harness

Safety-first diagnosis and fix support for CI failures

Error recovery (root cause isolation -> safe fix -> re-verification)

Integrated reviewer that renders verdicts from static/runtime/browser perspectives based on sprint-contract

Integrated scaffolder handling project analysis, scaffolding, and state updates

3-agent composition for Harness v3.

Agent that generates Remotion scene components

Integrated worker that cycles through implementation -> preflight self-check -> verification -> commit preparation, then hands off to independent review

Use this skill whenever the user asks to create CRUD for an entity, generate API endpoints, scaffold a resource, build a REST API for a table, or mentions entity generation. Also use when the user wants to add standard create/read/update/delete operations with authorization rules. Do NOT load for: UI component creation, form design, database schema-only discussions, or non-CRUD business logic. Auto-generates production-ready CRUD operations for entities — API endpoints, database models, validation, search, and pagination.

Use this skill whenever the user asks to review code, review a plan, check quality, analyze scope, run security checks, examine PRs or diffs, or runs /harness-review. Also use when the user wants a second opinion on changes, performance review, or pre-merge quality gate. Do NOT load for: code implementation (use harness-work), new features, bug fixes, project setup, or release. Unified review skill for Harness v3 — multi-angle code, plan, and scope review with optional dual-reviewer and security analysis.

Use this skill whenever the user mentions setup, initialization, starting a new project, CI setup, Codex CLI setup, harness-mem, agent configuration, symlinks, mirrors, or runs /harness-setup. Also use when the user needs to configure the harness environment or onboard a new repository. Do NOT load for: code implementation (use harness-work), code review (use harness-review), release (use harness-release), or planning (use harness-plan). Unified setup skill for Harness v3 — project initialization, tool configuration, 2-agent setup, memory config, symlinks, and mirror sync.

Use this skill whenever the user asks to sync progress, check what's done, see current status, asks 'where am I', 'how far along', or runs /harness-sync. Also supports --snapshot for progress snapshots. Do NOT load for: creating new plans (use harness-plan), code implementation (use harness-work), code review (use harness-review), or release. Syncs progress between Plans.md and actual implementation — detects drift, updates task markers, and runs retrospectives.

Use this skill whenever the user asks to implement, execute, build, code, 'do everything', 'run all tasks', or mentions harness-work, breezing, team run, parallel execution, or --codex. Also use when the user selects specific task numbers or ranges to execute. Do NOT load for: planning (use harness-plan), code review (use harness-review), release (use harness-release), or project setup (use harness-setup). Unified execution skill for Harness v3 — implements Plans.md tasks from single task to full parallel team runs.

Use this skill whenever the user mentions SSOT, decisions.md, patterns.md, memory search, save learnings, record a decision, harness-mem, sync memory, promote patterns, merge plans, migrate memory, or recall past decisions. Also use when the user wants to persist cross-session knowledge or search for previously recorded patterns. Do NOT load for: code implementation, code reviews, ad-hoc notes, or in-session task logging. Manages SSOT (Single Source of Truth) memory — decisions.md, patterns.md, cross-tool memory search, and memory sync.

Use this skill whenever the user mentions NotebookLM, wants to create YAML for NotebookLM, needs structured slide content, or asks for presentation material generation. Also use when the user wants to convert project documentation into NotebookLM format. Do NOT load for: code implementation, bug fixes, code reviews, deployments, or general slide images (use generate-slide instead). Generates NotebookLM-compatible YAML documents and structured slide content.

Use this skill whenever the user asks to create UI components, build a hero section, design a landing page section, create feedback or contact forms, or generate front-end visual elements. Also use when the user mentions component generation, form creation, or UI scaffolding. Do NOT load for: authentication features, backend API implementation, database operations, or business logic. Generates UI components, hero sections, and feedback/contact forms with production-ready styling.

Use this skill when the user appears non-technical, asks 'what should I do next?', 'how does this work?', seems confused about the development process, or needs step-by-step guidance in plain language. Do NOT load for: experienced developers making direct implementation requests, code reviews, or technical debugging. Guides non-technical users (VibeCoders) through natural language development — explains what to do next, how the system works, and how to express requirements.

Use this skill when the user asks how the 2-agent workflow operates, wants to understand the collaboration process, or needs guidance on Cursor/CC roles and responsibilities. Do NOT load for: actual implementation work, executing handoffs (use cc-cursor-cc instead), or workflow configuration setup. Reference guide for Cursor ↔ Claude Code 2-agent collaboration workflow — explains roles, handoff patterns, and process flow.

Use this skill whenever the user mentions CI failures, GitHub Actions errors, build pipeline issues, test failures in CI, red builds, or needs to debug why a pipeline broke. Also trigger when the user pastes CI logs or error output from a remote build. Do NOT load for: local builds, local test runs, standard implementation work, code reviews, or project setup. Diagnoses and fixes CI/CD pipeline failures — analyzes logs, identifies root causes, and applies fixes.

Auto-triggered when reviewing PRs that modify the Feature Table in CLAUDE.md or docs/CLAUDE-feature-table.md. Internal use only. Do NOT load for: standard code implementation, general PR reviews, project setup, or non-Feature-Table documentation changes. Quality guardrail for Claude Code update integration — detects doc-only Feature Table additions without corresponding implementation and requires implementation proposals.

Use this skill whenever the user asks to visit a URL, click elements, fill out forms, scrape content, automate a website, log into a service, take a screenshot of a page, or perform any browser-driven workflow. Also use when the user needs UI testing, visual regression checks, or web data extraction. Do NOT load for: sharing or embedding URLs in code, reading local screenshot image files, or non-browser file operations. Automates browser interactions — navigate pages, fill forms, take screenshots, extract web data, and test web apps.

