AGENTS.md Generator Integration
June 13, 2026 ยท View on GitHub
This document makes the ID and agentsgen relationship explicit at the repo-contract level.
Boundary
IDowns portable human context, freshness, privacy, and handshakes.agentsgenowns repo-scoped agent context and command manifests.SETcan orchestrate both layers, but the bridge should also work withoutSET.
Repo-local handoff contract
When a repository runs agentsgen pack --autodetect, it can publish docs/ai/id-context.json.
That file is the repo-local handoff surface for ID-compatible workflows. It should point at the repo artifacts that matter most for execution:
AGENTS.mdRUNBOOK.mdagents.entrypoints.jsondocs/ai/how-to-run.mddocs/ai/how-to-test.mddocs/ai/architecture.mddocs/ai/data-contracts.md- optional
repomap,graph,agents.knowledge.json, and proof-loop task artifacts when present
Recommended flow
- Load the owner profile and handshake from
ID. - If present, load
profiles/<owner>/soul.mdas the fast human-context bootstrap layer. - Load
docs/ai/id-context.jsonfrom the active repository. - Merge human context from
IDwith repo context fromagentsgen. - If orchestration is present, let
SETcallpre_taskandweekly_reviewaround the repo workflow.
Non-goals
docs/ai/id-context.jsondoes not replaceprofile.core.md,profile.extended.md, or interop exports.soul.mddoes not replace the canonical source profile files; it is a compact derived layer.agentsgendoes not own user preferences or privacy policy.IDdoes not replace repo-local contracts such asAGENTS.md.
Practical summary
Use ID for the human.
Use agentsgen for the repository.
Use SET when you want one orchestration entrypoint for both.