Multi-model AI orchestration where behaviour is Markdown, not code. One coordinator routes scoped task packets to 71 role-based specialists across 5 model families (Codex / Claude / Gemini / Grok / Kimi) — each working in an isolated git worktree, each reviewed by a rival family before it lands. One tmux session, no server.
This repository also configures its own agents. See what claude-vibe-squad tells them →
Skill Claude CodeCodex
Use when resuming work in a repository after a human or another process may have edited the tree—diff from the recorded anchor, inspect untracked files, treat changes as intent, and rebuild the plan and verification baseline from current contents.
Skill Claude CodeCodex
Use when diagnosing why a page or site may not be crawled, indexed, or understood by search engines: inspect robots, meta, canonical, sitemap, status, headings, internal links, and structured data; tie each recommendation to crawl, index, or intent without inventing ranking or traffic metrics.
Skill Claude CodeCodex
Use immediately after one defect is confirmed to sweep for structurally related siblings across current code, history, copies, vendored forks, services, or languages; encode a repeatable Semgrep or AST query, gate each hit independently, and propose one root-cause fix.
Skill Claude CodeCodex
Use immediately before declaring a task or session done—run the discipline checks that catch scope creep, leftover artifacts, unsolicited refactoring, and inflated prose, and block the completion claim on any failure unless the operator supplies a recorded override reason.
Skill Claude CodeCodex
Use when raw footage or a rough assembly must become a finished video deliverable—establish the story spine, refine pacing and continuity, finish dialogue, music, and SFX to a loudness target, add readable titles and a consistent grade, then verify platform-specific exports. It owns edit-to-delivery finishing, not…
Skill Claude CodeCodex
Use when analyzing why specific content spread or designing a platform-specific test of a sharing hypothesis—define cohorts, metrics, alternatives, confidence, and a stopping rule from observed data. Do not claim causality or access paid/credentialed analytics without the required gates.
Skill Claude CodeCodex
Use when a known-good render and candidate captures must be compared reproducibly: lock viewport, scale, theme, locale, and time; mask declared dynamic regions; produce tolerance-bounded pixel/perceptual deltas; and classify each delta before replacing the reference set.
Skill Claude CodeCodex
Use when a rendered UI has been built or changed and is about to be accepted — at the S4 Verify gate of a web, game, or other visual capability, before any ship step. Drives the "seen, driven, and measured" acceptance pass; the pixel-diff half belongs to visual-regression-baseline.
Skill Claude CodeCodex
Use when a human or synthesized voice read needs performance direction before or across recording or generation passes—define character or brand intent, mark emphasis, pace, pauses, tone shifts, and pronunciation, then maintain continuity across takes. This skill shapes delivery; route identity clearance for the…
Skill Claude CodeCodex
Use when a UI or media deliverable needs a criterion-by-criterion WCAG conformance decision, including keyboard, focus, accessibility-tree, contrast, reflow, motion, and screen-reader evidence. An automated score or screenshot alone cannot pass.
Skill Claude Code
Use right before anything leaves the house—a bounty submission, PR, published doc, or client deliverable—to strip internal process tells (phase names, task IDs, state//worktree paths, specialist names) and AI-cadence tells while keeping every technical claim and its evidence intact. A register change at the boundary…
Skill Claude Code
Use when sizing a single dispatch so it finishes inside one lane wall and fits the assembled-prompt ceiling—read assembledpromptbytes against assembledpromptlimit from the preflight, split work that cannot finish in one wall, and name what the lane must NOT do. scope-estimation sizes a corpus to read; this sizes the…
Skill Claude Code
Use when choosing the operator-approved mode (project versus bounty) before launch—approval of the work does not approve that workflow. Not for constructing fields or reviewer routing.
Skill Claude CodeCodex
Verify authorization, isolation, egress, credentials, persistence, evidence, and cleanup gates before provisioning or using a sandbox for PoCs, untrusted binaries, fuzzing, or other risky tests.
Skill Claude Code
Continuously execute a complex multi-phase implementation plan through configurable worker-agent loops until the plan is complete or a genuinely human-level blocker is reached. Uses fresh implementation and review contexts, reviewer-led repair, fresh re-review, durable multi-orchestrator state, and main-orchestrator…
Skill Claude CodeCodex
Use when you have a confirmed on-chain vulnerability hypothesis and must demonstrate it with a passing proof-of-concept — author an attacker contract or crafted instruction sequence for an EVM reentrancy, flash-loan, delegatecall-overwrite, or proxy-upgrade primitive (or a Solana CPI-privilege, account-substitution…
Skill Claude CodeCodex
Use when a single bug-bounty observation is too weak to pay out but several low-severity leads might compose: turn individually-inert findings into one demonstrated exploit chain whose end state proves realized impact (RCE, auth-bypass, privilege-escalation/ATO, funds theft, private-data). Not for a finding that…
Skill Claude CodeCodex
Use when auditing a Cosmos-SDK / CometBFT chain or appchain — custom modules (x/), app wiring (app/), and the ante chain. Covers the classes that halt chains or move funds, starting with non-determinism in consensus code.
Skill Claude CodeCodex
Use when auditing a cross-chain bridge, interoperability L1, cross-chain executor, or message-passing protocol (observe -> attest -> execute). Maps the trust/value pipeline and walks the bug classes that actually pay before declaring a surface clean.
Skill Claude CodeCodex
Use before spending real effort on a bug-bounty lead, and again immediately before submitting a finding, to prove the vulnerability and its specific composite chain are not already known — public advisories, prior audits, Solodit, and our own past submissions — so the squad never burns a report on a duplicate. Fires…
Skill Claude CodeCodex
Use when the audit target is a DeFi protocol — AMM, lending market, yield vault, stablecoin, or perps — and you must author the economic properties generic campaigns miss, such as token-conservation, constant-product k, share accounting, collateralization, and oracle-deviation invariants, then falsify them.
Skill Claude CodeCodex
Use when reviewing a Solana governance, multisig, or admin workflow where an unprivileged proposer may obtain signatures on an opaque durable-nonce transaction and hold it for later first submission; prove attacker-controlled exact-message approval, nonce causality, and an intrinsic unauthorized state or funds…
Skill Claude CodeCodex
Use when a contract validates a smart-contract-wallet signature by low-level-calling isValidSignature and you suspect the revert-data trick — the call's EVM success flag is discarded, so an attacker-deployed verifier that reverts with the 0x1626ba7e magic value in its reason is misread as a valid signature and…
Skill Claude CodeCodex
Use during an operator-authorized web assessment when a suspected backend template or expression sink is blind or suppresses output: force reflected evaluation exceptions or build an HTTP-status oracle to demonstrate scoped code execution or data extraction with a benign control.
At most 3 mods per repository are shown here, and a mod shipped inside a plugin is left to that plugin's page — the rest are on their repository pages: