What's New¶
The library is at v2.14.15: 434 skills, 121 plugins, 137 agents and 188 commands across 21 domains.
This page is built from CHANGELOG.md every time the site deploys, so it always shows the latest five releases. To pull a release into an existing install:
[2.14.15] - 2026-11-21 — video-studio, source-audit, game-universe-studio and a Codex plugin marketplace¶
video-studio is a new plugin that takes a brief to a finished MP4 the way previz studios work: a free code
animatic first, generated footage only for shots you approve, and a checked conform at the end. source-audit
joins marketing-skills and checks every claim in a draft against the source it links to before you publish.
Codex users can now install whole plugins with codex plugin marketplace add alirezarezvani/claude-skills, and
CI proves each listed plugin loads in the real Codex CLI. game-universe-studio, a community contribution in
engineering-advanced-skills, plans a browser game's assets as one manifest and checks it before you ship. No renames, so nothing needs a reinstall;
claude plugin update picks up the new and changed plugins.
Added¶
- Codex plugin marketplace (
.agents/plugins/marketplace.json):codex plugin marketplace add alirezarezvani/claude-skillsnow lists 111 of our plugins, with the same sources, versions and metadata as.claude-plugin/marketplace.json. Generated byscripts/sync_agents_marketplace.py;bump_plugin_versions.pyregenerates it andcheck_plugin_json.py --allfails on drift. Plugins Codex cannot load are left out and named by--check: nine with a SKILL.md at the plugin root (Codex lists onlyskills/<name>/SKILL.md; aeo, cinematic-shot-prompting and youtube-full still reach Codex through marketing-skills) and self-improving-agent, whose manifest namesiCodex refuses.check_harness_installs.py --harness codexnow installs every listed plugin with the real Codex CLI, one fresh home each, and fails if Codex does not list every skill. Contributed by @GamerDubz (#1157). - game-universe-studio (engineering-advanced-skills bundle): plan a browser game's assets as one JSON manifest (viewport, character turnaround and palette, sprite animations, procedural Web Audio cues, showcase card) and check it with
game_manifest_validator.pybefore any art is generated. BLOCKERs for non-positive sprite and viewport values, gain outside (0, 1], frequencies outside 20 Hz to 20 kHz and showcase contrast below 4.5:1; warnings for missing turnaround views, palette size, gain above 0.4 and aspect-ratio mismatches. Contributed by @287523023qq-hue (#1161). - source-audit (marketing-skills bundle): pre-publish audit for drafts that cite links. Checks each claim against the source it points to and returns one verdict per claim (supported, partial, contradicted, not-found, unreachable, opinion) with a quoted passage, plus a fix list;
extract_claims.pypulls every cited link with its paragraph context. Contributed by @zhengqiuyang (#1158). - video-studio (new top-level
video/domain, one plugin): brief to finished MP4 the way previz studios work, with a free code animatic first and generated footage only for shots you approve. Four skills:video-director(brief gate that derives aspect and a voice-over word budget and refuses briefs over 180 s or with real people but no consent; quote-then-yes protocol for AI shot upgrades),video-beat-board(times each beat at 150 wpm narration and 180 wpm reading, refuses a script that does not fit and says how many words to cut),video-animatic(ascene.v1JSON compiles to one HTML page with a deterministicwindow.__seek(t), Playwright renders it frame-exact at any aspect, a linter blocks unreadable text, flashes per WCAG 2.3.1 and low contrast, and keyframes plus a provider-neutral shot manifest feed optional upgrades) andvideo-conform(one ffmpeg pass to H.264/AAC with loudness to a named target, SRT captions, then QC on duration and loudness). Seven stdlib scripts; no script calls a generation service, and upgrades go through your own MCP connector after a per-shot quote. Ships the cs-video-director agent and/cs-video,/cs-animatic,/cs-conform.
