Skip to content
Aftergraph Knowledge Plane

Agent Guide

This portal is a compiler, never an owner: point your agent at the compiled surfaces below and have it verify claims against the source commits each page cites. Start agents at the Developers overview and the Quickstart before going deeper.

The four surfaces to give your agent

Section titled “The four surfaces to give your agent”
  1. Index — https://docs.aftergraph.org/llms.txt: the whole plane in one page, with the build commit and freshness boundary stamped at the top.
  2. Scoped indexes — one per area, same host under /llms/: llms/platform.txt, llms/research.txt, llms/standards.txt, llms/aie.txt, llms/agents.txt.
  3. Context packs — https://docs.aftergraph.org/context/index.json: per-page portable bundles (one JSON pack per route).
  4. MCP boundary — Docs MCP (read-only): the read-only tools over the built site (count lives with the server, not this page); every answer carries site_commit, and answers must not be treated as fresher than that commit.

Site search is Starlight’s built-in Pagefind search (noted in astro.config.mjs); agents working from a checkout can also use it, but the four surfaces above are the scripted entry points.

No official SDK or unified CLI ships with the plane yet, so prompts must not ask agents to install or import one — curl or server-side calls against the API Reference, plus the Reference and How-to guides, are the documented path.

Template 1 — orient on the current docs

Section titled “Template 1 — orient on the current docs”
You are working with the Aftergraph Knowledge Plane docs (a compiler over
canonical repos, not the owner of any spec or claim).


1. Fetch https://docs.aftergraph.org/llms.txt and note the site commit and
   freshness boundary at the top. Substitute <SITE_COMMIT> below with it.
2. Treat no page as fresher than <SITE_COMMIT>. Every page's provenance block
   cites its source repo and commit — verify technical claims there, not from
   memory.
3. My question (against docs slice <DOCS_SLICE>): <YOUR QUESTION>

Template 2 — scoped research with context packs

Section titled “Template 2 — scoped research with context packs”
Scope: <platform | research | standards | aie | agents> (docs slice <DOCS_SLICE>).


1. Fetch https://docs.aftergraph.org/llms/<SCOPE>.txt and follow only the links
   it lists.
2. For any page you rely on, also fetch its context pack via
   https://docs.aftergraph.org/context/index.json and cite the pack's
   repository + commit for each claim.
3. If a claim needs the read-only MCP, use only the tools documented at
   /developers/mcp-boundary/ and report the site_commit from each response.
4. Task: <YOUR TASK>

Template 3 — verify before you assert

Section titled “Template 3 — verify before you assert”
Before stating any Aftergraph contract, claim, or endpoint:


1. Check it in the Reference (/developers/reference/) or the API Reference
   (/developers/api-reference/).
2. Confirm the page provenance (source commit) is within docs slice
   <DOCS_SLICE> / site commit <SITE_COMMIT>.
3. If the docs say no SDK, CLI, or endpoint exists for it, report that
   instead of inventing one.

<SITE_COMMIT>, <DOCS_SLICE>, <SCOPE>, and <YOUR QUESTION> / <YOUR TASK> are placeholders — fill them per run, never hardcode a commit SHA or version number into a saved prompt.

Was this page helpful?