Agent Plugins Marketplace

Recently added

Plugins grouped by the day they were added to the directory, newest first.

Sep 14, 2026

  1. Package any app repo as an ambient-library capability with one install standard (init/validate/publish); use for "make a distro", "package this for ambient-library", "publish capability", "distro-kit"

    Claude Code1 skill
  2. Lifecycle operator for the cognitive-mirror database (not the live extractor) — bootstrap deep-mine of raw agent session logs, weekly Mine/Harvest maintenance, quarterly blind audit; use for "bootstrap the mirror", "deep mine my session logs", "audit the mirror", "mirror status", "is my profile still accurate". Prefer over cognitive-mirror when maintaining/auditing/bulk-mining logs rather than extracting from the current conversation.

    Claude Code1 skill
  3. Integrated insight system combining Eigenthinking, latent-space cartography, and wisdom-of-crowds validation; use for "run deep field", frontier-quality insight beyond default reasoning.

    Claude Code1 skill
  4. Builds rigorous text-comprehension assessments with detailed feedback; use for "test my understanding of this text", "comprehension quiz".

    Claude Code1 skill
  5. Extracts named frameworks, methodologies, and publishable IP from a long conversation or session transcript; use for "what did I figure out here", "find the framework in this".

    Claude Code1 skill
  6. Builds a four-stage pipeline turning scattered Claude chat history into a searchable Obsidian external brain; use for "I can't find what I discussed with Claude", "second brain for my AI work".

    Claude Code1 skill
  7. Installs the multi-agent .context/ continuity system into a project; use for "install context-mgr", "set up multi-agent context here", "scaffold .context/", "add the shared brain to this project".

    Claude Code1 skill
  8. Resolves a folder's declared .aai/connectors.md needs against whatever the current harness actually offers (MCP tools, CLIs, or nothing), caching successful matches in .aai/memory/; use for "wire up this connector", "resolve connectors.md", "what can reach X here".

    Claude Code1 skill
  9. Applies the Lenses/Operations/Recipes structured-thinking system to complex problems and decisions; use for "think through", "what am I missing", "run a pre-mortem", "use mental models".

    Claude Code1 skill
  10. Live extractor for the user's cognitive profile — observes the current or recent conversations to model how they think, decide, and evaluate; 5 modes /loms (current chat), Mine (recent chats), Diff, Query, Harvest (decision instances for DSPy); use for "/loms", "look over my shoulder", "mine my conversations", "update my cognitive profile", "harvest my decisions", "how do I typically approach X", "what's my pattern when". Prefer deep-mirror for bulk raw-log mining, scheduling, or auditing. Data lives in ~/.aai/.

    Claude Code1 skill
  11. Preserves conversation context across compaction and hands off work across sessions, agents, and CLI windows — distills since-last-checkpoint state into the project's .aai/checkpoint.md before /compact or /clear, or runs the .context/ store check-in/check-out protocol; use for "checkpoint", "checkpoint before compact", "save context before clearing", "session check-in", "wrap up".

    Claude Code1 skill
  12. Exports and archives a chat conversation or session as portable markdown — conversation only, or with the artifacts the session produced; use for "export this conversation", "capture this chat", "save this session", "export with artifacts". Defaults to the cross-harness .aai session archive; redirectable to any path.

    Claude Code1 skill
  13. Multi-role team for brand-building long-form content with structured intake, end to end; use for "help me write a post about X", "turn this idea into a piece" when the goal is personal/company brand.

    Claude Code1 skill
  14. audit-mcp0stars

    Security-audits a third-party MCP server repo before installation; use for "audit this MCP", "is this MCP safe", "check this repo before I install it".

    Claude Code1 skill
  15. audit-fix0stars

    Adversarial six-lens audit (errors, omissions, oversights, blind spots, conflicts, duplications) of a plan or change, then root-cause fixes; use for "audit this", "pressure test", "poke holes", "what did we miss".

    Claude Code1 skill
  16. Makes a folder ambiently intelligent and maintains it — interviews for purpose when it's unclear, writes the .aai/ scaffold, vendors canonical capabilities into .ailib/, forks them, re-syncs; use for "make this folder ambient", "set up an agentic folder for X", "stamp .aai here", "interview me about this folder", "add a capability to this folder", "re-sync my vendored skills".

    Claude Code1 skill
  17. Autonomous multi-role team (Researcher, Strategist, Writer, Editor, Publisher) for thought-leadership articles; use for explicit role commands like "Use Strategist mode" or a full AIMM article run.

    Claude Code1 skill
  18. Sends a newsletter email to all members of a named Google Contacts group; use for "send this to [group]", "send the newsletter", "email the group". Previews recipients and content before sending.

    Claude Code1 skill
  19. Library of 54 named AIMM prompt commands (skeptic, canon-lock, blind-spot-scanner, transcript-miner, ...); use for "/aimm:<name>", "run the skeptic", or any request naming an AIMM command.

    Claude Code1 skill
  20. Routes a stuck problem to the right kind of mental model from a 2,375-model library, runs a recipe, and returns a five-part answer (plain answer, cognitive mirror, reasoning trace, recipe answer, optional next move); use for a decision, diagnosis, or stuck problem needing the right kind of model, not a pile of frameworks.

