claude

forge-state

Lab bench for open-source projects. 14 skills: state spine, orchestrator, 4 harvesters (slack + github + rss + watchlist), researcher, builder, experimenter, packager, reporter, publisher, the EXP-0006-emitted agentic-rl runner, and forge-introspector (self-reflection on capability walls).

0 on GitHubby David Olssonv0.1.4MITupdated 1 month agoproductivity
/

Plugin forge-state from marketplace forge-state in Atomic-47-Labs/forge-state-public. Commands read from the repo's manifest — 27 days ago.

What it installs

Skills (14)

  • forge-agentic-rl

    Use to run a multi-turn LLM-agent rollout with GRPO-style group-relative advantages and AgentRL-style task renormalization, inside a no-secrets Docker sandbox. Wraps the `agentic-rl-runner` Python package shipped by EXP-0006. Suitable for benchmarking any chat-completion API (Anthropic, OpenAI, vLLM

  • forge-builder

    Use when a forge experiment is at phase `researched` and needs to be cloned, pinned, classified, and built inside the Docker sandbox. Invoked by forge-orchestrator. The riskiest skill — proves the path before downstream skills run. Trigger phrases include "/forge-builder EXP-NNNN", "build EXP-0007 i

  • forge-experimenter

    Use when a forge experiment is at phase `built` and needs a bounded real-world experiment run against it inside the Docker sandbox. Skipped automatically for `build-failed` experiments (they route straight to reporter). Trigger via "/forge-experimenter EXP-NNNN", or invoked by forge-orchestrator. Pi

  • forge-harvester-github

    Wild-source intake. Use GitHub Code Search (gh api) to find public repos containing one of forge's tracked agent-instruction files — SKILL.md, AGENTS.md, program.md, .claude/, .cursor/, .codex/ — and enqueue the freshest, highest-signal ones as forge candidates. Targets the strongest cross-experimen

  • forge-harvester-rss

    Wild-source intake. Watch RSS / Atom feeds of high-recipe-density sources — MarkTechPost, Hugging Face blog, Anthropic blog, Cameron Wolfe substack, the dottxt blog, opensourceprojects.dev — for posts that look like *system specs* rather than news. Filter for recipe-style headlines ("Using X and Y t

  • forge-harvester-slack

    Use when the orchestrator's nightly walk begins, or when the user says "/forge-harvester-slack", "check for new forge candidates", "any new lab-bench marks", or asks to drain Slack reactions into experiments. Reads David's own 🧪 reactions in #development (per locked manifest), creates new candidate

  • forge-harvester-watchlist

    Wild-source intake. Watch a configured list of GitHub authors / orgs that have historically produced forge-quality work — Karpathy, dottxt-ai, motiful, Mentra-Community, HKUDS, safishamsi, calesthio, nolly-studio, github (Spec Kit), pinokiocomputer — and enqueue any new public repos (or major new re

  • forge-introspector

    Read forge's own capability walls and propose how to overcome them. Walks every published experiment's "What I didn't run" section, every entry in state.json.policy_notes[] / follow_up_notes[], and every error-shaped activity event, clusters the limitations by theme, and emits capability work orders

  • forge-orchestrator

    Use when starting a nightly forge walk or when the user runs "/forge-orchestrator" (optionally with --once or --dry-run). The conductor of the forge facility — thin router that walks the experiment queue, dispatches each non-terminal experiment through the §6 lifecycle, respects state/forge.lock (fl

  • forge-packager

    Use when a forge experiment is at phase `experimented` and needs to be turned into a shareable, runnable artifact — a deploy/ bundle (Dockerfile, compose.yaml, RUN.md). Also pushes the image to the configured registry when packaging.push_image is true (token at env $GHCR_TOKEN, registry read from ma

  • forge-publisher

    Use when a forge experiment is at phase `reported` and needs its outputs shipped — gist (auto), scsiwyg blog post (gated on manifest.surfaces.blog.enabled — currently off), and a work-state event emission. Scrubs secrets before every send. Trigger via "/forge-publisher EXP-NNNN" or invoked by forge-

  • forge-reporter

    Use when a forge experiment is at phase `packaged` (success path) or `build-failed` (failure path) and needs a publishable writeup composed. Build failures get reports too — "this does not build on a clean machine" is a finding worth publishing (spec §2.5). Trigger via "/forge-reporter EXP-NNNN", or

  • forge-researcher

    Use when a forge experiment is in phase `candidate` and needs its research block filled — what the project is, who built it, why it's notable, comparables, and license. Trigger via "/forge-researcher EXP-NNNN", or invoked by forge-orchestrator on every candidate during a nightly walk. Writes researc

  • forge-state

    Use when any forge-* skill or user request needs to read, write, or validate the forge facility at ~/forge/ — including init, allocating an EXP-NNNN id, advancing an experiment's phase, scrubbing secrets from outputs, or appending to state/activity.ndjson. The only skill permitted to write the subst

View source on GitHub

Similar plugins

caveman

claude

Talk like caveman. Cut filler. Keep technical details explicit.

22 skills · 1 command · 3 agents · hooks

JuliusBrussee/caveman 119,462

claude-mem

claudecodex

Persistent memory for Claude Code: the next session starts where the last one left off.

29 skills · 1 MCP server · hooks

thedotmack/claude-mem 116,618

Taste Skill

claudecodex

tasteskill 2.5.1: professional company websites at the figma.com and x.ai level, with one section per screen, real photographs, real product UI, effects kept for tech and gaming pages, and a real-browser checker and finish gate.

1 skill

Leonxlnx/taste-skill 114,391

career-ops

claudecodex

Open-source AI job search that runs locally in your AI coding CLI: scan job portals, evaluate listings into a structured A-H report with a 1-5 score, tailor your CV, track applications

5 skills · hooks

career-ops-hq/career-ops 82,253

mem0

claudecodex

Cross-session memory and token savings for coding agents.

6 skills · 1 agent · 1 MCP server · hooks

mem0ai/mem0 68,893

pake

claudecodex

Package any website or local web build into a lightweight desktop app with Pake CLI.

1 skill

tw93/Pake 63,463

Maintain forge-state?

This page updates itself from your repo. Add the badge to your README, or get it Featured to pin it above the organic results with a followed link to your site.

claude & codex pluginslisted