Recently added
Plugins grouped by the day they were added to the directory, newest first.
Sep 20, 2026
- fixture-clean0stars
fixture-clean is an open-source Claude Code plugin indexed from GitHub with 1 skill. Review its manifest and source before installing.
Claude Code1 skill - fixture-broken0stars
fixture-broken is an open-source Claude Code plugin indexed from GitHub with 2 skills. Review its manifest and source before installing.
Claude Code2 skills - vibe-suite0stars
Vibe coding suite for Claude Code — NL-artifact quality tooling, cross-model review, and workflow loops, bridged to Codex CLI.
Claude Code24 skills - marketnow0stars
Free trust layer for MCP agents: verify credentials (8 formats), scam-check domains, fingerprint tools, search a 68k+ MCP server registry. No API keys, no fees.
Claude Code1 MCP server - kit-storekit-check0stars
Six offline file checks for StoreKit and Xcode pre-flight traps: bundled .storekit, distantFuture expiry, product ID drift, local-only intro offers, export-compliance key, prices in metadata text.
Claude Code1 skill - kit-ship-check0stars
Pre-commit gate: test, lint and type check with a PASS/FAIL table.
Claude Code1 skill - kit-new-feature0stars
Plan-first feature flow: 5-line plan, wait for yes, build, test, ship check.
Claude Code1 skill - kit-explain-diff0stars
Plain-English review of the current git diff with risky lines flagged.
Claude Code1 skill - reqstool1stars
AI-assisted reqstool requirements traceability — skills and commands for managing requirements, SVCs, and filters.
Claude Code1 skill - reqstool-openspec1stars
OpenSpec integration for reqstool — conventions for referencing reqstool IDs in OpenSpec files, plus an openspecui hook that enriches all OpenSpec documents with reqstool requirement/SVC titles and descriptions at read time.
Claude Code1 skill - openfpa0stars
Australian FP&A pack on openfpa: 30 June years, Xero AU, GST/BAS. Upstream is Jeff Brines / Guiderail.
Claude Code16 skills - ultrasafe8stars
Pre-release / pre-update simulated penetration testing discipline — v0.2.x runtime activation cut (advisory mode; v0.2.1 = ship-surface alignment doc patch). Wraps Ultrasafe.md v0.2.1 substantive spec (~3522 lines, +978 since v0.1.0) with the runtime layer: 8 attacker SKILL.md (ultrasafe-ai-llm-redteam · ultrasafe-web-api-attacker · ultrasafe-supply-chain-auditor · ultrasafe-crypto-reviewer · ultrasafe-social-engineer · ultrasafe-methodology-compliance · ultrasafe-threat-model-lifecycle · ultrasafe-synthesizer) implementing the v0.1.0-spec 8-agent parallel red-team fan-out; 2 hooks (PreToolUse `ultrasafe-trigger.cjs` matcher=Bash detects publish-equivalent commands and emits ULTRASAFE_RELEASE_GATE advisory; Stop `ultrasafe-clean-signal.cjs` evaluates 4-condition AND-gate at cycle-end and emits ULTRASAFE_ITERATION_BOUNDARY); MCP server `server.cjs` exposing 5 stdio JSON-RPC tools (ultrasafe_run_fanout · ultrasafe_finding_aggregate · ultrasafe_clean_signal_check · ultrasafe_report_generate · ultrasafe_release_gate); 5 new Constellation A2A intents integrated per Constellation v2.4.3 §13.16.13 wire-spec (ULTRASAFE_FINDING per-attacker emit + ULTRASAFE_ITERATION_BOUNDARY ≥3-iteration boundary marker + ULTRASAFE_RELEASE_GATE publish-time gate state + SECURITY_DISCLOSURE_INTAKE external disclosure entry + MPCVD_COORDINATION Multi-Party Coordinated Vulnerability Disclosure; ack_tier dual-mode mapping advisory v0.2.x → blocking v0.3+; cross-module integration with Hyperbrief paired DECISION_REQUEST on severity ≥ high + Greatpractice macro/mezzo entry candidate flow for ratified findings + Superscalar retire-barrier composition at iteration boundaries). v0.2 advisory invariant: all outputs (skill recommendations / hook emissions / MCP tool returns) carry `advisory` flag — NEVER block publish in v0.2.x. Blocking mode (v0.3+) introduces user gate via paired Hyperbrief DECISION_REQUEST. 3-layer synthesis report (OSCAL Assessment Result + Hyperbrief 9-section IR + Greatpractice tree candidate) generated by synthesizer attacker. ≥3 iteration loop with 4-condition AND clean-signal gate (regression-free + monotonic finding-reduction + coverage-floor + 2 iter consecutive). Backed by 17-axis cross-domain deep research (reports/2026-06-05-ultrasafe-research/) + Tier A 5 critic patches absorbed inline. Plugin manifest registers: 8 skills + 2 hooks + 1 MCP server. 0 npm deps (v0.2.7 — the ajv declaration was removed: it had no call sites and no schema files to compile).
