Plugins and skills for AI agents. The skills follow the Agent Skills open standard, so their bodies and bundled scripts are portable across tools that speak it. The packaging in this repo — hooks included — targets Claude Code, which consumes it as a plugin marketplace; other formats may join as the collection grows.
Add the marketplace, then install a plugin:
/plugin marketplace add 4nrry/agent-plugins
/plugin install agent-fleet@4nrry
Or paste 4nrry/agent-plugins into the Claude Desktop Add marketplace dialog.
| Plugin | What it is |
|---|---|
| agent-fleet | Rules for multi-agent work — model routing per phase, fan-out sizing, output schemas, and citation verification. Two hooks, because the prose alone measurably did not fire. |
| agents-md | Write, slim and audit a repo's AGENTS.md: what goes in, what comes out, per-tool limits (Claude Code, Antigravity, Codex), and a stdlib checker for missing scripts, dead paths, oversize files and the CLAUDE.md that hides AGENTS.md from Claude Code. |
| crew | Cheap headless workers through three agent CLIs, without fighting Claude's harness: Claude writes the task, sets scope and gate, and reviews the diff; the worker runs one slice under its own CLI's hooks, installed for the run only. Backends: agy (Antigravity CLI, Gemini; default), opencode (free Zen models, weaker shell guard) and GitHub Copilot CLI (Free plan: --model auto only, paid in AI credits, no shell). The verdict never reads exit 0 as success — in agy 1.3.4 timeouts, denied actions and an unmet --json-schema all exit 0 (observed, n=1 each). One skill per backend plus a shared delegation doc, injected by a UserPromptSubmit hook when the prompt names the CLI, and a watcher that ends the background commands agy would otherwise wait 30 min for. |
| game-facts | Pins the installed game's version before the agent answers from training memory. A UserPromptSubmit hook reads Steam appmanifests and injects appid, buildid and update date — so the version never has to be typed into the prompt. |
| homelab | Software you maintain yourself on a Linux box, where the official docs stop. A Palworld dedicated server (the config the server erases on shutdown), WireGuard through NetworkManager (the IPv6 leak that passes the IP test), and tarball installs (one versioned layout, and why a self-updating app must not live in /opt). Machine state stays out of the repo. |
| media-agent | Producao de midia numa maquina propria com GPU de 8 GB: locucao a -16 LUFS com say separado da legenda, trilha e bancos com a licenca de cada item, retoque no ComfyUI so com modelos comerciais, turntable no Blender headless, video como codigo edicao de video pronto pelo MCP do Kinocut, imagem e video gerados pelas assinaturas de Gemini e Grok, e a ElevenLabs alem da voz: SFX, trilha, limpeza de gravacao, legenda e dublagem. Um hook que avisa onde o vipsthumbnail -o vai gravar de verdade. |
| mobile-agent | What breaks silently in a mobile app: the device layer (adb, emulator, UI dumps) and the bundler layer (Expo, Metro, prebuild). Six narrow PreToolUse guards that warn and never block. |
Each plugin's own README carries its components, its measurements, and what those measurements do not establish.
Improvement claims about plugins in this repo are comparisons between run
records committed under bench/plugins/<plugin>/results/, collected per
bench/PROTOCOL.md — per-agent token splits, resolved
model IDs, paired arms with hash-checked prompts, and outcomes tagged by
source (script measurement vs orchestrator assertion). A claim without
records is marketing and does not belong here.
Records collected before that protocol existed, or by a foreign harness, are published as raw output with their hashes and are labeled imported in the first paragraph of their claims file, together with every field they lack and every claim that lack forbids.
Findings that belong to no single plugin — a fact about model routing or
harness behaviour that any plugin would reuse — go in
bench/shared/.
just check
Runs bench/validate.py: manifests carry no duplicated
metadata, every hook command points at an executable file, every run record
conforms to bench/schema/run.schema.json,
every eval_ref resolves and hashes to the eval_sha256 beside it, every
bench/ path cited in a plugin description exists, shellcheck passes, and every
script advertising --self-test passes it. CI runs the same one command.
.claude-plugin/marketplace.json entries are {name, source}; plugin.json is
the single authority for plugin metadata
plugins/<plugin>/ the installable plugin, with its own README
bench/PROTOCOL.md validate.py how records are collected, and the checker
bench/schema/ run record + eval file shapes
bench/shared/ findings belonging to no single plugin
bench/plugins/<plugin>/ evals, results and scripts for that plugin
MIT — see LICENSE.