agentleFS
Sign inStart free
jbellsolutionsjbellsolutions/hermes-super-agent/CLAUDE.md

hermes-super-agent

CLAUDE.md4 starsChanged 7 months agoMITView on GitHub

These rules apply to any Claude Code session running inside this repo. They override defaults. When work matches one of the layer roles in ECOSYSTEM-PLAYBOOK.md, route to that runtime — don't reinvent. Examples: - "Run shell automation for an hour" → OpenClaw runtime, not raw subprocess. - "Browse and extract structured data" → browser-use runtime, not custom Playwright. - "Generate code for a fixture" → Codex CLI / Aider / Claude Code subagents per job tag. - "Look up 10…

The file

We split this CLAUDE.md into the rules it gives an agent

4 rules. About 0 are common, 1 is distinctive, and 3 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈1

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 3

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Very specific:

    Single-state guarantee for channels: Every channel adapter (packages/channels/*) writes conversation memory through Hermes' Vault adapter.

    line 11
  2. Distinctive:

    No comments unless the WHY is non-obvious.

    line 23
  3. Very specific:

    When work matches one of the layer roles in ECOSYSTEM-PLAYBOOK.md, route to that runtime — don't reinvent.

    line 28
  4. Very specific:

    Use the /explain skill (backed by packages/manifest/explain.py).

    line 36

≈ “Common” is an estimate: at least 1 in 5 of the 26,980 CLAUDE.md files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
CLAUDE.md

CLAUDE.md — agent-os project rules

These rules apply to any Claude Code session running inside this repo. They override defaults.

The unbreakable rules

  1. Never edit vendor/. Submodules are vendored upstream OSS (Hermes, OpenClaw, browser-use, Aider, agi-1, awesome-hermes-agent, NemoClaw). Edits there break the auto-updater. If a change is needed: open an upstream PR, or wrap in packages/.
  2. Never start a new framework wrapper. This repo exists because there were 14. Frustration → upstream issue or a thin adapter in packages/runtimes/<name>/. Read ETHOS.md.
  3. Single-state guarantee for channels. Every channel adapter (packages/channels/*) writes conversation memory through Hermes' Vault adapter. No channel-local state. No per-channel side memory.
  4. Smoke tests are non-negotiable for the upgrader. packages/upgrader/streams/<name>.py MUST have a smoke check before a promotion path is added. A bad upstream commit silently promoted is the only failure mode that takes the system down.
  5. Default to Hermes. The router (packages/orchestrator/adapters/job_router.py) handles tagged jobs. Untagged → Hermes itself. Don't overengineer the routing.
  6. Markdown vault is the source of truth. Human-readable in vault/. Supabase mirrors. Never make Supabase the primary.

Code style

  • Python 3.11+, ruff for lint, mypy for types, pytest for tests.
  • TypeScript via pnpm workspace for packages/webapp/ and packages/dashboard/ only.
  • No comments unless the WHY is non-obvious.
  • Functions over classes when reasonable.

Adopted-tool routing

When work matches one of the layer roles in ECOSYSTEM-PLAYBOOK.md, route to that runtime — don't reinvent. Examples:

  • "Run shell automation for an hour" → OpenClaw runtime, not raw subprocess.
  • "Browse and extract structured data" → browser-use runtime, not custom Playwright.
  • "Generate code for a fixture" → Codex CLI / Aider / Claude Code subagents per job tag.
  • "Look up 10 recent articles" → Exa runtime, not browser-use.

When asking the agent-os system itself questions

Use the /explain skill (backed by packages/manifest/explain.py). It walks the system graph in vault/graph/system.yaml. If the answer isn't there, the manifest is incomplete — fix the manifest, not the explanation.

What "done" looks like

  • Tests pass: uv run pytest
  • Lint passes: uv run ruff check && uv run mypy packages/
  • Manifest aggregates: uv run python -m agent_os.manifest.aggregator
  • Smoke green: uv run pytest tests/smoke/
  • For UI changes: pnpm -F webapp dev and verified in browser.

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.

Check your setup, free