claude-code-harness

Adversarial evaluation of code changes — runs build, tests, plan compliance, and security review. Use after implementing changes or before creating a PR. Usage: /evaluate [story-id] [--quick]

Build a feature from a tracker task (local task, GitHub issue, or Todoist task) or plain description — understand, plan, execute, evaluate, and PR in a streamlined flow, or loop back on a rejected PR with `--rework <PR#>`. Lighter than /story — designed for solo devs and small teams. Usage: /implement <issue-id, task-title, or description> [--discuss] [--research] [--quick] [--auto] [--full] [--autonomous] [--rework <PR#>]

Goal-driven story execution — understand → define goal → plan → execute → verify → e2e gate → PR. Done = goal met (acceptance + e2e gate green), not just compiles. Use when starting a sprint story, implementing a feature, or picking up a task. Usage: /story <story-id> [--auto] [--autonomous]

Reconcile merged PRs and completed work against open tracker items. Mode-aware: works in local, tracker, and both modes. Supports --dry-run and --import-backup [file]. Usage: /sync-tracker [--dry-run] [--import-backup [file]]

Execute XML tasks from the story plan wave by wave (Phase 3 only — no understand, no plan, no PR). Use when resuming a story that already has a task plan. Usage: /run-tasks <story-id> [--auto]

Root cause diagnosis when the 3-attempt rule triggers — stop retrying and diagnose. Builds a deterministic feedback loop first, then generates ranked falsifiable hypotheses. Mandates regression tests and instrumentation cleanup. Aliases: /debug, /diagnose. Usage: /debug

Run local build, tests, integration testing, and the feature's e2e goal gate. Stack-agnostic — reads commands from tasks/lessons.md (enterprise) or tasks/notes.md (solo). Use before committing, after code changes, or when another skill asks you to verify. Usage: /local-test [1|2|3|e2e]

Strict RED-GREEN-REFACTOR enforcement with vertical slicing. One behavior at a time — write a failing test, make it pass, clean up. Forbids horizontal slicing and refactoring while RED. Usage: /tdd <feature or behavior to implement>

Plan an effort too big for one session as a map of decision tickets on the tracker, resolving one per session until the way is clear — then hand the resolved spec to the build pipeline as a grill-summary. Mode-aware (local/tracker/both). Usage: /wayfinder <loose idea> | /wayfinder <map ID> [<ticket ID>]

Check for and apply updates to your claude-code-harness installation. Resolves target (project/global/both), fetches the harness source on demand, shows the changelog, and applies updates with human confirmation. Usage: /update-harness [--global|--project] [--pin <version>|--latest|--local <path>]

Decompose planning artifacts into tracker tasks with proper hierarchy — issues (stories) grouped by milestone, with sub-issues (tasks) for breakdown. Backend-aware: creates GitHub issues, ADO work items, Todoist tasks, or local task files depending on your tracker mode. Reads from grill-summary, research, architecture, decision-brief, or PRD — no single artifact required. Usage: /to-issues [--milestone "<name>"] [--project "<name>"]

Deep behavioral bug investigation — system compiles but does the wrong thing. Traces data flow through 5 mandatory iterations, then produces an XML task plan for /run-tasks. Use when something is wrong with the output, logic, or answers — NOT for build failures (use /debug for those).

Personal assistant — answers questions about sprint status, blockers, PRs, and todos from your task files, then offers to update them in sync. Usage: /pa <question>. E.g. "/pa what's blocking me?", "/pa status of #10167", "/pa mark the N8N blocker resolved"

Read open issues, prioritize your work, and create a simple plan. Designed for solo devs — no sprints, no story points, just priorities. Usage: /plan [milestone-name]

Sprint planning — reads issues from your tracker and codebase, creates the sprint file, and surfaces gaps. Use when starting a new sprint, planning sprint N, or setting up a sprint file. Usage: /sprint-plan <N>

Detect and interactively resolve drift across the 7 enterprise task files and project artifacts (PRD, ARCHITECTURE.md, ADRs, work items). Invoked automatically when the drift-check hook hard-blocks on a contradiction. Usage: /sync-tasks [--report-only]

Decompose planning artifacts into Todoist tasks with proper hierarchy — milestones as uncompletable parent tasks, work items as subtasks with descriptions, priorities, and dependencies. Reads from grill-summary, research, architecture, decision-brief, or PRD — no single artifact required. Usage: /to-todoist [--project "X"] [--section "Y"] [--dry-run]

Route incoming issues through a 5-state workflow (needs-triage → needs-info / ready-for-agent / ready-for-human / wontfix) with bug/enhancement categorization. Applies labels, adds context notes, reproduces bugs. Usage: /triage <issue-id>