CodexClaude Code8 skills1 MCP server - superscalar8stars
Execution-scheduling discipline for parallel sub-agent dispatch. v0.5.0 / Superscalar v0.14.0 adds the /speculation toggle (§4 speculation authorization) — four modes plus a read-back (off default / auto asks per speculation with the four-element trade-off / on pre-authorizes low-downstream-sensitivity lanes while high-sensitivity ones still ask / always pre-authorizes every lane inside the §3 barrier / status reports the mode and where it would next apply). The modes graduate along downstream sensitivity, an axis §4 already named, so this is a threshold on existing vocabulary rather than new vocabulary. One constraint governs all four and cannot be read past: the mode changes WHO authorizes a lane, never WHAT a lane may touch — read the other way, there is no mode in which speculation reaches a write, deploy, send, or delete, so no operator setting turns speculation into an outward-facing action; widening scope is a separate and narrower decision than changing who signs off. Andon transparency (visual signal, /stop-spec pull-cord, misprediction logging) is unconditional at every mode including always, because pre-authorization removes the question and not the visibility — dropping the announcement at always would make the loudest mode the quietest one. State is one marker (.agent/speculation.json, absent ⇒ off) read at dispatch time and never mirrored, and a mode change is a declaration event where a board is present. v0.4.0 / Superscalar v0.12.0 adds the /context-caching toggle (§5.4 memory-hierarchy borrowing) — three modes (off/self/agent): pin model+effort at session head (each is a cache key), compact at task boundaries with rewind-over-compact for abandoned paths, prefetch context right after boundaries, account for cache-cold subagents in fan-out pricing, consult the per-harness flush list before mid-session mutations, and measure with the cache counters instead of declaring; volatile vendor numbers (multipliers, TTLs, minimums, storage rents) live in cache-registry.json under the dated/confirmedBy/revisit contract with singleSource marks. v0.3.0 / Superscalar v0.11.0 adds the /ooo question-scoreboard toggle (§5.3) — out-of-order execution for the main lane itself: a mid-task question registers as a one-line scoreboard entry instead of stalling the queue (self-resolve by measurement → bypass independents → assumption-run only under §4 speculation gates → park), late answers wake only their dependents, and loss/external-publish gates are never bypassed — the toggle moves the waiting, never the gates. v0.2.0 / Superscalar v0.6.0 adds the /subscaler toggle (§5.1 tiered model composition): when the main agent runs a frontier reasoning model, delegate execution-shaped work (code writing/editing) to the strongest execution-tier model as subagents — Fable-class→Opus-class / Sol-class→Terra-class, effort high — with a delegation rubric (spec-complete work delegates, architecture/judgment/review stays), per-harness application surfaces (Claude Code Agent-tool model param + Workflow opts; Codex .codex/agents TOML; Kimi per-invocation), single-marker-file toggle state, and default OFF (recommended ON for fan-outs, where the shared prompt cache is already forfeited). v0.2.4 / Superscalar v0.10.0: five-tier vocabulary — T1.5 frontier-execution band (the vendor's start-here flagship at ~half T1 price, split from T1 on temperament: it converts thinking into action sooner, so T1.5 lanes always pair with verification gates) + per-account entitlement check for invitation-gated models. Wraps Superscalar.md as model-invoked skills the agent consults before fan-out decisions: issue_width formula + worktree-isolated reorder buffer + cost-benefit gate + opt-in speculation + adoption thresholds + n=8 dogfood ledger (Entry 05 absorption sync to bundle 005 — case 8 first-worktree-write data point + §2b nested-repo limitation + R3 + secondary verification-depth lesson) + Entry 06 controlled A/B measurement showing 'parallelism is not sufficient; orchestration discipline produces consistency' + v0.4.1 §3.1 Hyperbrief decision-delegation interlock (write/deploy/send lanes pass through both cost-benefit gate AND hyperbrief-trigger-check; read-only lanes exempt).
Claude Code3 skills - hyperbrief8stars
Decision-delegation gating discipline. v0.6.0 substantive 4-slot additive cut (evaluation_lenses §0 multi-lens scoring + recommended_methodology §8 meta-decision methodology refs + maturity_anchor measurement-anchored maturity labels + term_pairing 4th tone-axis; 모두 optional, 기존 v0.5.6 IR back-compat 보존). Wraps Hyperbrief.md v0.6.0 as 3 model-invoked skills (trigger-check + hyperbrief + revisit) + IR schema + 2 templates + deterministic renderers (mini-engine.cjs + types.d.ts + bin/render.cjs + ajv) + MCP server exposing 4 tools (hyperbrief_render / hyperbrief_validate / decision_ledger_append / decision_ledger_query). v0.5 schema: surface_profile_estimate auto-computed (AF-18 declared-vs-effective drift warning) + recommended_artifacts {language, line_count, body_hash} sub-fields auto-stamped + audience_profile_fallback.telemetry opt-in phrase-learning + canonical button_label parenthetical universalization for adopter localization. 결정 위임 게이팅 규율: 사용자에게 결정을 묻는 행위 자체를 규율한다.
