Skill Claude CodeCodex
Trigger when the user asks whether implemented code still matches its spec, weeks or months after implementation: "check spec drift", "does the code still match this spec?", "has this spec drifted?". Given a spec@1 (via specfilepath) and a workspace root, reads the spec from disk and the implementation from the…
Skill Claude CodeCodex
Trigger when the user asks whether a spec matches its source: "is this grounded?", "does this spec match the requirement?", "check this spec against the requirement". Given a draft spec@1 plus the originating requirement@1 and optional research-report@1, checks whether each acceptance criterion is grounded in the…
Skill Claude CodeCodex
Trigger this skill when the user says "verify this", "check this spec", "check this plan", "check this requirement", "check this PR", "is this done?", "review this for completeness", "does this meet the criteria?", "gate this", or "quality check" — also before stage transitions: starting implementation from a spec…
Agent
Delegate to this subagent to execute exactly one task from a plan@1 via the full TDD cycle. Input is a single task object from a plan@1, a workspace manifest from recon, and optionally precisiontests from mutator. When specfilepath is set, reads the task's coverscriteria acceptance criteria from that file before…
Agent
Delegate to this subagent after implementer commits and before exit-gate runs. Input is the workspace manifest from recon and the list of files touched by implementer in the current task cycle. The agent detects the workspace language (Cargo.toml → rust uses cargo-mutants; package.json → typescript/javascript uses…
Agent
Delegate to this subagent after a implementer task commits and before exit-gate runs. Input is the set of commits from a completed implementation task, the original task specification, and the workspace manifest (which carries the language field). The agent loads the language-specific hazards reference (rust-hazards…
Skill Claude CodeCodex
Activate when the user says "implement this", "execute the plan", "write the code", "build this feature", "generate tests for X", "write tests for this spec", "refactor this without changing behavior", or "explain what this code does" — also when handed a plan@1 to execute, or a spec@1 exists but no plan. Every task…
Agent
Delegate to this subagent after recon produces a source map. Input is the source map from recon plus the original research question. The agent reads every identified source and extracts findings relevant to the question, with a citation, confidence level (confirmed, likely, or assumed), and source attribution for…
Agent
Delegate to this subagent when a proposed approach or research-report@1 needs risk review before a spec is written. Input is a proposed approach or research-report@1 JSON, optionally accompanied by context about the target environment. The agent identifies technical risks that would materially affect the spec or plan…
Agent
Delegate to this subagent after reader produces a findings list. Input is the findings array and reasoning from reader, along with the original research question. The agent synthesizes findings into a coherent research-report@1, surfaces contradictions between sources, identifies patterns across findings, and lists…
Skill Claude CodeCodex
Activate when the user says "I need to understand X before deciding", "is there a way to do Y", "what are the options for Z", "research this", "investigate this", "what does the codebase already do for X", "prior art on Y", or "what are the risks of approach Z" — also when the approach for a spec or plan is unclear…
Agent
Delegate to this subagent when a requirement@1 draft from intake needs review for missing or underspecified dimensions before a spec can be written. Input is a requirement@1 JSON object. The agent evaluates gaps in priority order: testability of donewhen criteria, specificity of stakeholder, and explicitness of…
Agent
Delegate to this subagent when you have a raw need statement in free text — a problem description, user story, or feature request — and need it converted into a structured requirement@1 draft. Input is unstructured text only; no schema is required from the caller. The agent infers id, statement, stakeholder, why, and…
Agent
Delegate to this subagent when a set of requirement@1 drafts needs ranking before planning starts. Input is a list of requirement@1 objects. The agent ranks them by impact (who is affected and how badly), urgency (signals in why/donewhen indicating time pressure), and dependency order (a requirement that another…
Skill Claude CodeCodex
Trigger when the user asks about overall backlog health: "what's in the backlog", "audit the backlog", "what requirements are missing coverage". Cross-references all open requirement@1 objects against existing specs and implementation files. Returns per-requirement status (covered, partial, missing, duplicate) with…
Skill Claude CodeCodex
Trigger when the user describes a raw need in free text: "we need X", "the problem is Y", "users are asking for Z". Converts unstructured text into a structured requirement@1 draft — id, statement, stakeholder, why, donewhen — without asking clarifying questions. Fields that cannot be inferred are omitted, not…
Skill Claude CodeCodex
Trigger when a requirement@1 draft from weaver/capture-need has gaps to close before a spec can be written: "is this requirement complete?", "what's missing from this requirement". Given a requirement@1, evaluates gaps in priority order — testability of donewhen, specificity of stakeholder, explicitness of outofscope…
Skill Claude CodeCodex
Trigger when the user asks whether a specific requirement already exists or relates to other work: "is this already captured?", "does this overlap with an existing requirement?", "what does this relate to?". Given one requirement@1 (or a small named set), searches the workspace for related specs and implementation…