Jarvis Skill Reference

Disciplined diagnosis loop for hard bugs and performance regressions. Reproduce → minimize → hypothesize → instrument → fix → regression-test.

Test-driven development with red-green-refactor loop. Builds features or fixes bugs using TDD methodology.

Build a throwaway prototype. Routes to a runnable terminal app (state/logic questions) or multiple UI variations on one route.

Finds deepening opportunities in a codebase — consolidate tightly-coupled modules, improve testability and AI navigability.

Verify a code change actually does what it's supposed to by running the app and observing behavior in the browser/terminal.

Review the current diff for correctness bugs. Pass --comment to post findings as inline PR comments. Levels: low / medium / high / max.

Launch and drive this project's app. Finds the project skill for launching; falls back to built-in patterns (CLI, server, TUI, browser).

Initialize a new project — scaffolds CLAUDE.md, sets up project structure and initial configuration.

Full PR review — reads the diff and comments inline on GitHub. Broader than /code-review, includes architecture observations.

Turn any folder of files into a navigable knowledge graph. Community detection, HTML viz, GraphRAG JSON, and audit report. Runs on Obsidian vault by default.

Tell the agent to give broader context or a higher-level perspective. Use when you need to understand how something fits into the bigger picture.

Turn the current conversation context into a PRD and publish it to the project issue tracker. Synthesizes what's already known — no interview needed.

Break a plan, spec, or PRD into independently-grabbable GitHub issues using tracer-bullet vertical slices.

Triage issues through a state machine. Labels: needs-triage → needs-info → ready-for-agent / ready-for-human → wontfix.

Relentless interview about a plan or design until reaching shared understanding. Resolves every branch of the decision tree with recommended answers.

Like /grill-me but challenges your plan against the existing domain model. Updates CONTEXT.md and ADRs inline as decisions crystallize.

Create new agent skills with proper SKILL.md structure, progressive disclosure, and bundled resources. Outputs install-ready skill directory.

Autonomous multi-session goal execution. Dispatches agent waves, commits progress, loops until the goal ships. Use --session <id> to resume.

Check status of the current /goal session — what phases are done, what's in progress, what's next.

Compact the current conversation into a handoff document so another agent can pick up exactly where you left off. Saved to OS temp dir.

Write a dated session log to the Jarvis Obsidian vault and add the entry to Sessions/_Index.md. Run at end of every work session.

Write a standardized wake-up handoff after an overnight build — gathers real git/PR state and summarizes what's left for Kevin.

Delta-aware 3-bucket Slack-formatted update (CRM admin/rep · Personal · non-CRM). Covers only what shipped since last delivery. Scans for leaked secrets.

Run a prompt or slash command on a recurring interval. Omit the interval to let the model self-pace. Good for polling CI, babysitting PRs, autonomous loops.

Create, update, list, or run scheduled remote agents (routines) on a cron schedule. Also handles one-time scheduled runs ("run this at 3pm").

Deploy FLUXATH CRM to Railway. Commits, runs clean-worktree safety audit, pushes, runs railway up --detach, verifies healthz 200. Never deploy from dirty worktree.

Write a structured error_slug.md post-mortem after a mistake. Indexes it in MEMORY.md and checks for recurring patterns (3 = escalate to feedback).

Ultra-compressed communication mode. Cuts token usage ~75% by speaking like a caveman while keeping full technical accuracy. 6 intensity levels.

Ultra-compressed commit message generator. Conventional Commits format. Subject ≤50 chars, body only when "why" isn't obvious.

Compress memory files (CLAUDE.md, todos, preferences) into caveman format to save input tokens. Overwrites original; backup saved as FILE.original.md.

Ultra-compressed code review comments. Each comment is one line: location, problem, fix. No throat-clearing.

Quick-reference card for all caveman modes, skills, and commands. One-shot display — does not activate a mode.

Security-focused code review. Checks for OWASP top 10, injection flaws, auth bypasses, secrets in code, and insecure configurations.

Configure Claude Code via settings.json — hooks ("from now on when X"), permissions ("allow npm"), env vars ("set DEBUG=true"), MCP servers.

Customize keyboard shortcuts — rebind keys, add chord bindings, modify ~/.claude/keybindings.json.

Scans transcripts for common read-only Bash and MCP tool calls, then adds a prioritized allowlist to .claude/settings.json to reduce approval prompts.

Forge a new skill from scratch — interviews you on triggers, behavior, and resources, then generates a complete SKILL.md-based skill directory.

Forge a new specialized agent — generates agent definition with tools, system prompt, and routing rules.

Improve an existing skill — analyzes usage patterns, identifies gaps, and suggests or applies targeted improvements to SKILL.md.

Find the right skill for a task — searches across all installed skills and recommends the best match for your current goal.

Audit installed skills for quality — scores trigger clarity, description completeness, and usage example coverage.

Commit staged changes, push to origin, and create a GitHub PR — all in one step with a generated commit message and PR description.