agentleFS
relvo-labsrelvo-labs/agent-runtime/AGENTS.md

agent-runtime

AGENTS.md0 starsChanged 47 days agoApache-2.0View on GitHub

Operating contract for automated and human contributors to relvo-labs/agent-runtime. .agents/skills/ is the only skill root in this repository. Do not create .claude/skills/, .cursor/rules/, skills/, docs/skills/ or any other parallel instruction root. tools/skills/validate.ts fails the build if a second active root appears. If another tool needs to see these skills, symlink into the tool's expected location from outside version control — never the reverse, and never inside .agents/skills/ (symlinks under the skill root are a validation error). Before mutating any area…

The file

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

5 rules. About 0 are common, 2 are distinctive, and 3 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈2

    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:

    Do not create .claude/skills/, .cursor/rules/, skills/, docs/skills/ or any other parallel instruction root.

    line 9
  2. Distinctive:

    Before mutating any area of this repository, read the skill that owns it.

    line 17
  3. Distinctive:

    Use that exact pnpm version.

    line 40
  4. Very specific:

    Every third-party version lives in the catalog: block of pnpm-workspace.yaml.

    line 42
  5. Very specific:

    All eight public packages have published, immutable 0.2.0 and 0.3.0 versions.

    line 67

≈ “Common” is an estimate: at least 1 in 5 of the 40,246 AGENTS.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
AGENTS.md

AGENTS.md

Operating contract for automated and human contributors to relvo-labs/agent-runtime.

1. Canonical skill root

.agents/skills/ is the only skill root in this repository.

Do not create .claude/skills/, .cursor/rules/, skills/, docs/skills/ or any other parallel instruction root. tools/skills/validate.ts fails the build if a second active root appears. If another tool needs to see these skills, symlink into the tool's expected location from outside version control — never the reverse, and never inside .agents/skills/ (symlinks under the skill root are a validation error).

2. Read before write

Before mutating any area of this repository, read the skill that owns it.

If you are about to change…Read first
protocol schemas, wire DTOs, JSON Schema, state machinesruntime-contract-evolution
a provider SPI implementation or capability descriptorprovider-adapter-development
workspace acquisition, leases, cleanupworkspace-lifecycle
runtime commands, side effects, cleanup, subscriptionsruntime-lifecycle-coordination
package boundaries, the dependency DAG, tsconfig graphpackage-architecture
anything exported from a package entry pointpublic-api-evolution
exports maps, build output, tarball contentspackage-artifact-validation
dependencies, the lockfile, install scriptspnpm-supply-chain
version intent for a published packagechangesets-release
CI workflows or the local gatelocal-ci-parity
the release workflow or tools/releasenpm-release

.agents/skills/INDEX.md is generated. Run pnpm skills:index after adding or editing a skill; pnpm skills:check verifies it is not stale.

3. Toolchain

  • Node: use nvm use (reads .nvmrc, currently 24.20.0).
  • Package manager: pnpm, pinned exactly by packageManager in the root package.json. Use that exact pnpm version; Corepack is not assumed to exist. Do not use npm or yarn to install this workspace's dependencies.
  • Every third-party version lives in the catalog: block of pnpm-workspace.yaml. Packages must reference catalog:, never a literal range.

4. The gate

One command reproduces CI locally:

pnpm gate

CI runs the identical step list (tools/repo/gate.ts is the single source of truth for both). No gate step may require provider credentials, network access to a model provider, or a publish token — including the release step, which checks the release workflow's structure without being able to publish anything.

5. Hard boundaries for this foundation

  • @relvo-labs/agent-runtime must never import a concrete provider adapter. It composes neutral SPIs only. pnpm dag:check enforces this mechanically.
  • Provider-native identifiers, handles and checkpoints must not appear in public DTOs.
  • existing workspaces are borrowed. Nothing in this repository may delete, reset or otherwise destructively mutate a borrowed workspace.
  • Zod schemas are authoritative. TypeScript types are inferred from them; JSON Schema is generated from them. Never hand-write a type or a .json schema that duplicates one.
  • All eight public packages have published, immutable 0.2.0 and 0.3.0 versions; neither version may ever be named in a dispatch again. A 2026-09-27 registry observation found latest at 0.3.0 for all eight. Provider 0.4.0 is not published. Publication happens only through .github/workflows/release.yml: manual workflow_dispatch on main, an explicitly typed name@version scope, a credential-free gate and fail-closed preflight first, then an npm-release environment approval. It refuses to run while any version intent is pending, and it never chooses a version. Read .agents/skills/npm-release/SKILL.md and docs/release.md before changing it; an agent may not initiate a publication.

6. Provenance and licensing

This project is Apache-2.0. Material adapted from third-party MIT sources is attributed in NOTICE and in the ## Provenance section of the skill that carries it, always with a pinned upstream revision. Sources with absent or ambiguous licensing are recorded as "reviewed, not copied" and their text must not be incorporated.

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.