code-steward

Write complete, traceable Epics using Liminal Spec methodology. Covers User Profile, Flows, Acceptance Criteria, Test Conditions, Data Contracts, and Story Breakdown.

Publish a detailed epic as two handoff-ready artifacts: a PO-friendly business epic with grouped ACs and a developer story file with full AC/TC detail and Jira section markers.

Explore product direction and produce a PRD before feature scoping. Use when requirements are unclear, scope spans multiple features, or stakeholder alignment is needed.

Orchestrate story-by-story implementation with agent teams. Manages teammates, routes verification, and makes judgment calls across the full implementation cycle.

Orchestrate the full Liminal Spec pipeline with agent teams. Manages drafters and verifiers from orientation through technically enriched stories.

Transform a Feature Spec into implementable Tech Design with architecture, interfaces, and test mapping. Validates the spec as downstream consumer.

Write a single functional story with epic-quality rigor. Story-sized epic for focused changes that don't warrant the full pipeline.

Perform inline technical design and embed results into a functional story. Story-sized tech design + enrichment with codebase analysis, architecture, interfaces, and test mapping.

Orchestrate a comprehensive multi-pass code review by launching specialized review subagents (standard, security, performance, architecture, test-coverage, silent-failure) in parallel, then synthesizing all findings into a single verified report. Use this skill whenever the user asks for a "full code review", "comprehensive review", "run all reviews", "multi-pass review", "thorough review", "complete code review", "review everything", or wants multiple review types run together on the same code. Also trigger when the user wants to review local work without a PR — phrases like "review my changes", "review my branch", "review what I've done", "review my uncommitted changes", "review this branch vs main", "diff review against main", or "what's wrong with my changes" should trigger this skill. This skill handles all scope modes: PR, uncommitted local changes, branch-vs-branch comparison, and specific files or directories. Do NOT trigger for single-type review requests like "security review" or "check performance"

Perform an architecture and design review — the "tech lead" pass that evaluates whether a change fits the existing codebase architecture. Use this skill when the user asks to "review architecture", "architecture review", "design review", "check coupling", "review code structure", "does this fit the codebase", "abstraction review", "module boundary check", "dependency review", "is this well-designed", "review module design", "check for tight coupling", "review layering", "does this belong here", "check dependencies", "review the design of this", or any request focused on whether code is in the right place, at the right abstraction level, with the right boundaries. Also trigger when the user asks about circular dependencies, leaky abstractions, separation of concerns, module organization, or API surface design — even if they don't use the word "architecture" explicitly.

Analyze code for performance bottlenecks, inefficient algorithms, N+1 queries, memory leaks, and resource management issues. Use this skill whenever someone asks for a "performance review", "check for performance issues", "is this code efficient", "review for bottlenecks", "N+1 query check", "memory leak check", "optimization review", "will this scale", "check query efficiency", "review for resource leaks", "find slow code", or any request to evaluate how code will behave under load, at scale, or with real-world data volumes. Also trigger when reviewing a PR or codebase and the user mentions performance, scalability, or resource usage as a concern.

Synthesize and verify multiple code review outputs into a consolidated, verified report. Use when you have outputs from multiple review skills (standard, security, performance, architecture, test-coverage, silent-failure) and need to verify every finding against the actual code, deduplicate, resolve contradictions, and produce a final consolidated report. Trigger phrases — "synthesize reviews", "consolidate review results", "verify and merge reviews", "create final review report", "combine review findings".

Perform a security-focused code review to find exploitable vulnerabilities. Use this skill whenever the user asks for a 'security review', 'security audit', 'vulnerability scan', 'check for security issues', 'review for vulnerabilities', 'OWASP review', 'check for injection', 'auth review', 'is this code secure', 'pentest this code', 'find security bugs', 'check for XSS', 'check for SQL injection', or any request to examine code from an attacker's perspective. Also trigger when reviewing PRs that touch authentication, authorization, user input handling, cryptography, or data exposure — even if the user just says 'review this PR' and the changes are security-sensitive.

Perform an exhaustive silent failure audit — systematically trace every error handling path to ensure nothing fails silently. Use this skill whenever the user asks to "check for silent failures", "review error handling", "are errors being swallowed", "error handling audit", "check for empty catch blocks", "review error messages", "catch block review", "are failures visible", "error propagation review", "check error logging", "find swallowed errors", "audit catch blocks", "review fallback behavior", or any request to verify that errors are properly surfaced, logged, and actionable. Also trigger when reviewing code that contains significant error handling logic (try-catch blocks, error callbacks, fallback patterns, retry mechanisms) even if the user only says "review this" — silent failure analysis is warranted anytime error handling is a meaningful part of the changes.

Perform a standard code review — the general-purpose "senior developer eyes on the code" pass. Use this skill whenever the user asks to "review this code", "code review", "check code quality", "review this PR", "look over my changes", "is this code well-written", "check for bugs", "review my implementation", or any general request to evaluate code quality, correctness, or readability. This is the default review type — trigger it for any review request that doesn't specifically ask for security analysis, performance profiling, architecture review, test coverage audit, or exhaustive error handling analysis. Even casual requests like "take a look at this", "does this look right", or "anything wrong here?" in a code context should trigger this skill.

Review test coverage quality and completeness — the "if this code breaks next month, will the tests catch it?" pass. Analyzes behavioral coverage gaps, test quality, test design issues, and missing test scenarios. Use this skill whenever the user asks to "review test coverage", "check test coverage", "are the tests good enough", "review tests", "test quality review", "what tests are missing", "check for test gaps", "is this well tested", "audit test coverage", "review my test suite", "test review", or any request specifically about test quality, test completeness, missing test scenarios, or whether tests adequately protect against regressions. Also trigger when the user asks "will these tests catch bugs", "are these tests sufficient", "test gap analysis", "what should I test", "do I need more tests", or "review the test changes in this PR". Do not trigger for general code review requests — those go to standard-review.

Generate and update wiki-style documentation for codebases using AST-aware structural analysis. Use when the user wants to generate documentation, update stale documentation, create a documentation wiki, or assess documentation coverage for a repository.

Build agents, tools, and multi-agent systems with the Codex Agent SDK (formerly Codex SDK). Use this skill whenever the user wants to plan, design, implement, debug, or extend anything involving the Codex Agent SDK, Codex-agent-sdk, @anthropic-ai/Codex-agent-sdk, claude_agent_sdk, agent automation, programmatic Codex usage, MCP tool servers, agent hooks, subagents, or SDK-based agent orchestration — even if they just mention "agent SDK", "building agents programmatically", or want to create tools for Codex to use. Covers the complete API for TypeScript (default) and Python.

Generate and update wiki-style documentation for codebases using AST-aware structural analysis. Use when the user wants to generate documentation, update stale documentation, create a documentation wiki, or assess documentation coverage for a repository.

Build agents, tools, and multi-agent systems with the Claude Agent SDK (formerly Claude Code SDK). Use this skill whenever the user wants to plan, design, implement, debug, or extend anything involving the Claude Agent SDK, claude-agent-sdk, @anthropic-ai/claude-agent-sdk, claude_agent_sdk, agent automation, programmatic Claude usage, MCP tool servers, agent hooks, subagents, or SDK-based agent orchestration — even if they just mention "agent SDK", "building agents programmatically", or want to create tools for Claude to use. Covers the complete API for TypeScript (default) and Python.