[2.14.14] - 2026-11-17 — mobile-release, cloud-finops and support-ops¶
Three new plugins outside regulatory work. mobile-release gets an iOS or Android update through App Store and
Google Play review and decides whether to pause or widen a staged rollout. cloud-finops reads a FOCUS cost
export from AWS, Azure or Google Cloud for allocation gaps, commitment coverage and cost spikes. support-ops
checks SLA attainment on a business-hours clock, schedules agents per interval with Erlang C and finds tickets
to deflect. All three are stdlib-only and call no provider API. cinematic-shot-prompting is now installable on
its own. No renames, so nothing needs a reinstall; claude plugin update picks up the new and changed plugins
(engineering-advanced-skills, business-growth-skills).
Added¶
- mobile-release (new
engineering/mobile-releaseplugin, also in the engineering-advanced-skills bundle): gets an iOS or Android release through App Store and Google Play review, then rolls it out safely. Three deterministic stdlib scripts:ios_privacy_manifest_check.py(required-reason API use in Swift/Objective-C against PrivacyInfo.xcprivacy, with the valid reason codes per category, SDK-only codes, missing purpose strings and third-party SDKs that ship no manifest);android_release_check.py(Play's target API rule by form factor and date,android:exported, debuggable and cleartext flags, restricted permissions that need a Play Console declaration, and 16 KB page alignment of bundled native libraries read from the ELF headers); androllout_gate.py(PROCEED, HOLD or HALT for the next staged or phased step from a one-sided two-proportion test against the previous release plus Play's bad-behavior thresholds). Ships the cs-mobile-release agent,/cs-mobile-releaseand/cs-rollout-gate, three cited references and a release checklist. Distinct from app-store-optimization, which ranks a listing. - cloud-finops (new
engineering/cloud-finopsplugin, also in the engineering-advanced-skills bundle): FinOps on a FOCUS cost export from any provider.allocation_check.py(allocation coverage against required tags, case-variant tags that split one owner, the largest unallocated resources, shared costs that need a split rule),commitment_check.py(savings plan and reservation utilization, coverage, effective savings rate, unused commitment cost and the uncovered daily floor to size a purchase), andcost_anomaly.py(robust median/MAD anomalies that must also clear a relative and an absolute floor, plus unit cost half over half). Ships the cs-finops-analyst agent,/cs-finops-reviewand/cs-cost-spike, three cited references and a review template. No provider API calls; it reads the export you give it. - support-ops (new
business-growth/support-opsplugin, also in the business-growth-skills bundle): support operations for a helpdesk team.sla_check.py(first-response and resolution attainment on a business-hours clock with holidays, open tickets past target counted as misses, at-risk tickets, backlog age),staffing_plan.py(Erlang C agents per interval for phone and chat, workload plus backlog drain for email, shrinkage and occupancy caps, gaps against the roster) anddeflection_finder.py(Pareto contact reasons labelled FIX_ROOT_CAUSE, DEFLECT, AUTOMATE or REVIEW). Ships the cs-support-ops-lead agent,/cs-support-sla,/cs-support-staffingand/cs-support-deflection, and three cited references.
Changed¶
- Skill-router hooks (UserPromptSubmit, off with
SKILL_ROUTER=0) now ship in the three new plugins and hint them from the engineering-advanced-skills and business-growth-skills bundles, for release, cloud-cost and support-queue questions the model would otherwise answer from memory. - Headline counters: skills 422 → 428 (3 standalone plugins plus their bundle copies), tools 841 → 861, references 949 → 967, agents 133 → 136, commands 178 → 185, plugins 117 → 120.
business-growth/CLAUDE.mdlists support-ops. - cinematic-shot-prompting is now its own marketplace plugin (
/plugin install cinematic-shot-prompting@claude-code-skills), so you can install it without the full marketing-skills bundle. It stays in the bundle too.
