distill-project.mjs hardcodes ~/.npm-global/bin/ruflo — fails on Homebrew/other-prefix installs

[Tigon32]

What I observed

scripts/distill-project.mjs resolves ruflo via a hardcoded path with no fallback:

const RUFLO = process.env.RUFLO_BIN || path.join(HOME, '.npm-global/bin/ruflo');

On a machine where ruflo is installed globally via Homebrew (/opt/homebrew/bin/ruflo → /opt/homebrew/lib/node_modules/ruflo/bin/ruflo.js, confirmed working: ruflo v3.34.0), ~/.npm-global/bin/ruflo does not exist, so the script dies immediately:

ruflo is not at ~/.npm-global/bin/ruflo — install it with `npm i -g ruflo@latest`

...even though ruflo is installed and on PATH. Had to work around it with RUFLO_BIN=/opt/homebrew/bin/ruflo node distill-project.mjs.

Why it matters

This is the same class of bug as #18 (stack-sync.mjs hardcoded prefix, fixed) — and health-repair.mjs already carries the correct fix for this exact problem, with its own comment explaining why:

/**
 * Find ruflo HONESTLY.
 *
 * This was hardcoded to `~/.npm-global/bin/ruflo` — Stuart's prefix, not everyone's. ...
 * Telling someone their tool is missing when it is on their PATH is the product lying...
 */
function resolveRuflo() {
  const preferred = path.join(HOME, '.npm-global/bin/ruflo');
  if (fs.existsSync(preferred)) return preferred;
  const which = spawnSync('sh', ['-lc', 'command -v ruflo'], { encoding: 'utf8', timeout: 10_000 });
  const found = String(which.stdout || '').trim().split('\n')[0];
  return found && fs.existsSync(found) ? found : null;
}

distill-project.mjs just never got this fix applied — it's a straight port of resolveRuflo() from health-repair.mjs.

Suggested fix

Replace distill-project.mjs's RUFLO const with health-repair.mjs's resolveRuflo() (or extract it to a shared helper — at least 10 scripts in this repo hardcode ~/.npm-global independently: agentdb-fleet-doctor.mjs, capability-registry.mjs, calibrate-router.mjs, learning-replay-execution.mjs, onboarding-console.mjs, etc. — a shared resolve-ruflo.mjs would prevent this exact regression from recurring script-by-script).

Environment

• macOS, Node installed via Homebrew, ruflo installed globally via Homebrew (/opt/homebrew/bin/ruflo → v3.34.0)
• ruvnet-brain v4.0.4 (marketplace checkout, commit 9de40a5)

[stuinfla]

🤖 Automated acknowledgment — received and opened. The maintainer has been paged and this is being worked. The next update here will be a fix, findings, or the maintainer in person.

[stuinfla]

You were right on every point, including the diagnosis — and thank you for naming health-repair.mjs as the file that already had the answer. That's what made this fixable properly instead of patched again.

Fixed in #110, but not the way you suggested, and I want to be straight about why. There were two existing resolvers, not one:

• scripts/health-repair.mjs (2026-07-21) — preferred path, then sh -lc 'command -v ruflo'
• scripts/capability-registry.mjs:177 (2026-07-22) — preferred path, then a plain PATH walk, no shell

The second is the first one hardened: -l sources the user's entire profile as the price of answering "where is ruflo?". Copying health-repair's body into a shared module would have re-introduced, into shared code, a bug this repo had already fixed one file over. So the hardened form was extracted to plugin/scripts/ruflo-bin.mjs and both call sites now import it.

Resolution order, each step earning its place: RUFLO_BIN (authoritative, returned as given), then ~/.npm-global/bin/ruflo, then a PATH walk (your case), then null — said plainly, never a guessed path the user gets blamed for. Windows .cmd handled.

You also identified the shape correctly: this was #18 again. Seven files carried that hardcode; two are fixed, and the remaining five (health-repair, agentdb-fleet-doctor:25, onboarding-console:2161, nightly-wrapper.sh:81,84, learning-replay-contract) are tracked as the same class rather than five future reports.