agentleFS
AJschoonsAJschoons/robowright/CLAUDE.md

robowright

CLAUDE.md1 starsChanged 9 days agoMITView on GitHub

This repository is the system itself: the skills, agents, launcher, hooks and dashboard that run Robowright Sessions. A change here changes what runs on the Owner's machine. - Everything a session writes is untrusted input. That covers: - run.json, events.jsonl and .stop-gate-count in a session directory; - LEDGER.md, HANDOFF.md and HANDBACK.md; - the session directory's own name, and any directory created beside it; - a robowright.json under the sessions root, and a project folder's robowright.json: Planners write it, and a…

The file

Files
CLAUDE.md

Robowright — working in this repository

This repository is the system itself: the skills, agents, launcher, hooks and dashboard that run Robowright Sessions. A change here changes what runs on the Owner's machine.

Trust model

  • Everything a session writes is untrusted input. That covers:

    • run.json, events.jsonl and .stop-gate-count in a session directory;
    • LEDGER.md, HANDOFF.md and HANDBACK.md;
    • the session directory's own name, and any directory created beside it;
    • a robowright.json under the sessions root, and a project folder's robowright.json: Planners write it, and a Worker can too wherever your own settings grant it the projects root.

    A session controls all of these, and a prompt injection can control the session.

  • The Owner-side tools that read them run outside the sandbox. The launcher, accept.sh, role-start.py, run-cost.py, session-state.py, the roster, the dashboard and the worker hooks all do. A value from a session file reaches a command line, a JSON blob, git or the Owner's terminal only after a shape check. SECURITY.md → "Values a session can write" lists every value and its check.
  • The limits a session runs under are instructions, not enforcement. Never widen them from a value a session wrote.

Every change

This is the maintainer's workflow for the main clone. A contributor works through pull requests, per CONTRIBUTING.md.

  • The maintainer signs every main tip. Signing is opt-in in accept.sh; the maintainer always accepts this repository with accept.sh -S, never rewrites main, and is the only one who pushes it. A session never pushes.
  • Tests never carry a working payload. A security test proves a refusal or a validity (.claude/rules/tests.md).
  • bash tests/run.sh ends EXIT=0. It includes tests/conventions/, the lint for the rules below.
  • Wire the hooks: git config core.hooksPath tools/hooks. The pre-commit hook runs the scrub gate and the conventions lint on the staged tree. Never commit with --no-verify.

Where the standards are

  • .claude/rules/. Each rule loads when you read or edit a file it covers:
    • session-inputs.md: scripts, hooks, tools/ and install.sh;
    • dashboard.md: the dashboard and its tests;
    • public-content.md: Markdown, tests and examples;
    • tests.md: test files.
  • CONTRIBUTING.md: the bar, commits, and the conventions a change keeps.
  • SECURITY.md: what a session can do, every session-written value and its check, and what the checks do not cover.
  • .claude/robowright/project.md: this repository's project hook (build, tests, kind checks, commits).

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.