Run 5 critique axes on an ARCHITECTURE.md — NFR-architecture fit, failure modes, cost stress-test, security posture, operability. Reports severity-tagged findings. Read-only — does not modify the architecture doc. Usage: /architect-critique <path-to-ARCHITECTURE.md> [--prd <path-to-PRD>]

Design system architecture from a PRD or grill-summary — produces ARCHITECTURE.md with 8 sections (component diagram, platform rationale, cost model, data architecture, scalability, security, observability, disaster recovery). Cloud-agnostic with optional platform-specific extensions. Usage: /architect [<path-to-PRD-or-grill-summary>]

Pre-PRD assumption pass — runs 4 inline phases (decompose, data-quality-audit, assumption-extract, evidence-rank) and produces a Decision Brief with tiered evidence thresholds and a risk-ranked test plan. Stops teams from building the wrong thing. Usage: /decision-brief

Generate the full project-level spec stack from ARCHITECTURE.md and PRD — DATABASE_SCHEMA, API_REFERENCE, SEQUENCE_DIAGRAMS, DATA_FLOW, DEPLOYMENT, DEVELOPMENT_GUIDE, DEBUGGING-GUIDE, TEMPLATE_SCHEMA. The missing link between DEFINE and BUILD. Usage: /design-artifacts [all | doc-name ...]

Decision-tree interrogation of a plan, design, or proposal. Asks serial questions with recommended answers until shared understanding is reached. Use when stress-testing a Decision Brief, PRD, architecture doc, or refactor proposal. Usage: /grill-me <plan or design to stress-test>

Stress-test an architectural plan or design against CONTEXT.md and ADRs. Like /grill-me but anchored in documented domain language, glossary terms, and prior decisions. Updates CONTEXT.md with resolved terms and proposes ADRs when decisions are hard-to-reverse + surprising + trade-off-driven. Usage: /grill-with-docs <plan or design to stress-test>

Find modules where the interface is as complex as the implementation, propose deepening refactors that increase locality and leverage. Reads CONTEXT.md and ADRs, applies the deletion test, updates the glossary with new terms. Usage: /improve-codebase-architecture [area or module]

Run 6 critique checks on a PRD — metric validity, NFR specificity, failure-mode coverage, assumption traceability, rollback plan, intent clarity. Reports severity-tagged findings. Read-only — does not modify the PRD. Usage: /prd-critique <path-to-PRD> [--brief <path-to-decision-brief>]

Throwaway prototyping — creates 1-3 candidate approaches in _prototype/<feature>/, writes a decision.md comparing trade-offs, and cleans up losers after the user picks a winner. Usage: /prototype <feature or question to explore>

Research an external API, integration, or unfamiliar library and cache findings in research.md with provenance-tagged claims. Future agents (planner, executor) read this file for context. Usage: /research <topic> [--urls <url1> <url2> ...]

Drive a PR to zero review threads — fetch, categorize, fix, reply, repeat. Use when handling Code Rabbit comments, fixing PR review feedback, or working through the CR cycle. Usage: /babysit-pr <PR_ID>

Deploy to cloud and verify — branch-test before merge, or post-merge production verification. Use when deploying, verifying a deployment, re-ingesting data, or running smoke tests. Usage: /deploy [branch-test|post-merge|re-ingest|smoke-test]

Learning effectiveness dashboard. Shows how learnings are performing, promotes high-scoring ones to permanent rules, archives ineffective ones. Usage: /calibrate

Self-improvement loop for the harness. Reads recent sessions, evaluations, blockers, and lessons, surfaces recurring friction patterns, and proposes specific harness edits. Never auto-applies. Usage: /improve-harness [days]

High-level map of unfamiliar code — shows relevant modules, caller relationships, dependencies, and how a piece of code fits in the bigger picture. Conversational output, no file artifact. Usage: /zoom-out [file or module path]

Generate a Product Requirements Document for a new feature — clarifying questions, stories sized to one context window, ordered by dependency. Use when planning a feature, starting a project, or asked to write a spec or PRD.

UI/UX design intelligence for web and mobile. Includes 50+ styles, 161 color palettes, 57 font pairings, 161 product types, 99 UX guidelines, and 25 chart types across 10 stacks (React, Next.js, Vue, Svelte, SwiftUI, React Native, Flutter, Tailwind, shadcn/ui, and HTML/CSS). Actions: plan, build, create, design, implement, review, fix, improve, optimize, enhance, refactor, and check UI/UX code. Projects: website, landing page, dashboard, admin panel, e-commerce, SaaS, portfolio, blog, and mobile app. Elements: button, modal, navbar, sidebar, card, table, form, and chart. Styles: glassmorphism, claymorphism, minimalism, brutalism, neumorphism, bento grid, dark mode, responsive, skeuomorphism, and flat design. Topics: color systems, accessibility, animation, layout, typography, font pairing, spacing, interaction states, shadow, and gradient. Integrations: shadcn/ui MCP for component search and examples.