    Claude Code
  21. The complete writing toolkit in one install: researcher, writer, editor, project-brief, voice-profile-trainer, and the writing-team orchestrator that drives them end to end.

    Claude Code
  22. Canonical agent, skill, and capability library for Codex, exposed through one lazy-loading ambient router skill.

    CodexClaude Code3 skills
  23. Reflective journaling with adaptable persistent project, memory, or file storage.

    CodexAgent Plugins1 skill
  24. dobbel0stars

    Regnskap for små norske AS og holdingselskap. Les regnskapet, registrer kjøp fra kvitteringer, bokfør banken, hent rapporter og forbered MVA og årsoppgjør. Alt agenten fører blir utkast du godkjenner.

    CodexClaude Code5 skills1 MCP server
  25. fleet0stars

    The fan-out lane: a backlog turned into parallel work across sessions and worktrees — the one thing a flow may never do. fleet-issue-sweep triages a repository's open issues for relevance and for collision with work in flight, proposes the stale ones for closure, and dispatches up to five independent worker sessions, one issue and one worktree each, then waits them out and writes the brief. fleet-resolve-issue is what each worker runs: one issue to a pull request when the change proves itself, or to a parked worktree with a handoff brief when a human has to look. fleet-morning-brief re-reads a past sweep from its manifest and worker result files. An L1 extension over delivery: it owns no flow and holds no gate, and trades Personal Validation for a pull request nobody merges unread.

    Claude Code3 skills
  26. devbook0stars

    The devbook convention: .arc42, .domain, .tech, .design, and .ai folders of addressed Markdown chapters, each carrying a parseable meta block with status ladders, a shared approved rung, runnable test-case links, an opaque ext namespace a layered plugin keeps its own state in, and annotation fences that keep a review note in the chapter beside the passage it is about, compiled into derived _meta indexes and a reference-graph canvas. Ships bidirectional to-spec-<kind> and from-spec-<kind> converters between chapters and code, a versioned contract with runnable migrations, and enforcement tooling. Ships the shape and the check, never a flow: the delivery engine carries a chapter change and reads these rules from the repository.

    Claude Code14 skills
  27. Owns a repository's .devbook/config.json and explains the stack that reads it. `setup` writes the engine-owned keys — bindings, extensions, policy, gates — before anything is installed, then hands every component its own install skill; `update` moves the whole configured stack forward in one run, fanning out to every adopted component’s own install skill and skipping what this machine has not installed or this checkout has not enabled, migrations included; `guide` answers a question about this marketplace from what is on disk — what devbook, the delivery engine, the bridges, the surfaces, and the fan-out lane are and how they fit, which version of each plugin is installed against the newest published, and which are enabled; `adoption` reports where the .ai adoption record no longer matches what is installed and hands the write to flow-ai. All four are backed by a read-only report that names the file behind every fact. Declares no dependency, devbook included: it is named for the folder it writes into, not for a plugin it needs, and one it names but cannot find is reported as not installed.

    Claude Code4 skills
  28. Review, comment, and hand-off workflows over devbook chapters. Requests a review and hands a chapter to a named reviewer, reviews one against its folder's rules and its evidence, records each finding as a devbook annotation fence beside the passage it is about, runs the approval decision that writes devbook's own approved rung, and sweeps the adopted folders for what is awaiting whom and which approvals have gone stale. An L1 extension: the state it remembers is three ext.devbook-collaboration keys in the chapter's own meta block, so it adds no field to devbook's schema and needs no devbook release of its own. Ships devbook-collaboration:install, which puts its one rule in the repository so both hosts apply it on a matching read.

    Claude Code5 skills
  29. delivery0stars

    The host-neutral delivery engine: 16 flow-* staged procedures, five of them one per devbook folder, 3 shared phases, and the pull-request lane, sequenced by the flow-runner agent through a closed set of extension points — spec, implement, verify, app.start, qa.run and deliver as services, session.start, flow.start, data.prepare, docs.update and flow.end as chores — with human gates a repository adds but never removes, and a run surface resolved from the live tool list so none is a normal outcome. Bindings, extensions, policy, and gates are declared per repository in .devbook/config.json.

    Claude Code26 skills
  30. A run surface: the live dashboard a run is watched in. An MCP server implementing all three surface capability groups — lifecycle (open_dashboard, start_run, record_prompt, set_run_context, update_stage, finish_run, list_runs, get_run), render (render_diagram, render_markdown), and export (export_report) — with a run timeline carrying stage output, QA scenarios and their evidence, a Mermaid diagram viewer and a Markdown document viewer, and hook-captured tool and token telemetry. Rendered inline as an MCP App where the host supports it, served on 127.0.0.1 where it does not. Declares no dependency and names no engine.

    Claude Code1 MCP server
  31. A headless run surface: where a run becomes recorded rather than watched. An MCP server implementing two surface capability groups — lifecycle (open_dashboard, start_run, record_prompt, set_run_context, update_stage, finish_run, list_runs, get_run) and export (export_report) — keeping stage status, output, gate decisions, QA scenarios and the handoff marker on disk so a resumed session picks the run up where it stopped, and writing the run's Markdown report at the end. No page, no port, nothing rendered: the right surface for scheduled and unattended runs. Declares no dependency and names no engine.

