Make Your CLI Agent Ready

Publish a structured manifest, predictable machine output, explicit command boundaries, and reproducible evidence so coding agents can call your tool correctly.

Submit a CLI tool Read the agent-ready standard

Agents Need a Clear Operating Contract

An agent cannot fill in implied context like an experienced user. It needs exact installation, headless authentication, stable output, distinguishable errors, and explicit write operations.

Machine output

Provide JSON, YAML, CSV, or field selection where a command result must be parsed.

Safety boundary

Separate preview, read, write, and delete paths and name every approval point.

Agent artifacts

Ship a skill, AGENTS.md, policy, or a manifest that can generate them.

Version evidence

Record the tested version, environment, commands, exit codes, and failure paths.

Publish a Minimum Manifest

A manifest is operating data that an agent and registry can read, not marketing copy. Start with the common install, output, authentication, and representative commands.

Record risk per command. Sources should point to official documentation, repository, installation, or release pages. Never label an unexecuted path as verified.

clifinder.manifest.json

{
  "schemaVersion": 1,
  "name": "example-cli",
  "command": "example",
  "installation": [{ "manager": "brew", "command": "brew install example" }],
  "output": { "structured": true, "formats": ["json"], "flags": ["--json"] },
  "authentication": { "headless": true, "secretEnv": ["EXAMPLE_TOKEN"] },
  "commands": [
    { "command": "example inspect --json", "risk": "R0", "requiresConfirmation": false },
    { "command": "example apply", "risk": "R2", "requiresConfirmation": true }
  ],
  "sources": ["https://example.com/docs/cli"]
}

Maintainer Checklist Before Submission

Make the invocation reproducible

Every item below directly affects whether an agent can use your CLI reliably and safely.

1Provide at least one official installation path with version pinning.
2Make --help and --version stable without a TTY.
3Provide machine-readable output for core read commands.
4Write diagnostics to stderr and task results to stdout.
5Distinguish usage, authentication, network, and target errors with exit codes.
6Document minimum permissions and supported headless credentials.
7Provide dry-run, preview, or plan before remote writes.
8Make delete and irreversible commands obvious and require explicit targets.
9Publish releases and call out breaking output or authentication changes.
10Include reproducible test commands and expected results.

Submit the Official Source and Use Context

Send the tool name, official documentation or repository, and the agent workflow that should be reviewed. Submission alone never makes an entry Verified.

Start a submission