skill-issue

Use when the user wants to turn a rough goal into a verifiable goal with an explicit definition of done - ready to run directly via /goal on Claude Code or Codex, or pasted as a prompt. Type /cds-create-goal to use.

Review past coding-agent sessions across Claude Code, Codex, Cursor (IDE + CLI), and Pi over a time window, then produce a structured markdown report plus a single-file HTML dashboard that explains what was worked on, what finished, what's left and WHY (needs review? blocked on a human? superseded by another thread? got distracted?), how parallel worktrees/directories of one project relate (overlap, sequencing, divergence, A/B experiments), and which leftover tasks are orchestratable long-runs vs need hands-on iteration. Use for a morning kickoff or catch-up: 'what did I work on yesterday/this week', 'catch me up on my projects', 'agent standup', 'summarize my coding sessions', 'what's left to finish', 'how should I start my day', 'review my agent sessions'. Default window is yesterday; supports a week or any custom range.

Orchestrate model-parallel code review with selected skill references. Runs a Claude reviewer subagent and a Codex CLI reviewer in parallel with those skill references listed in reviewer prompts as additional review lenses, then synthesizes all findings through a judge agent with semantic deduplication, contradiction detection, and dependency ordering.

Get a second opinion on your current conversation from a different AI model. Routes to Claude from Codex and Codex from Claude using a fresh local shell session.

Review a plan or implementation for model integrity, service boundaries, and canonical architecture direction. Use when a change touches service boundaries, data models, or trust boundaries.

Clarify what should be built before planning begins. Use when the problem is under-defined, the intended outcome is unclear, or there are materially different directions to explore before committing to a plan.

Directly simplify existing code changes for clarity and minimal complexity. Use after writing or modifying code when the user wants unnecessary complexity, defensive patterns, over-abstraction, excess state, or excess line count removed by making edits. Applies to any language. Triggers on direct requests to simplify code, reduce complexity, clean up changes, or make code more elegant. Supports an explicit manual `mode:review` argument for findings-only simplicity review.

Convert a plan of record into Linear-ready issues with dependencies, acceptance criteria, and sequencing. Use when a reviewed plan needs to become executable work in Linear.

Check whether an implementation matches its plan of record. Use when a plan exists and code has been written against it. Flags deviations, missing phases, unimplemented acceptance criteria, and scope drift.

Challenge and strengthen a plan or brainstorm before implementation begins. Use when a plan of record or brainstorm document exists and needs a second pass for scope, sequencing, testing, complexity, or operability gaps.

Create a concrete plan of record from a clarified idea, existing code, or a spec. Use when a task crosses module boundaries, touches service boundaries, or requires choosing between viable approaches.

Classify and route review feedback before implementation begins. Use when review findings exist from any source and need validation, prioritization, and routing to the right next action.

Testing philosophy and conventions. Use when writing, generating, modifying, or auditing test files. Covers behavior-driven testing, async patterns, database testing, and language-specific practices for Swift, Python, and TypeScript.

Builds polished single-file HTML artifacts (reports, decks, formal code reviews, audits, specs, explainers) from markdown, data, or analysis using an editorial newsprint design system. INVOKE ONLY ON EXPLICIT REQUEST that names veneer or asks for a shareable single-file HTML artifact — e.g. "/veneer", "veneer this", "make a shareable HTML report/deck/code review/audit", "single-file HTML artifact for X", "broadside/folio/dispatch this". Also does the reverse — reads veneer HTML and extracts clean markdown for model re-ingestion. Do NOT trigger on a generic "make an HTML page", app or product UI, dashboards inside a running app, web scraping, or marketing/landing pages — those route to frontend-design, firecrawl, or other skills.