Recently added

Turns any subject into a self-teaching campaign using Benjamin Franklin's two modes: acquisition, then reconstruct-and-diff refinement. Every check is external - Claude never grades the work.

Turn a career into a verified record, then render resumes, LinkedIn profiles, and executive letters from it. A recognition harness reconstructs what you actually did at autobiography depth; a validator refuses to let any document claim more than the evidence supports.

Entwurf async receive mode for Claude Code — a portable, drop-in plugin that arms a PER-SESSION mailbox watch at SessionStart (watchPaths = $CC_MAILBOX_ROOT/<session_id>/inbox.signal) so an IDLE subscription session can be woken by an addressed external file-write (agy send-message parity; pair with cc-enqueue-addressed.sh). Only the targeted session wakes — siblings stay idle. This is the plugin form that a bare .claude/skills/ skill CANNOT achieve: plugin hooks register at startup, so the SessionStart hook actually fires and registers watchPaths. A bare skill's hooks register only on invocation (mid-session), after SessionStart has already passed.

Minimum plugin shape that entwurf's `skillPlugins` validator accepts. Copy this directory anywhere on disk, point `skillPlugins` at the absolute path, and add your own `skills/<name>/SKILL.md` siblings to `hello/`. The entwurf bridge does not require any other key in this manifest — additional fields are forwarded to the Claude Agent SDK as-is.

Garden-native meta-bridge receive for Claude Code (entwurf 1.0.0). On every native session open the SessionStart hook idempotently upserts a meta-record at <pi-agent-dir>/meta-sessions/<garden-id>.meta.json (making the native session a garden citizen addressable by garden id) and arms a per-garden idle-wake watch at <pi-agent-dir>/meta-mailbox/<garden-id>/inbox.signal. CwdChanged re-arms; UserPromptSubmit backfills the record only (cannot re-arm). FileChanged is the addressed doorbell that wakes an idle session when a sender pokes its signal. Installed via the meta-bridge local marketplace with --scope user; do NOT hand-edit — use ./run.sh install-meta-bridge / doctor-meta-bridge.

Garden-native meta-bridge BIRTH for GitHub Copilot CLI (entwurf, #82). Copilot's hook schema has no `args` array and its exec form is a single `exec` string, so this unit is a sibling of entwurf-meta-receive rather than a variant of it. On the first prompt of a Copilot session (measured: sessionStart is deferred to the first prompt, userPromptSubmitted fires just before it) the hook idempotently upserts a meta-record at <pi-agent-dir>/meta-sessions/<garden-id>.meta.json, making the Copilot session a garden citizen addressable by garden id. It also writes a SENDER marker keyed to the Copilot host pid, so the session's MCP children can say which citizen is calling. This birth unit does NOT arm a receiver, and that stays true now that one exists: Copilot's doorbell is a forked first-party EXTENSION (installed separately with 'run.sh install-copilot-receive'), and only the process holding that watch may claim it. Birth gives identity and who-sent; the extension gives replyability. Installed via the meta-bridge local marketplace into Copilot; do NOT hand-edit.

Stage-keyed skills for the full Raycast-extension lifecycle: api-changelog (derive @raycast/api release notes Raycast doesn't publish), scaffold (new), develop (change code, modernize, house-style audit fix), ship (dep hygiene, house-style audit, Store compliance, PR, mirror, cleanup), greptile-loop (drive an open Store PR's automated review to 5/5), review-pr (inbound review of a third-party PR).

Rewrite AI-sounding text so it reads naturally without changing what it says.

Rewrite AI-sounding text so it reads naturally without changing what it says.

Anban Creator plugin for OpenAI Codex — WeChat, Seednote, Montage, ecommerce, and agent-assisted publishing workflows

Personal agentic development workbench with role-named agents (software-engineer, code-reviewer, product-manager, project-manager, research-librarian, ux-product-designer) backed by action-named skills for code, planning, research, product strategy, design, and ASO work

/aitmpl opens a component browser above the prompt: the aitmpl.com catalog (agents, commands, MCPs, settings, hooks, skills, loops, mods) read live from the site's public JSON through $.http.fetch. Pick a type or a trending component, filter, page, open a component and install it from there ($.process.run with the fixed CLI argv and a validated path, no shell), drop its install command into the prompt, or open its page in your browser. Hooks session.start, command.run and ui.render on AbovePrompt; browsing costs no tokens.

Stack-agnostic, ad-hoc orchestration framework. karta synthesizes a binder of work items (karta-plan) and delivers it in parallel waves onto a per-binder integration branch (karta-deliver), building each item in an isolated git worktree (karta-build) and verifying it against its own acceptance check — behavioral (karta-verify) or visual (karta-validate). Works across stacks (frontend, backend, CLI, data, IaC, …) while keeping deep frontend support: component/icon/token mapping, DTCG conformance, and a design-validation loop.