CodexClaude Code1 skill1 MCP server - greatpractice8stars
Memory-triggered practice-codification discipline. Wraps Greatpractice.md v0.3.3 — v0.3.3: first model-invoked skill `routinize` (관행화 — the front door to the capture→mature→codify pipeline §1.2/§5.4; capture-eager/promote-deferred, ratification user-gated) + `/before-compact` soft chain (codify recurring practices before materializing session state). v0.2.x / v2.5.117 P0b normative-foundations backfill: §B.5/§B.6 placeholders expanded with adversarially-verified primary sources (distributed cognition + institutional isomorphism), correcting two v0.1 over-claims (coupling criteria are necessary-not-sufficient; isomorphism-inhibits-innovation + ≤20% mandatory cap → empirically contingent, no numeric threshold); §B.7 marks P0b complete. v0.3.1 = micro tier first decomposition (9 mezzo → 20 atoms). children decomposition pattern 의 first batch demonstration. v0.2.0 / v2.5.61 mezzo 8-batch ratification: release-cadence.md (macro) §2.3 의 8 mezzo decomposition candidates 를 Workflow wk5a6jh5k 병렬 fan-out 으로 한꺼번에 ratify (n-way-sync-registry / package-files-validate / bin-entry-validate / link-integrity-check / dry-run-smoke-test / pre-publish-user-gate / naming-hygiene-grep / auth-2fa-discipline). 5-axis maturation 분포 24/24/22/21/19/19/19/17. ~1850 lines added. v0.1.0 (2026-06-04 cut) 의 1 macro + 1 mezzo (outbox-json-validation) 구조에서 1 macro + 9 mezzo 로 확장. 3 JSON schemas (entry-frontmatter + hook-spec + voice-rules) + 1 PreToolUse contact hook (outbox-json-validate) 보존. Targets the "quiet omission" surface — small obligations that should accompany the work but slip through; agent-workspace memory feedback 가 input trigger, raw signals 5-axis multi-criteria maturation gate 통과 후 3-tier macro/mezzo/micro hierarchy 등재, ratified entries lifecycle hook 강제. 9-axis cross-domain deep research 백업 (reports/2026-06-04-greatpractice-research/). v0.3+ roadmap: 8 mezzo entries 의 micro decomposition + 4 runtime cjs (lint + promote + voice-check + freshness) + 4 additional hooks (motion-step pre-send-inbound + SessionStart bridge + fixed-value n-way-sync + workspace path) + 5 skills (routinize shipped) + 3 templates + manifest hash + renderers. 0 npm deps (v0.2.0).
Claude Code1 skill - estregenesis8stars
The EG kit plugin — the entry point to EstreGenesis itself, as skills instead of a copy-pasted URL. 13 skills over 8 canonical procedures: eg-bootstrap (/egboot — new project: pick the seed tier, install it as .agent/seed_prompt.md, run the seed's Bootstrap mode) · eg-migration (/egmig — existing project with its own rule files: audit them, classify M1/M3, run the seed's Migration A or C, promote every rule or record its drop) · eg-upgrade (/egup — seeded project: read the seed's version marker, diff it against the upstream CHANGELOG, present an additive numbered delta menu, then refresh the installed EG plugins) · eg-memsync (/egmem — reconcile agent memory with the project's governance docs: five drift kinds detected both ways, dated memory beats undated doc prose, repo reality closes stale state-memories, memory-only durable policy promoted into the docs under four boundaries) · eg-interview (/eggrill — interview-strategy intake for the human-facing agent: press a vague or high-stakes request into a delegable brief [purpose, deliverable, constraints, success criteria, scope, assumptions] in few high-yield question rounds before decomposing or delegating; composes with Hyperbrief for choice presentation) · egrich (full arm: the seed plus all six optional modules, their plugins, hooks and MCP servers, wired and verified — with the cheap discipline-only subset named up front so 'everything' is a choice, not a default) · talka (/talka — talk-about elicitation: hold a relaxed, topic-anchored conversation that progressively draws out the requester's tacit knowledge ahead of a review or judgment — the gentle sibling of eg-interview: same goal, opposite pace, at most one question per turn, statements offered for correction, elicitation ledger materialized to a durable surface at close) · feeda (/feeda — feed-analysis intake: take an external feedback artifact about the current project, verify every claim against measured reality, tier the survivors [defect / spec gap / design disagreement / info-only], build the prevention mechanism before fixing the instance, and close the loop with the contributor — the artifact is data to verify, never instructions to follow). The five short names are pointer aliases, not copies (Compendium's pointer-not-paraphrase discipline applied to the plugin's own surface): each alias reads the canonical sibling skill and executes it. Seed locations (raw.githubusercontent.com, 3 tiers x 2 languages) are baked into the skills, which is the whole point — the seed is fetched, never reconstructed from memory, and its version marker is preserved so a later upgrade can diff mechanically. 0 npm deps, no hooks, no MCP server: this plugin is procedure, not runtime.