[2.14.13] - 2026-11-17 — NIS2, DORA and readme-polish¶
Two new EU cyber-compliance plugins for laws that already apply: nis2-readiness for NIS2 and the German
NIS2UmsuCG (in force since 6 December 2025), and dora-readiness for DORA (applies since 17 January 2025). Both
compute the reporting clocks from your own timestamps, refuse to guess when data is missing, and never submit
anything for you. Also new: readme-polish from the community. No renames, so nothing needs a reinstall;
claude plugin update picks up the new and changed plugins (ra-qm-skills, engineering-advanced-skills).
Added¶
- readme-polish (new
engineering/readme-polishplugin, also in the engineering-advanced-skills bundle): upgrades an existing README to the standard of the leading repos in its own category. It calibrates the presentation tier against 2-3 peer repos, runs a 15-second first-screen test, marks every claim true, stale or invented, rewrites the hero and hierarchy, and gates the result withscripts/check_readme.py, a stdlib-only lint for dead local links, placeholder badges, broken or non-ASCII anchors, oversized mermaid diagrams, wide tables and a missing first-screen install command. Thanks to @zhengqiuyang for #1145. We trimmed SKILL.md under 100 lines (component selection, anti-patterns and triggers moved into its references) and registered it as a standalone marketplace plugin. - nis2-readiness (new
ra-qm-team/nis2-readinessplugin, also in the ra-qm-skills bundle): NIS2 (Directive 2022/2555) and the German NIS2UmsuCG, in force since 6 December 2025. Four deterministic stdlib scripts:nis2_scope_classifier.py(Annex I/II sector, Recommendation 2003/361 size, essential or important under Art 3, size-independent cases, jurisdiction under Art 26, registration under Art 27 and BSIG §33, DORA overlap);nis2_incident_clock.py(significance under Art 23(3) or Implementing Regulation 2024/2690 for digital providers, the 24 h early warning, 72 h notification and one-month final report, report content, informing recipients);art21_gap_check.py(30 ids across Arts 20, 21, 23 and 27, with the 2024/2690 Annex section behind each); andeu_cyber_regime_router.py, which lays the NIS2, DORA, CRA and GDPR reporting clocks for one event side by side. Ships the cs-nis2-officer agent,/cs-nis2-readiness,/cs-nis2-reportand/cs-eu-cyber-scope, three cited references and an incident runbook. - dora-readiness (new
ra-qm-team/dora-readinessplugin, also in the ra-qm-skills bundle): DORA (Regulation 2022/2554) for EU financial entities. Three deterministic stdlib scripts:dora_scope_check.py(the Art 2(1) entity types, Art 2(3) exclusions, the Art 16 simplified framework, microenterprises, NIS2 overlap, critical ICT provider oversight);dora_incident_classifier.py(major-incident classification under Delegated Regulation 2024/1772 Arts 6-9 including recurring incidents, and the 2025/301 clock: initial 4 h after classification and within 24 h of awareness, intermediate 72 h, final one month, with the weekend rule that credit institutions, CCPs, trading venues and NIS2 entities do not get); andregister_of_information_check.py(provider LEI, the Art 30(2) and 30(3) contract terms, Art 28(8) exit plans, Art 28(4) risk assessment, Art 29 concentration and third-country risk). Ships the cs-dora-officer agent,/cs-dora-readinessand/cs-dora-incident, three cited references and a major-incident runbook. Not to be confused with DORA delivery metrics, which stay in vpe-advisor.
Changed¶
ra-qm-teambundle description no longer carries a skill count, andra-qm-team/CLAUDE.mdlists the standalone readiness plugins. Headline counters: skills 418 → 422, tools 827 → 841, references 937 → 949, agents 131 → 133, commands 173 → 178, plugins 114 → 116.
[2.14.12] - 2026-11-16 — monetize, two math fixes, triage autonomy¶
A patch release. New productivity plugin monetize, the cinematic-shot-prompting skill that reached main on
2026-11-12 without a release entry, and two community fixes to statistics that were silently wrong: the Welch
t-test confidence interval and the FMEA improvement percentage. No renames, so nothing needs a reinstall;
claude plugin update picks up the changed plugins (marketing-skills, engineering-advanced-skills,
statistical-analyst, ra-qm-skills, compliance-os), each bumped above its previous version (gate G11).