Use this skill whenever the user mentions login, signup, authentication, OAuth, session management, payments, subscriptions, billing, Stripe integration, or checkout flows. Also use when the user needs to protect routes, add role-based access, or implement payment webhooks. Do NOT load for: general UI components, database schema design, non-auth API endpoints, or business logic unrelated to auth/payments. Implements authentication and payment features using Clerk, Supabase Auth, or Stripe.

Use when user says 'breezing', 'do everything', 'run the whole plan', 'team run', 'full auto', or wants all tasks executed end-to-end with parallel workers. Do NOT load for: single-task implementation, planning, code review, release, or setup. Team execution mode — runs Plans.md tasks with full parallel team orchestration. High-level alias for harness-work in team mode.

Use this skill when the user mentions Cursor PM handoff, 2-agent plan validation, CC-Cursor round trip, brainstorm review with Cursor, or cross-agent plan sync. Do NOT load for: direct implementation work, single-agent coding tasks, or standard code editing. Validates brainstormed ideas with Cursor PM agent, updates Plans.md, and hands off between agents in the Cursor ↔ Claude Code 2-agent workflow.

Use this skill whenever the user mentions deploying, pushing to production, Vercel setup, Netlify config, deployment monitoring, health check endpoints, or production analytics. Also use when the user wants to verify a deployment is healthy or set up post-deploy monitoring. Do NOT load for: feature implementation, local development, code reviews, or project initialization. Configures and executes deployments to Vercel or Netlify, including analytics setup and health checks.

Use this skill whenever the user asks for a project slide, one-page visual summary, project introduction image, or visual overview of a project. Also use when the user wants to present a project visually or create a promotional image for a repository. Do NOT load for: presentation decks (multiple slides), video generation, text-only documentation, or code implementation. Generates project introduction slide images using Nano Banana Pro (Gemini image API) with 3 visual patterns, quality checks, and retries.

Use this skill whenever the user mentions video generation, product demo videos, visual documentation, animated project overviews, or '/generate-video'. Also use when the user wants to create a video walkthrough of features or a release video. Requires Remotion setup. Do NOT load for: embedding video players in UI, live demo recording, video playback features, or slide generation. Auto-generates product demo videos using Remotion — architecture overviews, feature demos, and release announcement videos.

Use this skill whenever the user asks to check, list, search, export, read, or update Google files; when a Google Drive/Docs/Sheets URL or file ID needs parsing; when selecting a Google account or verifying auth; or when troubleshooting gogcli access errors. Also use when the user pastes a Google URL and wants to interact with it. Do NOT load for: general local file operations, non-Google cloud storage (S3, Azure), or standard shell commands. Operates Google Workspace files via gogcli CLI — Drive, Sheets, Docs, and Slides.

Use this skill whenever the user asks to create a plan, add tasks, update Plans.md, mark tasks complete, check progress, sync status, or says 'where am I' or 'what's next'. Also use when the user runs /harness-plan, /harness-sync, or needs to organize work into actionable tasks. Do NOT load for: code implementation (use harness-work), code review (use harness-review), or release tasks (use harness-release). Unified planning skill for Harness v3 — task planning, Plans.md management, and progress sync.

Use this skill whenever the user mentions releasing, version bumps, creating tags, publishing, cutting a release, or runs /harness-release. Also use when the user asks about the release process or wants to finalize and ship changes. Do NOT load for: code implementation (use harness-work), code review (use harness-review), planning (use harness-plan), or project setup. Unified release skill for Harness v3 — automates CHANGELOG updates, version bumps, git tags, GitHub Releases, mirror sync, and release validation.

Use this skill when the user asks about coding principles, development guidelines, safe editing practices, or VibeCoder guidance. Also use when another skill needs to reference general development standards. Do NOT load for: actual code implementation (use harness-work instead), code reviews, or project setup. Reference for development principles, guidelines, and VibeCoder best practices — including diff-aware editing, repo context reading, and safety guardrails.

Auto-triggered by harness-work when session --resume/--fork flags are present. Do NOT load for: user-facing session management, login state, app state handling, or direct user requests. Internal workflow skill — controls session resume/fork(branch) for /work, updates session.json and session.events.jsonl.

Use this skill when the user starts a new session, asks for a status check before beginning work, or wants to verify the environment is ready. Do NOT load for: mid-session implementation work, code reviews, or ongoing task execution. Initializes a work session with environment checks, Plans.md status overview, and readiness verification.

Use this skill when the user asks about previous sessions, wants to continue from where they left off, references past work history, or needs to recall what was decided in earlier sessions. Do NOT load for: code implementation, code reviews, ad-hoc information, or in-session note-taking. Manages cross-session learning and memory persistence — records patterns, decisions, and context that carry across sessions.

Auto-triggered by the orchestration system at /work phase boundaries. Do NOT load for: user-facing session management, login state, app state handling, or direct user requests. Internal workflow skill — manages session state transitions per SESSION_ORCHESTRATION.md, escalated transitions on error, and initialized restoration on session resume.

Use this skill when the user runs /session, asks to list active sessions, check inbox messages, broadcast to other sessions, or manage Claude Code session state. Also use for session lifecycle operations. Do NOT load for: application user sessions, login/auth state, or authentication features. Unified session management — initialization, cross-session memory, state tracking, inbox, and broadcasting.

67 hooks across 26 events

Structured output style optimized for Plan/Work/Review workflow. Provides progress tracking and phase-specific output formats.