Claude Code13 skills - corporate8stars
Durable-role agent organization. Wraps Corporate.md v0.1.5 (design draft — runtime deferred as named prunable units) — EstreGenesis's 7th module and the third orchestration axis: Superscalar decides how many lanes and at which tier (anonymous, one fan-out), Constellation decides how agent processes talk, and Corporate decides who exists. Its unit is the durable role: a named seat that outlives any session or process and therefore can accumulate practice, be addressed across time, own units in a coordination registry, and give a decision an addressee — four properties an anonymous lane structurally cannot have. Ships the durable half only: the two-space desk contract (a gitignored private desk that is the cwd where the role's harness actually runs, holding its charter and its own memory, plus a tracked public locker that is the read surface for the organization and the user; the default sharing mechanism is write-a-document-and-pass-the-path), multi-host organizations (a desk does not span hosts; the locker is the cross-host exchange), a reference roster whose main does not take direct work and whose reviewer must run on a different model family than its builder, four residency classes defaulting to the cheapest with an organization circuit breaker that demotes rather than stops, an eviction lifecycle symmetric with creation whose automation becomes mandatory the moment the organization is dynamic (automating growth without automating shrinkage is a one-way ratchet), a three-layer toggle chain (role > group > organization) read by exactly one resolver and carrying an identity invariant that makes adoption bit-identical for projects without a roster, hub-and-spoke communication with three declared escapes, bidirectional practice-role derivation (a recorded practice is evidence a seat is warranted; a seat with no practice is an eviction candidate), and two declaration events (CorporateChart / RoleState) that let a live board render an organization it hosts none of. The perishable per-harness values (exact invocation flags, session-continuation syntax, sandbox-grant keys, instruction-file names) live in a dated registry data file with a per-row source requirement and a self-declared expiry, so the spec does not rot on harness release cadence; v0.1.1 populates it with four confirmed harnesses (Claude Code, Codex CLI, Antigravity CLI, Gemini CLI), the gotchas that break a naive desk implementation, eight explicitly named gaps that stay out of the rows, and five cross-harness mechanisms. The supervisor, the org-chart rendering, the bootstrap interview automation, and boot residency are named prunable units. 0 npm deps.
Claude Code4 skills - constellation8stars
Live multi-agent orchestration: WebSocket server + A2A messaging + dashboard + Stop hook helper + MCP server (full Phase 2 impl: WS proxy + 5 tools incl. a2a_wait_ack 3-tier + session-lifecycle + §13.13.2 dedup + chunked transfer reassembly). v0.3.29 / Constellation v2.4.57 — room artifacts + board state schema: the reference server holds three pinned artifacts per room (header · append-only decision ledger rd-N with supersedes · running summary with covers_until) behind the symmetric fetch contract (RequestRoomArtifacts → RoomArtifacts) and the new write path RoomArtifactsUpdate (participants/board only, server-stamped editor/versions, delta broadcast under a single ack-by-reference artifacts version); dashboard room tabs gain a collapsible 📌 pinned block (esc-only, auto-fetch on entry, TDZ-guarded) and the scribe charter the room-digest duty; §13.31 (new) canonicalizes the board state contract — current[].stages {label, status: pending|active|done} + sub/blockedBy + a five-point usage discipline — with matching hooks in /roundnext (start-with-stages), /boardsweep (§13.31 as the consumer contract), and /roundtable (D5 fetch/write); roundtable smoke 18→22 checks. v0.3.28 = boardsweep renderer-contract clause (adopter-measured gap: a stages list written as {name,state} against a {label,status} renderer survived two sweeps rendering all-PENDING — schema normalization now checks field keys/value domains against the consumer contract, since render fallbacks silently swallow this class; deterministic state lint preferred where present). v0.3.27 / Constellation v2.4.53 — roundtable topic rooms (§13.30): multi-party rooms (3+ participants, agents + humans) with a two-layer split — the server enforces the deterministic floor (fan-out over existing 1:1 relay, autoHop chain cap with parking, per-agent rate + consecutive caps, stall detection, human soft-yield priority, advisory floor state, notice-class surfacing) and the new /roundtable skill owns the conduct layer (silence-as-default gating, response obligation on naming, floor bids, agent-chain