Summary

On a gbrain-detected install, setup (~line 1271) runs:

bun run gen:skill-docs:user --host claude

but package.json in v1.53.0.0 defines only gen:skill-docs — there is no gen:skill-docs:user script. So every ./setup on a gbrain machine prints:

gbrain detected — regenerating Claude SKILL.md with brain-aware blocks (~250 token overhead per planning skill)...
error: Script not found "gen:skill-docs:user"
  warning: gen:skill-docs:user failed — run 'bun run gen:skill-docs:user' manually if you want brain-aware blocks

Effect: the brain-aware blocks are silently not injected into planning-skill SKILL.md files on user-local gbrain installs — i.e. the headline behavior the v1.53.0.0 CHANGELOG says shipped never runs from setup.

The CHANGELOG says the script was added (but it wasn't)

From the v1.53.0.0 CHANGELOG:

./setup runs this at the end of install and conditionally regenerates Claude-host SKILL.md with bun run gen:skill-docs:user (added package.json script) ...
Use bun run gen:skill-docs:user for user-local installs.

The alias is absent from package.json, so this reads as done but isn't.

Root cause + one-line fix

scripts/gen-skill-docs.ts already implements the user-local behavior behind a flag:

// Use `bun run gen:skill-docs:user` (which adds --respect-detection) for user-local installs.
const RESPECT_DETECTION = process.argv.includes('--respect-detection');

So the missing alias is just (add after the existing gen:skill-docs entry):

"gen:skill-docs:user": "bun run scripts/gen-skill-docs.ts --respect-detection",

Verified locally — running the underlying command directly regenerates all claude-host SKILL.md with brain-aware blocks and leaves 0 stale:

bun run scripts/gen-skill-docs.ts --respect-detection --host claude

Environment

• gstack v1.53.0.0
• gbrain v0.33.x (postgres), gbrain_local_status: "ok"
• macOS, bun, Claude host

Secondary data point — ship/SKILL.md over the token ceiling (re: your TODOS.md rebaseline)

While verifying, the generator flags ship/SKILL.md over the 160000-byte (~40K token) ceiling. Measured both ways:

So the brain block is only ~600 tokens of the overage — the bulk is the v1.53 redaction-engine content now in the ship template. Sharing in case it helps the rebaseline the CHANGELOG mentions tracking.