@agent_press/agentpress
v1.0.0
Published
agents.txt for any repo. Tell autonomous AI agents what they're allowed to do, in 60 seconds.
Maintainers
Readme
AgentPress
robots.txttold crawlers what to crawl.llms.txttold LLMs what to read.agents.txttells autonomous AI agents what they're allowed to do.
AgentPress drops an agents.txt and the supporting machine-readable surfaces into any repo in under 60 seconds, so coding agents (Claude Code, Cursor, Devin, Aider, Continue, Replit Agent) know what's allowed, what's prohibited, and what requires human approval.
Install
npm install -g @agent_press/agentpress
agentpress initPython fallback:
pip install agentpress-static
agentpress initWhat agentpress init does
In one interactive minute it adds:
agents.txtat the repo root — the human + machine-readable contract.well-known/agentpress.json— structured entrypoint map for tools.well-known/ai-ingestion.json— crawler/agent ingestion policyagentpress/agent-instructions.json— operating contract with allowed/prohibited/requires-human-approval boundaries- GitHub Action template — CI lint that fails PRs which violate the contract
- README badge — proof of adoption that links back to AgentPress
CLI
| Command | What it does |
|---|---|
| agentpress init | Interactive wizard to drop agents.txt + supporting files |
| agentpress lint . | Validate the agent-readable surfaces; CI-friendly (exit 0/1, --json) |
| agentpress doctor | Health check — verify install, files, schema validity |
| agentpress receipt | Generate a cryptographic proof receipt that an agent ran a check |
The safety contract
Every agents.txt declares three sets of actions:
- Allowed — agents may do these without asking. (read code, run tests, file PRs)
- Requires human approval — agents must pause for sign-off. (schema migrations, billing changes, deploys)
- Prohibited — agents must refuse. (secret exfiltration, deceptive tracking, bypassing 2FA)
This is the contract. AgentPress generates it, validates it, and gives you receipts.
Why now
Coding agents are landing PRs in 2026. Most repos have no machine-readable answer to "what's safe?" agents.txt is the answer — same lineage as robots.txt, sitemap.xml, llms.txt. Adopt early, sleep better.
Add the badge
Once you've run agentpress init, drop the badge into your README:
[](https://github.com/barneywohl/agentpress)→ See docs/BADGE.md for variants. Then submit yours to the registry.
Live endpoints
- Site: https://agentpress.pages.dev/
- npm: https://www.npmjs.com/package/@agent_press/agentpress
- PyPI: https://pypi.org/project/agentpress-static/
- Spec:
docs/AGENTSTXT_SPEC.md - Registry:
registry/(live: https://agentpress.pages.dev/registry/) - GitHub Action:
actions/setup-action - Reference parsers:
@agent_press/core(TS) ·agentpress-core(Python)
License
MIT. See LICENSE.