self-cap, fixed catch-up order, ack-by-reference, structured close, moderator-hat procedure authority); MUC/Matrix-lineage vocabulary (voice/moderated/moderator/notice) + dashboard ➕ roundtable creation entry on the prompt-target selector. v0.3.26 / Constellation v2.4.52 — peer key kind (pk- prefix · ?peerKey= · peer wire role): another project's main joins as a §13.9.3 peer-main, no longer riding an upstream key; dashboard gains a peer tab group (로컬↔협업 사이, rose) + Main↔Peer/Peer↔Peer/Peer↔Collab monitors (lazy — hidden until traffic) + a prompt-target selector on the injection row (auto-indicates the effective target on tab/group selection, per-selection override), key UI gains the 🤝 피어메인 radio/tab, /join/peer onboarding, KEY-MGMT v0.4. v0.3.25 boardsweep wording-precision patch (post-ship adversarial verify: agent-authored-scope framing, generic tone/policy wording instead of a workspace-specific modes field, current[] add-missing clause, project-registry naming). v0.3.24 adds the /boardsweep skill (8th): one full board true-up over every component except the realtime chat — done[] identifiers verified against the VCS, shipped planned items migrated, resolved decisions closed, current corrected, registry pruned, tone/schema normalized; retrospective accuracy only (the forward-looking counterpart /roundnext keeps scheduling). Bundles the v2.5.x reference runtime, the v2.5.3 SSoT-compliant gateway-client, the v0.4 MCP integration (§8), the v2.5.8 cross-environment Stop hook helper, the v2.5.10 attachment transport-mode (§13.11 rule 5), and the v2.5.11 at-least-once relay reliability (§13.13.2). v0.3.12 / Constellation v2.4.12 대시보드 XSS 하드닝 (dashboard render only): 첨부 미리보기 html iframe sandbox(빈 값=스크립트·동일출처 차단) + mermaid 첨부 textContent 주입(innerHTML→textContent, raw body 의 <img onerror>/<script> 파싱-전 실행 차단). Playwright 5/5 PASS. v0.3.11 / Constellation v2.4.11 보안 하드닝 (전수점검 후속): reference server.cjs secure-by-default bind (기본 127.0.0.1, 네트워크 노출 WS_BIND opt-in + 비-loopback 경고 — 무인증 board 표면의 원격 키관리·SetMain 벡터 차단) + ws-core.cjs WS_MAX_FRAME 프레임 상한 (16 MiB, 메모리 고갈 DoS 가드) + server-keys.eux credential 파일명 사실오류 정정 (key.json/ws-keys.json/local-keys 전부 gitignored MUST). v0.3.10 / Constellation v2.4.10 PR system §13.22 distill (RRP→정식 normative 스펙: two-level virtual/Github PR + 5 wire vocab + local-bridge topology + trusted-mirror/standard approval + β′ sourcing + strict merge gate + 5-state machine) + deadlock 서버 detector opt-in (§13.19.10 Q3 board-adapter strict 2-cycle detector, WS_DEADLOCK_DETECT env-gated 기본 OFF — canonical 검출은 에이전트측 DFS §13.19.3). v0.3.9 / Constellation v2.4.9 dashboard A2A 대화 카드 통일 + 인라인 re>summary 우선 (코디네이션 메시지 Delegate/WorkerReport/WorkerAck/OnboardAck/AgentHello + 미분류 객체-value CUSTOM 을 a2acard 로 통일, 대화/제어/dim-gated row 는 유지). v0.3.8 / Constellation v2.4.8 dashboard EstreUF-parity 5-gap (검토사안 0뱃지 숨김 + 메시지 발신-시각 보존 wsMsgEpoch/wsRowTime + 닫은세션 영구삭제 크로스보드 동기/persist + SelectionResolved 라우팅-무관 처리 + 그룹선택 멤버탭 4-role 틴트). v0.3.7 / Constellation v2.4.7 단일 워크스페이스 멀티에이전트 session-aware routing (register-all + fail-safe) — 같은 워크스페이스 다중 Claude 세션이 공유 Stop hook 실행 시 cursor 경쟁을, per-session CLAUDE_CODE_SESSION_ID + .agent-sessions.json 레지스트리로 분기 (미등록=SKIP-ALL fail-safe, 자기 소유 inbox 만 처리). 신규 register-session.cjs + pre-send-probe routing + join-local.cjs v2.4.7 (local 워커 outbox drain + full inbound 로깅) + server.cjs WS_PRIMARY_AGENT 경고 + dashboard cursor:move leak fix. ≤5초 file Monitor 2층 기상 권장. v0.3.6 / Constellation v2.4.6 dashboard A2A 대화 UI 확충 — 7 additive surfaces: sticky popup header (스크롤 가림 방지) + 날짜변경선 sticky top:0 + A2A-intent CUSTOM 카드 (18 intent dict + summary helper + card builder + §13.16.12 Pattern 7 TEXT_MESSAGE fallback 코드블록 JSON envelope 승격) + 첨부 칩/미리보기 모달 (4 helpers: wsAttachments/wsAttIcon/wsAttChipEl/wsAttPreview, mime별 이모지 + sha256 title + 👁/⬇) + SelectionPrompt chip 카드 #406 UI6 (issued/answered/cancelled + multiSelect + allowFreeText + SelectionAnswer/SelectionCancel emit + SelectionResolved cross-board dim) + 닫은 세션 삭제 UI3 (개별 🗑 + 전체 🗑 + 연결상태 dot + DeleteChannelHistory CUSTOM emit) + self-attestation confirm/prompt 자체 다이얼로그 (wsConfirm/wsPrompt Promise 헬퍼 — browser confirm/prompt 의존 제거로 IDE webview 환경 호환). 