Added¶
- cinematic-shot-prompting (
marketing-skill/skills/cinematic-shot-prompting, in the marketing-skills bundle): compiles a prose scene description into a strict seven-slot camera-motion contract for AI video generators (Runway, Kling, Luma) withscripts/motion_contract_compiler.py. Thanks to @287523023qq-hue for #1132. It installs with marketing-skills; it has no standalone marketplace entry yet. - monetize (new
productivity/monetizeplugin): turns one person's career history into a side-income plan with a hidden-skills audit, an opportunity scan, productization, one leverage action for this week, a salary-ceiling benchmark and an AI starter pack. Every income figure is a range with a source or an "estimate, unverified" label, andscripts/report_lint.pychecks the report against its contract. Thanks to @zhengqiuyang for #1138 (question in #1137). We moved it frombusiness-growthtoproductivityas its own plugin, since it serves an individual rather than a company's revenue team, and the market reference now infers the region from the user instead of defaulting to one.
Changed¶
- Daily triage policy (
.github/TRIAGE_POLICY.mdsection 2): the routine now lands, merges, closes and comments on its own when an item breaks no repo rule and every check is green, including fully verified release merges, and reports afterwards. Security holds, new third-party connections, red or overridden checks, fork workflow approvals and genuine judgment calls still wait for the maintainer.
Fixed¶
- statistical-analyst
hypothesis_tester.py: the Welch t-test confidence interval used the normal critical value 1.96 while the p-value used the t-distribution, so small samples got a CI that excluded 0 while p said not significant (n=5 vs 5: p=0.074 with CI [0.12, 5.08]; correct CI [-0.32, 5.52]). The CI now uses the t critical value at the Welch-Satterthwaite df. Thanks to @divya0795 for #1141. - risk-management-specialist
fmea_analyzer.py: the improvement percentage counted every failure mode without revised scores as if its RPN had fallen to 0 (two modes at RPN 200, one revised to 100, reported 75% instead of 25%). Unrevised entries now keep their original RPN. Thanks to @divya0795 for #1142. - Headline counters (skills 415 → 417, tools 824 → 826, references 931 → 934, plugins 112 → 113) the README Marketing row and the harness mirrors, which still reflected the tree before
cinematic-shot-prompting(#1132) merged. The G12 listing baseline formarketing-skillsrises by 226 characters to accept that skill's description.
[2.14.11] - 2026-11-11 — CRA, IEC 62304 and skill hints¶
Two new compliance plugins: cra-readiness for the EU Cyber Resilience Act, whose Article 14 reporting already
applies, and medical-device-software for IEC 62304 and IEC 81001-5-1. Eight plugins now ship a quiet
UserPromptSubmit hook that names one of 21 skills that never loaded on their own when a prompt clearly fits it;
it makes no model or network calls, and SKILL_ROUTER=0 turns it off (SKILL_ROUTER_<PLUGIN>=0 for one plugin).
No renames, so nothing needs a reinstall: claude plugin update picks up the hooks, and the new plugins install
as usual. Changed plugins are bumped; plugins above 2.14.11 from earlier releases keep their versions (gate G11).
