Skip to content

GitHub Copilot CLI: Best-Effort Recall at Session Start

GitHub Copilot CLI is a best-effort adapter: neuron init wires a session-start hook, injecting recall once when a session begins. There is no per-turn hook point on this harness, so a memory recorded mid-session won’t surface again automatically until the next session starts.

Repo-level hooks are written to .github/hooks/neuron.json (project), mirroring the committed/local/user-global scope choice neuron init asks for on every harness.

neuron init writes .mcp.json, the same file Claude Code uses, so Copilot CLI can call neuron_remember, neuron_recall, and neuron_query_exec directly via MCP.

Copilot CLI gets agentStop, which forces a continuation with a reminder if nothing has been recorded to memory by the end of a turn — per Copilot CLI’s own documented agentStop semantics, not independently verified live the way Claude Code’s Stop hook was.

Copilot CLI’s preToolUse hook is documented as permission/gating-only — it has no context-carrying field at all, unlike Claude Code and Codex CLI’s PreToolUse. That means the pre-command pre-execution memory lookup available on those two harnesses has no equivalent here: this is a structural ceiling of Copilot CLI’s own hook model, not a gap a future neuron version can close. Copilot CLI keeps the neuron exec -- <command> CLI wrapper as its permanent path for pre-execution lookups, driven by instruction rather than by hook.

Source: ADR 0014 — Recall Adapter Architecture, 2026-08-10 amendment.