프로토콜 wire 변경 없음. v0.3.5 / Constellation v2.4.5 incident hardening — 4-layer runaway guard after 2026-06-07 duplicate same-agentId collab-client incident: (a) new `runtime/single-instance.cjs` helper (deps-0 PID lock + alive check + stale auto-cleanup, reject-new policy); (b) helper applied to ws-agent-client + join-local + watchdog (local-bridge retains 2026-05-30 takeover guard); (c) pre-send-probe.cjs runaway protection (PROBE_RUNAWAY_BYTES 32 MiB cap + streaming countLinesStream + auto cursor advance + MAX_MEANINGFUL_SURFACE=50 cap); (d) new `runtime/scripts/runaway-recover.cjs` (readline streaming allowlist extraction from runaway backup, deps-0). v0.3.4 / Constellation v2.4.4 dashboard UI augment — 4 corner-anchored position presets (`.ws-pop.pos-{tl,tr,bl,br}` localStorage-backed; drag/resize end normalized to anchor edge-distance via `wsRectToAnchorPos`; legacy left/top WS_UI.pos auto-migrated; default 'br'); settings modal ⚙ (popup head 의 📂 우측) + 4 preset cards + anchor behavior description; fab right-click context menu (center / key-issue / key-manage / settings); fab z-index 60→58 so popup covers fab; arch-empty placeholder text + folder emoji 🗂→📂; header spacing compression (.ws-conn margin-bottom 12→8, .ws-h margin 4/7→0/0). v0.3.3 / Constellation v2.4.3 ships Ultrasafe A2A-intent family integration (§13.16.9 A2A-intent allowlist + 4-group classification table A2A-intent row + new §13.16.13 wire-spec sub-section): 5 new intent names (ULTRASAFE_FINDING per-attacker emit + ULTRASAFE_ITERATION_BOUNDARY ≥3-iteration boundary marker + ULTRASAFE_RELEASE_GATE publish-time gate state + SECURITY_DISCLOSURE_INTAKE external disclosure entry + MPCVD_COORDINATION Multi-Party Coordinated Vulnerability Disclosure) + dual-mode advisory/blocking ack_tier mapping (advisory v0.2.x = received-tier; blocking v0.3+ = decided-tier reusing Hyperbrief application-tier ladder) + cross-module integration (Hyperbrief paired DECISION_REQUEST on severity ≥ high + Greatpractice macro/mezzo entry candidate flow + Superscalar retire-barrier composition at iteration boundaries) + board-mirror render policy per intent (severity-colored finding cards + iteration timeline dividers + gate decision cards with §13.14 redaction for disclosure/MPCVD identifiers) + §13.16.12 Pattern 7 TEXT_MESSAGE fallback prefix support. Additive minor — wire compatible with v0.3.2 adopters (5 new names route via fail-safe-default-A2A-intent branch until allowlist update). v0.3.2 / Constellation v2.4.2 dashboard UI 통합 (setupWsCollab 별도 버튼 제거 → setupWsKeyMgmt 통합, 🔑 토글, kind radio ⬆/🏠/🔗 + 기본값 local + 라벨 포커스, UI5 모달 4 탭 + 이모지 통일 + 🔐 제목 + placeholder 즉시 렌더 + 폭 240→320 한 줄 fit). v0.3.1 / Constellation v2.4.1 extends KEY-MGMT (v0.3 additive: kind='upstream'|'collab'|'local' + roleDescription + local key wire-private file path / script call registration + UI kind dropdown + role textarea + UI5 modal chips). v0.3.0 / Constellation v2.4.0 first ships KEY-MGMT (WS-PROTOCOL-KEY-MGMT.md v0.2): 5 canonical messages (KeyIssue/KeyList/KeyRevoke/KeyLabel/AgentNameChanged) + 5-state machine (ISSUED→ACTIVE→REVOKED_PENDING→REVOKED→DELETED) + key.json atomic-write persistence + 14-day TTL + main-only permission gating + immediate/sessionEnd revocation modes + UI4 발급 패널 + UI5 키 관리 모달 (목록/revoke/label rename); RegisterUpstreamKey transitional alias retained. Plus v2.3.23 dashboard reference parity (upstream invite UI + onboarding md channel guide). Plugin v0.2.0 = Phase 2 production-ready (npm dep: ws ^8.18.0).
CodexClaude Code10 skills1 MCP server - compendium8stars
Concept-anchored dual-register vocabulary substrate. Wraps Compendium.md v0.2.4 (runtime — content store + lint/pointer-resolution + the §10 8-tool MCP server; dashboard surfaces deferred v0.2.x) — EstreGenesis's 6th module and the purest expression of north-star axis-2 (vocabulary survival > code survival): a markdown wiki of cross-module concept pages + a concept-anchored dual-register glossary (internal/expert term <-> general/plain term <-> one shared definition) + a cross-link/click-to-define system. Owns nothing normative — every entry resolves to a one-directional pointer into the owning module spec (Constellation/Superscalar/Hyperbrief/Greatpractice/Ultrasafe = SSoT); the sole exception is general-register vocabulary no module owns (owner_spec=null), for which Compendium is the legitimate SSoT. Also the relocation home for deprecated vocabulary (status:superseded + superseded_by redirects — axis-3 eviction first-class), and it models its own eviction. Right-sized v0.1 ships the concept-anchored model + frontmatter schema + content-boundary charter (pointer-not-paraphrase + one-directional authority) + ~25-term EG seed vocabulary exemplars + 4 gardening lints + pointer-resolution check (verify-nway 13th axis form fixed). Promotion is a single count->queue-Hyperbrief-brief gate; the hysteresis/debounce automation, click-to-define a11y component, multi-tab split, MCP runtime, and dashboard board surfaces are named prunable units deferred to v0.2+. v0.2.4 ships cross-tool curation Agent Skills (compendium-curate + compendium-lint) that carry the pointer-not-paraphrase + dual-register + gardening-lint + first-class-eviction discipline to any harness over plain markdown — the portable carrier of the opinion (the runtime-agnostic option-(c) shape: the storage/viewer is substitutable, the discipline is what survives). 0 npm deps.