Stack-agnostic, ad-hoc orchestration framework. karta synthesizes a binder of work items (karta-plan) and delivers it in parallel waves onto a per-binder integration branch (karta-deliver), building each item in an isolated git worktree (karta-build) and verifying it against its own acceptance check — behavioral (karta-verify) or visual (karta-validate). Works across stacks (frontend, backend, CLI, data, IaC, …) while keeping deep frontend support: component/icon/token mapping, DTCG conformance, and a design-validation loop.

Development discipline for long-running projects: a seven-stage pipeline restated on every prompt and after every answer with the shape its report takes, a per-session task registry, capped task memory, a brief for every subagent, and warnings — optionally blocks — when two live sessions are editing the same files.

MCP bridge for live-editing open PowerPoint presentations via Office.js.

Skills and agents to automate repository maintenance, releases, and CI/CD workflows

vault-tools is an open-source Claude Code plugin indexed from GitHub. Review its manifest and source before installing.

AI Context Hub: shared real-time workspace context (tasks, decisions, presence) for Claude Code. Bundles the hub MCP server and /hub:* slash commands so sessions read from and write to the same project state as Cursor, Codex, and the dashboard.

The tagteam handoff contract for Claude — the /handoff skill and a SessionStart hook that surfaces the current cycle. The engine is the tagteam CLI (uv tool install tagteam).

Governança de projeto para desenvolver com agentes: a estrutura que sustenta o projeto (PRD, roadmap, demandas versionadas) e o ciclo que leva cada demanda da ideia ao registro do que foi feito.

Session analysis tools for cLens — analyze distilled agent sessions, compare performance, and deep-dive into backtracking patterns

Map markdown acceptance items to explicit implementation evidence for advisory spec-drift checks.

Bounded public search evidence with source-explicit supersearch.search.v1 receipts (pinned hermes-supersearch 0.11.0); evidence, not answers

Python quality gate CLI for Ruff, Pyright, and pytest with bounded auto-repair and escalation artifacts

CSV preflight validation and batch CSV quality checks that fail fast before pipeline runs.

A guided, project-local first experience for Claude Code that helps beginners start making something real.

Jailbreak regression benchmark for LLM endpoints with repeatable known-pattern attacks and deterministic scoring

Run the released Quick Gate CLI ([email protected]) as one deterministic JavaScript/TypeScript gate and read its gate-result/v1 verdict

Receipt-bound completion rail for Claude Code: SessionStart injects the completion contract, Stop runs the cached fast gate advisory-only, and PreToolUse denies a Bash commit/push/PR boundary command missing its matching receipt.

Developer-first Apify data APIs, Agent Skill, and MCP tools for public-data workflows including RAG, ecommerce, search, video, Reddit, jobs, and restaurant menus.

Screens submitted prompts with a local Little Canary service. Advisory mode flags suspicious input; block mode stops turns the service rejects.

Returns concise LintLang repair guidance after Claude Code changes a supported language-bearing file.

On-demand static analysis of AI system prompts: audit a prompt file for contradictions, coverage gaps, priority ambiguities and meta-paradoxes, locally and offline.

Blocks permanent-delete shell commands in Codex and redirects agents to a recoverable trash workflow.

Blocks permanent-delete shell commands in Claude Code and redirects agents to a recoverable trash workflow.

Recoverable deletion for coding agents: move files to trash, list them, and restore them with agent-trash.

Advisory-only Stop-hook drift gate for Codex — regex-only, zero-LLM, no network.

Legacy advisory Stop-hook bundle for the fixed English gate. Current Claude Stop compatibility is not certified in v0.1.7; use the CLI directly. Requires the hermeneutic package.

Orka is a multi-agent engineering harness for Codex and Claude Code: resilient Jira sprint scheduling, ticket decomposition, isolated implementation, bounded reviews, and merge-on-green.

Download Jenkins workspaces with hidden files and optional dependency, cache and artifact filtering.

Architectural conventions for any repo. Includes 1 skill.

How to write instruction files for AI agents

Product screenshots and demo clips go stale the moment the UI changes, and re-shooting them by hand costs an afternoon. Spectra drives your running app across web, macOS, iOS, and watchOS to capture screenshots, videos, and narrated walkthroughs, then files every asset in a searchable tagged library. You get publish-ready visuals for docs, launch posts, and App Store previews without reshooting.

demo is an open-source Claude Code plugin indexed from GitHub with 1 skill. Review its manifest and source before installing.

Human-gated delivery for Codex and Claude with shared dependency graphs, source and task decomposition, one dashboard, and explicit native host limits.

Spec-driven, review-enforced, hook-gated AI-native development workflow for Codex

Spec-driven, review-enforced, hook-gated AI-native development workflow for Claude Code

Spec-driven, review-enforced, hook-gated AI-native development workflow for Codex

Plan, coordinate, and execute durable project work across ChatGPT and Codex.