "show me the logs for X", "why is it stuck", or any question about a past or current workflow run that requires reading log data to answer. Also fire when the user pastes log output and asks for an explanation. Output is a Markdown diagnostic report with timeline, root cause, and evidence sections.