You do not need full coverage before Agents Remember becomes useful. Start small, then let onboarding grow as work touches new files.
Configure the Agents Remember MCP server, then request:
runtime_install()
Expose skills for your harness using the relevant install guide.
Ask the agent to run c-00-initialize-memory-repo for the target repository.
Default internal memory creates:
<repo>/ar-memory/
onboarding/
docs/
system/
settings.md
settings.json
sources.md
tools.md
Do not create onboarding content by hand before the memory root exists. The c-00-initialize-memory-repo skill owns the scaffold; the c-03-repo-bootstrap skill owns onboarding bootstrap.
Review <repo>/ar-memory/system/settings.json.
Start with a small eligible surface:
{
"version": 1,
"onboarding": {
"storage": {
"mode": "repo-sidecar"
},
"pathRules": {
"include": {
"paths": ["README.md", "docs/**", "src/**"],
"fileTypes": [".md", ".py", ".ts", ".tsx"]
},
"exclude": {
"paths": ["node_modules/**", "vendor/**", "dist/**", "build/**", ".env", ".env.*"],
"fileTypes": [".png", ".zip"]
}
}
}
}
See Path Rules for the fuller exclusion baseline.
Ask the agent to run c-03-repo-bootstrap.
A repo-level overview.md is enough to start. Larger repositories can add route-local overviews under the mirrored onboarding hierarchy when a package, module, or source slice needs durable context.
When a task touches a file, the agent should:
c-08-ar-coordination-context-resolver skillc-02-memory-quality-control skillc-05-create-or-update-onboarding-files skillThis avoids a giant up-front documentation project. The first task on a file pays the onboarding cost; later tasks benefit.