Planning And Task Breakdown
Decompose a spec into a verifiable task list — vertical slices, explicit dependencies, per-task acceptance criteria. Use when the user says "break this down", "task list", "make a plan", "what are the steps", or invokes /plan. End with `jutsu swarm doc-review <plan.md>` to surface gaps before tasks are considered ready for implementation.
Convergence Detect
Detect when an iterative loop has stopped producing new signal. Use when another skill (polish, session-retro, plan refinement) needs a stop condition that's smarter than "ran N rounds". Three signals — output shrinking, change rate slowing, content similarity rising — indicate it's time to stop.
Incremental Implementation
Implement features as thin vertical slices behind feature flags, each shippable on green CI. Use when the user says "implement this", "ship it in stages", "vertical slice", "incremental", or invokes /implement / /build. Reject big-bang refactors; demand a slice plan first.
Dcg
Destructive Command Guard. Installs a pre-tool-use hook that blocks unrecoverable shell commands (rm -rf /, git reset --hard, git clean -fd, rm of .env / credentials, force-pushing to main, fork bombs, shred / dd, etc.) before the agent executes them. Use when the user says "install dcg", "destructive command guard", "guard rails for destructive commands", or invokes /dcg.
Editorial Review
Multi-pass editorial review for long-form essays + articles. Run after drafting is complete, before publish. Four passes — structural / line-level / AI-tells & voice / resonance. Voice-configurable (default literary-intelligent-general). Composes deslop for final user-facing prose pass. Use when the user says "review this essay", "polish this article", "edit this draft", "check this for AI-tells"…
Verification Before Completion
Hard gate before claiming work is complete. Use when about to say "done", "fixed", "passing", "ready to commit", "ready to merge" — requires running the verification command FRESH and reading actual output before any success claim. Composed by polish, incremental-implementation, and pr-review.
Blunder Hunt
Multi-pass adversarial critique primitive. Use when another skill or the user invokes /blunder-hunt or asks for a "5x critique", "find what's wrong", or "blunder hunt". Composed by skills like pr-review, polish, and unstuck to surface bugs that single-pass review misses.
Security Audit
Multi-agent security audit. Runs claude / codex / antigravity in parallel through three CVSS-aligned lenses (claude=auth+data flow, codex=injection+priv-esc, antigravity=dep+supply-chain). Wraps `jutsu swarm security-audit --pr <n>` (diff mode) or `jutsu swarm security-audit <path>...` (files mode). Output is a threat model with attack vectors + prioritized mitigations. --full mode defaults ON be…
Spec Driven Development
Author a PRD-style spec before writing implementation. Use when the user says "write a spec", "PRD", "design doc", "spec out", "let's plan this", or invokes /spec. Produce a written, reviewable artifact (problem, scope, non-goals, acceptance criteria, open questions). End with `jutsu swarm doc-review <spec.md>` to surface gaps before declaring done.
Session Retro
End-of-session retrospective that mines the current conversation for learnings. Use when the user says "retro", "session retro", "what did we learn", "extract learnings", "session review", "conversation review", or wants to capture insights before ending a session. Also trigger when the user asks to turn repeated patterns into skills or memory, or says "before we wrap up".
Journal
Monthly development journal. Use when the user says "journal", "monthly journal", "what did we do this month", "monthly recap", "development diary", or invokes /journal. Optionally /journal {month} for a specific month (e.g., /journal march, /journal 2026-03).
Unstuck
Guided problem articulation when you're stuck. Use when the user says "unstuck", "I'm stuck", "this isn't working", "help me think through this", "I keep going in circles", or invokes /unstuck. Helps break out of unproductive loops by forcing clarity before the next attempt.
Deslop
Strip AI-generated tells from a draft without changing its substance. Use when the user says "deslop", "remove AI tells", "make this sound less like ChatGPT", or invokes /deslop. Composed by polish, journal, readme-update, and any skill that produces user-facing prose.
