Yanez Agent Authorization
View as Markdown Open in Claude Open in ChatGPT llms.txt

Use with AI agents

Every page on this site has a raw-markdown twin, and a compact agent reference lives at llms.txt. Point a chat assistant or a coding agent at either and it can help you wire agent authorization into your stack. There is nothing to install.

Open llms.txt

The file follows the llms.txt convention: a short summary, the facts an agent must not get wrong (who verifies the receipt, the HTTP routes, the credential rule), and links to every doc, SDK README, and example as raw markdown.

On every page

The toolbar at the top of every page is a one-click handoff:

  • Copy page copies the page’s raw markdown to your clipboard. Paste it into any chat.
  • View as Markdown opens the raw .md source so you can save it or pipe it somewhere.
  • Open in Claude and Open in ChatGPT start a new chat with a prompt that points the assistant at the page.
  • llms.txt opens the agent reference.

If your assistant can fetch URLs, prefer that over pasting: it stays in sync when the docs change.

For a chat session

Paste this into Claude, ChatGPT, Gemini, or any other assistant to seed the session:

You are helping me integrate Yanez agent authorization: an AI agent requests
verifiable human approval for a sensitive action, and the action executor verifies
the signed receipt before acting.

Start by reading https://yanez-compliance.github.io/yanez-agent-authorization/llms.txt,
then follow its links to the pages relevant to my task. Then help me with my task.

For a coding agent

Coding agents read a rules file from your repository on startup. Add the following block to whichever file your agent reads:

  • Claude Code: CLAUDE.md
  • Codex, OpenCode: AGENTS.md
  • Cursor: .cursor/rules/yanez.md
  • GitHub Copilot: .github/copilot-instructions.md
# Yanez agent authorization: agent context

Before touching Yanez agent-authorization code, read
https://yanez-compliance.github.io/yanez-agent-authorization/llms.txt and the pages
it links for the task at hand.

Rules that must hold in any code you write:

- A receipt authorizes nothing by itself. The action executor verifies the signature,
  verifies the approver's own signature over the decision, compares the signed terms
  with the proposed action structurally, applies
  its own freshness policy, and consumes single-use receipts. Gate single-use actions
  on `consumed_now: true`, never on `valid: true` alone.
- The `yak_` agent key comes from configuration (`YANEZ_AGENT_API_KEY` or a secret
  manager), never from prompts, tool arguments, command-line flags, or logs. Never
  pass it to a sub-agent.
- Send a random `Idempotency-Key` on every create and reuse it verbatim on retries.
  Never derive it from the request content.
- Stop on `rejected` or `expired`. Never create replacement requests in a loop.

Fetch the docs as markdown

Every page’s raw markdown lives in the repository. Swap the site URL for the raw GitHub URL:

BASE=https://raw.githubusercontent.com/yanez-compliance/yanez-agent-authorization/main

curl "$BASE/docs/integration-options.md"              # choosing an integration path
curl "$BASE/docs/http-quickstart.md"                  # the HTTP routes
curl "$BASE/docs/terms.md"                            # every field of what the human approves
curl "$BASE/docs/receipts.md"                         # receipt claims, signing, key rotation
curl "$BASE/docs/action-enforcement.md"               # the action executor's contract
curl "$BASE/openapi/agent-authorization.openapi.yaml" # full request and response schemas

# The agent reference
curl https://yanez-compliance.github.io/yanez-agent-authorization/llms.txt