neuron exec — Pre-Execution Memory Lookup

neuron exec -- <command> runs a command with a pre-execution memory lookup. Matches print to stderr, filtered through the relevance gate: a result whose top hit has no keyword match at all is rejected, regardless of semantic similarity. If the gate rejects every candidate, neuron exec still prints a line naming the command and how many candidates it rejected, so an empty result is distinguishable from an empty store. The command itself runs with inherited stdio, and its exit code passes straight through.

Usage

neuron exec -- npm test
neuron exec -- git commit -m "message with spaces"

Arguments are spawned as an argv array, so quoting is preserved — a multi-word -m value reaches the underlying command intact. To use shell operators (&&, pipes), pass the whole thing as one quoted string instead:

neuron exec -- "npm run build && npm test"

Which categories get searched

Which categories exec searches, and how many results it returns, comes from pullRules.onExec in neuron.yaml — matched against the command text by regex. See the config reference for the exact field shape.

Limitations

neuron exec is purely informational — it never blocks the command it wraps, even when it finds a highly relevant match. On Claude Code and Codex CLI, neuron init wires a pre-command hook that runs this same lookup automatically on every shell tool call, surfacing a hit as context instead of requiring the agent to remember to type neuron exec itself — see the harness adapter pages for which harnesses get that hook.

Source: README.md “Command execution gets the same treatment”.