examples/ contains exactly one file — claude-code-stop-hook.json — so the only documented way to wire this memory layer into an agent is Claude Code's hook system. The retrieval path itself (brain_ask.py) is a plain script with no dependency on any particular harness, so this is a documentation gap, not an architectural one.
What "done" looks like
One more example that wires brain_ask.py into a different agent runtime, plus a short examples/README.md saying what each example is and what it assumes. Any of these is fine — pick the one you actually run:
- OpenAI Codex CLI /
AGENTS.md
- a plain MCP server exposing recall as a tool
- a bare shell wrapper for people who use neither, so the fallback is documented
Constraints, so this stays reviewable
Keep it to one runtime. An example that was never executed is worse than no example, so paste the actual output you got — even if it is a short one — into the PR description. If you hit a case where brain_ask.py assumes something about its caller that is not written down anywhere (env var, cwd, ordering), that assumption is the more valuable half of this issue: say so, and we will either fix the script or document it.
What you need
A folder of markdown notes and the ability to run brain_ask.py once. The vault does not have to be large — a handful of notes with [[wikilinks]] between them is enough to show recall working.
Comment "claiming this" and it is yours.
examples/contains exactly one file —claude-code-stop-hook.json— so the only documented way to wire this memory layer into an agent is Claude Code's hook system. The retrieval path itself (brain_ask.py) is a plain script with no dependency on any particular harness, so this is a documentation gap, not an architectural one.What "done" looks like
One more example that wires
brain_ask.pyinto a different agent runtime, plus a shortexamples/README.mdsaying what each example is and what it assumes. Any of these is fine — pick the one you actually run:AGENTS.mdConstraints, so this stays reviewable
Keep it to one runtime. An example that was never executed is worse than no example, so paste the actual output you got — even if it is a short one — into the PR description. If you hit a case where
brain_ask.pyassumes something about its caller that is not written down anywhere (env var, cwd, ordering), that assumption is the more valuable half of this issue: say so, and we will either fix the script or document it.What you need
A folder of markdown notes and the ability to run
brain_ask.pyonce. The vault does not have to be large — a handful of notes with[[wikilinks]]between them is enough to show recall working.Comment "claiming this" and it is yours.