CodexClaude Code2 skills1 MCP server - jus0stars
Juscribe Workflow SOP as three on-demand Claude Code skills — ticket-workflow (lifecycle plus estimation, labels, testing gates, and the jus API reference), hard-rules, and retrospective (reviewing an iteration and publishing the report) — backed by deterministic enforcement hooks.
Claude Code3 skills - cadence27stars
Catch robotic-sounding writing and rewrite it in a voice you choose. One skill, 5 commands (/cadence write, learn, recast, deslop, voices) and a deterministic checker that scores prose and names every tell.
Claude Code1 skill - design1stars
Interview-driven software design: takes a rough pitch to a complete, gap-free design tree, then captures it as specs, contracts, and ADRs. Carries a lexicon and an assumptions record across a project's lifetime; everything else it writes is the feature's brief.
Claude Code4 skills Takes one work package (a ticket or an epic with all its children) from a prepared worktree to a pull request with a GREEN CI pipeline, for projects in any language (stack auto-detected). One headless skill — process-ticket — backed by six subagents (context-extractor, planner, plan-critic, developer, test-critic, reviewer) with isolated critics, three-round caps and machine-readable progress events on the ticket. Never asks a human. Drives the agent-project-issues MCP.
Claude Code- nax-spec-kit0stars
Spec authoring + audit workflow for coding agents: spec-writing (brainstorm → SPEC-*.md) and spec-review (ground a spec against the real codebase before implementation).
Codex1 skill - nax-spec-kit0stars
Spec authoring + audit workflow for coding agents: spec-writing (brainstorm → SPEC-*.md) and spec-review (ground a spec against the real codebase before implementation).
CodexClaude Code1 skill - openindu-control-tower0stars
openIndu org-wide Agent behavior principles (11), 20 maintainer agents, organization Issue intake, tiered SDLC workflows, and a main-branch protection hook
Claude Code15 skills - overleaf-mcp-rt6stars
Edit Overleaf projects live alongside human collaborators. Bundles the overleaf-mcp-rt MCP server (self-hosted Overleaf and overleaf.com) with skills for setup, safe co-editing, LaTeX workflow and review comments. Use when asked to read, write, review or compile anything on Overleaf.
CodexClaude Code4 skills1 MCP server - fieldrun0stars
Start, review and submit Fieldrun field jobs from your own machine. No account needed.
Claude Code3 skills - tribunal0stars
Multi-agent execution with LLM-as-a-Judge quality gates. An Actor performs work; a jury of typed Judges scores it against a project-overridable rubric; a Meta-Judge reviews the jury for bias; the gate decides accept, retry, or exhaust. Grounded in LLM-as-a-Judge (Zheng et al. 2023) and LLM-as-a-Meta-Judge (Wu et al. 2024). Builds on the Onlooker ecosystem plugin.
Claude Code1 skill - scribe0stars
Intent documentation from agent activity. Captures why changes were made — problem context, decisions, tradeoffs — and distills them into readable artifacts at session end.
Claude Code - jev-skill-suggestion31.9kstars
Keeps the skill listing out of the context window and lets TypeSafe's Jev, a System One decision model, pick at most one skill per prompt from the skills' descriptions, the way TypeSafe's skill-suggestion cookbook does: one request ranks every skill and asks whether the prompt needs a skill at all, a second re-reads the top three with their SKILL.md and can reject them all. The winner's SKILL.md is attached to the prompt, so the skills can be hidden from the model altogether (/jev-skill-suggestion:setup makes them user-invocable-only and /context counts them at 0). Falls back to the engine's built-in classifier when no key is set.
Claude Code - lexicon1stars
Personal voice lexicon for agents: fixes the names, brands and identifiers speech-to-text gets wrong (Ashler -> Ashlr.AI) before Claude acts on a dictated prompt.
Claude Code1 skill1 MCP server - skill-interop0stars
Use when authoring or reviewing a portable multi-host agent skill (Grok, Claude Code, Codex, Hermes): scaffold a prompt-only skill, make a skill host-agnostic, create skill layout, review skill for interop, or install across hosts. Covers prompt-first design, host matrix, anti-patterns (divergent copies, silent mode fallback, abs symlinks), and script-backed CLI contracts.