Brainstorm
Multi-agent ideation against a free-form prompt. Wraps `jutsu swarm brainstorm "<prompt>"` to run claude / codex / antigravity in parallel with three different angles (claude=long-horizon framing, codex=code-pattern grounding, antigravity=cross-domain analogy), synthesizes into a ranked list of approaches with tradeoffs. Use when the user says "brainstorm", "ideas for", "how should we approach",…
Dream
Pre-implementation interrogation. Walks any topic through 8 cognitive lenses to surface hidden assumptions, weak premises, orthogonal angles, temporal decay, and adversarial misuse BEFORE the topic becomes a spec / plan / commit. Use when the user says "what do you think about X?", "should we do X?", "is X worth pursuing?", "let's brainstorm" (note: this is NOT brainstorm — see "vs brainstorm pre…
Dispatch Parallel
Fan out N independent tasks to parallel subagents, collect their results, optionally synthesize. Use when another skill needs work that decomposes into independent pieces — review passes with different lenses, multi-file searches, multi-source data gathering. Composed by skills like blunder-hunt, pr-review, polish, journal.
Code Simplification
Reduce code without changing behavior. Apply Chesterton's Fence (don't remove what you don't understand) and the Rule of 500 (files over 500 lines usually need splitting). Use when the user says "simplify this", "clean up", "remove dead code", "tighten", or invokes /simplify. Composes polish (review-fix loop) and blunder-hunt (regression risk).
Polish
Automated post-implementation review-and-fix loop. Use after completing a feature or implementation plan to find and fix issues across multiple passes until convergence. Invoke with /polish. Composes blunder-hunt for the review pass, convergence-detect for the stop condition, and deslop for any user-facing prose at the end.
Pr Review
Adversarial multi-agent pull-request review. Runs claude / codex / antigravity in parallel via `jutsu swarm pr-review`, each with a tailored prompt (claude=architecture+correctness, codex=edge cases, antigravity=cross-file patterns), then synthesizes into a single markdown review with a disagreement table. --full mode adds a Pass-2 round-robin debate where agents critique each other. --strict add…
Refactor Plan
Multi-agent refactor planning — files + goal in, ordered step plan with risk per step out. Wraps `jutsu swarm refactor-plan <path>... --goal "<goal>"` to run claude / codex / antigravity with three angles (claude=architectural decomposition, codex=stepwise risk, antigravity=pattern consistency). Designed to feed into `planning-and-task-breakdown` for execution. Use when the user says "plan this r…
Readme Update
Detect README staleness and propose a targeted update. Use when the user says "update readme", "is the readme out of date", "readme refresh", "/readme-update", or after merging a feature that likely changes installation, usage, or public API. Reads the README's last-touched commit, walks the commits since, and proposes a section-by-section diff.
Scope Check
Mid-task scope mirror. Use when the user says "scope check", "am I scope creeping", "how big has this gotten", "check the scope", or asks about the size of current work. Also use proactively when you notice the work has grown significantly beyond the original ask.
Security And Hardening
OWASP-Top-10-aware security review of a diff or target file set. Use when the user says "security review", "OWASP check", "audit for vulns", "harden this", "security audit", or invokes /security. Composes blunder-hunt's hostile-input lens. Surfaces injection, auth gaps, broken access control, sensitive data exposure, and supply-chain risks.
Doc Review
Universal multi-agent QA gate for markdown artifacts (specs, plans, decision records, RFCs, design docs). Wraps `jutsu swarm doc-review <path>` to run claude / codex / antigravity in parallel with tailored lenses (claude=completeness, codex=implementability, antigravity=consistency), synthesizes into a single review with a disagreement table. Designed to be the final-review pass for any artifact-…
Autopilot
Intent-to-PR pipeline. Describe an outcome in natural language and the system runs the full development pipeline autonomously — brainstorm, spec, plan, build, ship — with multi-agent adversarial review at every artifact stage. Two gates total (post-brainstorm + PR review on GitHub). Cost-capped. Use when the user says "autopilot", "build this end to end", "take this from idea to PR", or invokes /…