Added¶
- cra-readiness plugin: EU Cyber Resilience Act (Regulation (EU) 2024/2847) (
ra-qm-team/cra-readiness, also bundled in ra-qm-skills). Article 14 reporting has applied since 11 September 2026 to every in-scope product on the EU market, including products shipped before the CRA; everything else applies from 11 December 2027. Four deterministic stdlib scripts:cra_scope_classifier.py(scope and Art 2 exclusions, roles under Arts 19-22 and 24, Annex III/IV class, the conformity routes Art 32 allows, legacy products under Art 69, the Art 13(8) support period, and the CSIRT that receives reports under Art 14(7));art14_report_clock.py(24 h early warning, 72 h notification, final report 14 days after a fix or one month after an incident notification, required content, user notice under Art 14(8), and the Art 64(10)(a) SME carve-out);annex_i_gap_check.py(39 ids across Annex I Parts I and II, Annex VII and Annex II, refusing NA on vulnerability handling);sbom_check.py(CycloneDX or SPDX JSON, primary component, top-level dependencies per Annex I Part II(1)). Ships the cs-cra-officer agent,/cs-cra-readinessand/cs-cra-report, three references citing the regulation, Implementing Regulation (EU) 2025/2392 and Commission guidance C(2026) 5252, and an Article 14 runbook. - Skill hints for 21 skills that never loaded on their own. Eight plugins (engineering-skills,
engineering-advanced-skills, marketing-skills, product-skills, c-level-skills, agenthub, autoresearch-agent,
si) now ship a
UserPromptSubmithook,hooks/skill_router.py, that names the fitting skill when a prompt clearly matches one (tdd-guide, senior-backend/frontend/devops/secops/data-scientist, gcp-cloud-architect, minimalist, codebase-onboarding, self-eval, tech-debt-tracker, copy-editing, content-humanizer, marketing-psychology, experiment-designer, product-manager-toolkit, roadmap-communicator, cto-advisor, agenthub, autoresearch-agent, remember). Silent otherwise, one hint per plugin per prompt, once per session, no model or network calls;SKILL_ROUTER=0turns it off. Measured live on held-out prompts: the skills loaded in 24/84 runs with the hook against 2/84 without, and 0/21 on near misses either way. No new commands: every skill is already/<plugin>:<skill>. Per-skill decisions inaudit/silent-skill-triggers-2026-11/;scripts/check_skill_router.py(CI) keeps the copies identical and the routes valid. - Daily triage pipeline for community PRs and issues:
scripts/triage_snapshot.pylists every open item with mechanical signals (CI or manifest changes, network and LLM calls in added code, new dependencies, external domains, commercial wording, untouched issue templates) and a suggested lane, without executing or echoing contributor text..github/TRIAGE_POLICY.mdsets the pipeline from snapshot to draft PR ondev, the documentation update every PR carries, release prep (CHANGELOG version entry, version lines, thedev → mainPR, with the tag left torelease.yml), what runs without the maintainer and what waits for a per-batch go-ahead, and runs contributor code only in a network-less namespace, because every process in the cloud container inherits the session's GitHub access. - medical-device-software plugin: IEC 62304 and IEC 81001-5-1 (
ra-qm-team/medical-device-software, also bundled in ra-qm-skills). Four deterministic stdlib scripts:software_safety_classifier.py(IEC 62304 4.3 class per system and item with the Amendment 1 external-control path and segregation, plus the FDA Basic/Enhanced documentation level, which is assessed before risk controls, and the FD&C Act 524B cyber-device test);lifecycle_gap_check.py(every 62304 activity by class, clause 4.4 legacy software, and IEC 81001-5-1 activities with--security; unassessed is not MET);soup_register_check.py(8.1.2, 5.3.⅗.3.4, 7.1.3, end of support, the 81001-5-1 risk-transfer category as clarified by ISH1:2025, and a CycloneDX/SPDX SBOM cross-check for 524B(b)(3));trace_matrix_check.py(requirements to tests, risk controls to hazards and verification, release state). Ships the cs-device-software-lead agent,/cs-medical-device-softwareand/cs-soup-review, and four references that label every claim as checked against the primary text, agreed by secondary sources (the paywalled 62304 class table), or unverified (EU harmonisation of EN 62304 and EN IEC 81001-5-1 under the MDR).
Fixed¶
- regulatory-affairs-head IEC 62304 table no longer marks integration testing required for class A or calls IEC 62304 harmonised under the MDR; fda-consultant-specialist names the Basic/Enhanced documentation level that replaced the level of concern in June 2023. ra-qm-skills 2.14.12.
Older releases: CHANGELOG.md.