Platform Health
Loading platform health data...
How does this work? (the cogs & how to grease them)
What it does
EduScan finds defects in the codebase (broken links, undeclared catalog files, security issues, etc). Most are surfaced for human triage. A few classes are mechanical enough to fix autonomously — and this system runs those fixes when you explicitly authorize it. Nothing happens automatically without operator action.
The 4 cogs (and how each gets greased)
Grease: run
node nexus.js full --publish
[2]
Promotion — a rule must be in AUTO_FIX_ELIGIBLE_RULES in _tools/nexus/publish.js for items to enter _auto_fix_queue.Grease: edit the file, add a rule code, re-run nexus [3] Master toggle (above) — global kill switch. CLI commands return "GATE: master toggle is OFF" until enabled.
Grease: click the slider [4] Per-template enable (list below) — per-rule gate. Even with master ON, each template must be individually enabled.
Grease: click
enable next to a rule
To make something get fixed (full workflow)
- Edit
_tools/nexus/publish.js— add a rule (e.g.'CAT-002') toAUTO_FIX_ELIGIBLE_RULES - Run
cd _tools/nexus && node nexus.js full --publish— grouped items appear in_auto_fix_queue - Open this panel — toggle master ENABLED + click
enablenext to the rule - Get an itemId:
node nexus.js autofix-statusor query Firestore - Run
node nexus.js autofix-apply <itemId>— claims, applies, validates, resolves (one file per call) - Review the diff,
git commit,firebase deploy --only hosting
What runs autonomously vs what needs you
Always automatic (no operator needed): Nexus rescan auto-resolves items whose defect disappeared (e.g., you fixed it manually); auto-reopens items whose defect reappeared; dead-claim reaper releases stuck claims after 10min; saturation alarm fires Discord webhook + Pulse banner if queue grows past 50 with 20%+ 24h growth.
Needs operator action: promoting a rule (step 1 above), enabling toggles, running autofix-apply, reviewing diffs, committing, deploying.
Safety — to STOP everything
Click master toggle DISABLED. Every CLI agent command immediately returns exit 2 with a GATE error. In-flight transactions complete; the dead-claim reaper releases anything stuck within 10min. Per-template enables are preserved (re-enabling later doesn't lose state).
Why three gates instead of one
Per Nancy review: each gate guards a different failure mode. Promotion gates "can this rule mechanically be fixed." Master toggle gates "is anyone watching." Per-template gates "are we comfortable with THIS specific class of fix." Defense in depth.
Full architecture: _docs/features/SELF_HEALING_PIPELINE.md · Template contract: _tools/nexus/fix-templates/CONTRACT.md
Quiz pass rate analytics require aggregation Cloud Function (planned for v2)
--
Lab completion rate analytics require aggregation Cloud Function (planned for v2)
--
Student behavior intelligence will feed back into triage queue automatically
v2