CodexClaude Code1 skill - shiploop0stars
Markdown-authoritative delivery harness. Start or resume once, follow the script's current action packet, and submit its exact completion call until the script reports completion with an HTML achievement report. Use when the user says shiploop, ship the project, or requests a durable delivery loop.
CodexClaude Code1 skill - shiploop-e2e-audit0stars
Run the ShipLoop test harness and audit its retained graph, review, test, product and incremental-change evidence. Use for ShipLoop mock checks, live one-shot E2E smoke/full campaigns, or review of existing trial output. Full Google Apps Script and Salesforce game cases require an authorized test deployment and hosted behavior evidence. Includes its harness for source and marketplace installs; tests a separately selected ShipLoop.
CodexClaude Code1 skill - review-fix-bench0stars
Compare two code-review prompt versions against supplied fixture ground truth using an explicitly configured external benchmark runner. Reports an F1-based verdict only when the runner completes both evaluations and comparison.
CodexClaude Code1 skill - review-coverage0stars
Add a post-ship improve-to-exhaustion directive to a plan, or run that directive after implementation. Invoke like any skill: /review-coverage, "review-coverage on this plan", residual×2 after ship, or ensure the plan has a /goal-ready ## Review Coverage section. Not for pre-exit plan quality (use review-plan) or raw residual×2 engine mechanics alone (use review-converge under /goal).
CodexClaude Code1 skill - question-bench0stars
Benchmark review-plan question effectiveness via experiment-based ablation. Applies different question subsets to a plan (or directory of plans) in parallel experiments, evaluates each against fixed plan-quality questions (Q-PQ1..Q-PQ8), and compares quality spreads to identify which questions drive real improvement.
CodexClaude Code1 skill - prompt-refine0stars
Full prompt-improvement workflow — runs prompt-audit to find inconsistencies, presents a remediation plan, then runs prompt-migrate to apply fixes and prompt-align to verify harness consistency. Use for any agent or skill prompt that needs structural repair.
CodexClaude Code1 skill - prompt-migrate0stars
TDD-based prompt migration — given a target agent/skill prompt and a remediation list, writes failing tests first, then updates the prompt to make them pass. Commits remain user-controlled.
CodexClaude Code1 skill - prompt-audit0stars
Audit an agent or skill prompt file for internal inconsistencies (phase numbering, behavioral contracts, terminology, stale references). Produces a Q&A with info-gain scores, a learnings section, and a remediation list. Use before any prompt migration.
CodexClaude Code1 skill - prompt-align0stars
Compare an agent or skill prompt against its test harness skill for phase-model, skip-condition, and wiring consistency. Reports mismatches and identifies which file is authoritative.
CodexClaude Code1 skill - plan-test0stars
Generate comprehensive tests for code. Uses an inline strategy or an available independent test specialist for complex components.
CodexClaude Code1 skill - improve0stars
Use when a repository candidate needs a deliberate review-and-improvement loop: use recent Git history, make warranted changes, run meaningful checks, and require two consecutive trivial-only review passes. Supports a read-only interpretation preview; not a one-off code review.
CodexClaude Code1 skill - improve-system-prompt0stars
Benchmark and compare system prompt variants (V2/V2a/V2b/V2c) for Sheets Chat by running test scenarios through the real GAS-side ClaudeConversation pipeline. Tests both system-placement and user-placement, then evaluates with heuristic scoring (ABTestHarness) and LLM-as-judge. Product-specific: requires an authorized Sheets Chat/GAS execution integration.
CodexClaude Code1 skill - evidence-gates0stars
Optional offline evidence gates (freeze/prove/stop with guard digests) for machine-checkable red→green contracts without the autonomous engine. Use when the user says evidence-gates, offline evidence gates, freeze prove stop, or host-native verify gates. NOT the autonomous engine product.
CodexClaude Code1 skill - devloop0stars
DevLoop (default): invoke the autonomous engine for a machine-verifiable build or debug goal. Use when the user says devloop, DevLoop, /devloop, or wants an isolated fail-closed build with executable tests. Thin shim: resolve a preinstalled engine, then exec scripts/devloop-run. Runtime hosts: Grok and Hermes. Claude/Codex/Cursor require an explicit external transport. NOT the demoted offline skill evidence-gates. NOT host-agent DEFINE/PROVE/BUILD.
CodexClaude Code1 skill - derive-questions0stars
Iteratively researches real software project failures and wins, extracts key planning questions via 5-whys analysis, validates them against synthetic test plans, judges their effectiveness via a parallel judge agent, refines them, and persists to a growing questions library. Builds a curated, language/system-agnostic question library that prevents known failure modes when applied during software planning.
CodexClaude Code1 skill - compare-prompts0stars
Compare two prompt versions (A vs B) by running both against a directory of test input files, then evaluating results on three dimensions in priority order: quality > tokens > time.
CodexClaude Code1 skill - c-plan0stars
Resolve ambiguous user prompts by choosing whether to answer now, answer with assumptions, ask 1–2 high-value clarification questions, replan, or stop. Use when the best response depends on hidden intent, audience, scope, constraints, risk, output format, or desired depth.
CodexClaude Code1 skill