    Claude Code1 MCP server
  32. The unattended lane, stacked on delivery: the eleven schedule-* skills that run with nobody watching, plus install, plus the catalog of triggers that fires them. Nine schedulable entry points that pick their own input and run a flow or a review — bug fix, merge review, package update, performance review, review, security review, week starter, weekly cost analysis, and what's new — and six trigger definitions naming a target, a cadence, the plugins that target needs, and the prompt for a cloud session that starts with nothing but the repository. delivery-schedule:install creates or updates them through whatever scheduler the live session exposes and records the selection under components.schedule; schedule-status reads their runs and logs; schedule-run fires one now. One capability, two host names: Routines in Claude Code, Automations in the GitHub Copilot app. Depends on delivery and names devbook: a trigger whose target is not enabled is reported and skipped, never scheduled.

    Claude Code12 skills
  33. oms0stars

    oms - see what every Claude and Codex subscription you own has left, and never let one spill into paid credit.

    Claude Code5 skills
  34. concise3stars

    English. The answer goes in the first sentence; cut preamble and process narration; keep every caveat that changes what the reader does. Governs chat replies, plans, decisions, commit messages, PR descriptions, cards, review comments and release notes — and the text written into code, from comments to screens. Forces its output style on while enabled, restates the style beside every prompt, and ships a SessionStart hook with self-update, a credit guard on the commands that publish text, fourteen commands — rewrite, pr, card, commit, comment, release, plan, decide, draw, status, handoff, audit, trim, woman — and an audit agent.

    Claude Code1 skill
  35. Unofficial packaging of the htmx 4 skill files published in bigskysoftware/htmx. Not an official Big Sky or htmx release.

    Claude Code4 skills
  36. Frame, plan, track and trace dev/config/architecture requests through plan files, a consolidated roadmap.md and linked GitHub issues (strict 7-phase workflow). Supports multi-IDE, GitHub-optional local mode, and autonomous test execution.

    CodexClaude Code1 skill
  37. weave3stars

    Weave is a dependency-free engineering layer for Claude Code and Codex. It wraps shell commands to strip noisy successful output while always preserving failures in full, keeps a compact .weave/state.md handoff so context survives across sessions and between agents, and applies one consistent minimalism-and-verification policy so agents reuse existing code, prefer native features over new dependencies, and back every completion claim with real evidence.

    CodexClaude Code6 skills
  38. Cross-harness delivery, review, Agent-Native, visual, and software-factory skills for production work.

    CodexClaude CodeAgent Plugins50 skills1 MCP server
  39. Developer companion to the Prerender Buddy platform for crawler audits and optional workspace evidence.

    CodexClaude CodeAgent Plugins2 skills1 MCP server
  40. ICS/OT Security Operations — CVE enrichment, event investigation, MITRE ATT&CK mapping, and adversarial quality review with convergence-driven analysis.

    Claude Code20 skills3 MCP servers
  41. qualiow11stars

    AI-driven exploratory testing for Claude Code: web (playwright-cli), iOS/Android simulators (Maestro via mobile-cli), and backend/API acceptance-criteria verification — with a versioned QA knowledge base.

    Claude Code11 skills
  42. rx0stars

    Team-blessed RX workflow commands: architect, implement, review, test, qa, pr, commit, and more. The canonical Scientist.com agentic-development command set for the RX Rails monorepo.

    Claude Code1 skill
  43. rranauro0stars

    Worktree-first ticket lifecycle commands: architect → start-ticket → review → cleanup. Configured per project via a dev-suite profile. Personal-share plugin, opt-in, not team policy.

    Claude Code
  44. local-llm5stars

    Local LLM integration for Claude Code — delegates mechanical coding tasks to a local AnythingLLM Desktop workspace to save tokens without sacrificing quality.

    Claude Code1 skill1 MCP server
  45. devops5stars

    Complete DevOps automation for Claude Code: hooks, skills, templates, and git hygiene.

    Claude Code13 skills3 MCP servers
  46. core0stars

    Nucleo de diretrizes de agente. Hooks deterministicos de verificacao, guarda de dependencias, maquina de fases e economia de contexto.

    Claude Code6 skills
  47. Deterministic Claude Code → wmux notifications via UserPromptSubmit / Stop / StopFailure / SubagentStop / SessionStart / PreToolUse hooks. Replaces wmux's regex-based agent detector with 100%-accurate hook signals when this plugin is installed.

    Claude Code
  48. omnimodal1.1kstars

    Omnimodal plugin for Codex and Claude Code: recognize and generate images, video, and audio through Qwen, Z.AI, or OpenAI-compatible APIs.

    Codex1 skill4 MCP servers
  49. feishu2codex1.1kstars

    Community Codex plugin packaging lark-cli skills for Feishu and Lark workflows.

    Codex28 skills
  50. exa1.1kstars

    Community Codex plugin for web search and full-page retrieval through the official Exa MCP server.

    Codex1 skill1 MCP server