Running Drive Automation

TL;DR: Chain 1 (n8n) classifies + routes incoming Drive files automatically. Chain 2 (10 CrewAI crews in agent_council.py) performs deep per-file evaluation and writes structured memos. Supabase logs every action. Smoke tests gate production runs.

⚠ DEGRADED STATE ACTIVE — n8n MCP is CONNECTION_ISSUE. Chain 1 automatic trigger is unavailable via MCP. See DEGRADED MODE section before proceeding. Chain 2, Supabase, and Google Drive components are fully operational.

FAIL_IF — n8n MCP Unavailable (Chain 1)

This block fires before any Chain 1 operation is attempted.

IF n8n MCP status = CONNECTION_ISSUE:
  STOP. Do NOT attempt MCP tool calls to trigger or monitor Chain 1.
  Do NOT simulate Chain 1 execution or fabricate routing results.

  Chain 1 manual trigger path:
    1. Navigate to https://krosebrook.app.n8n.cloud
    2. Workflows → "Drive Router Chain 1"
    3. Click "Execute workflow" (top-right button, lightning bolt icon)
    4. Monitor execution in the Executions tab on the left sidebar
    5. Verify results in Supabase: SELECT * FROM file_audit ORDER BY
       created_at DESC LIMIT 20;

  Chain 2, Supabase queries, Google Drive reads, and drift checks are
  UNAFFECTED by this failure. Proceed with those operations normally.

  "Just do it" override: If user explicitly says "try the n8n MCP anyway,"
  attempt the call once and report the exact error. Do not retry silently.

DEGRADED MODE — n8n MCP: CONNECTION_ISSUE

Component	Status	Notes
Chain 1 — Drive Router (auto trigger via MCP)	❌ UNAVAILABLE	n8n MCP CONNECTION_ISSUE; use manual trigger at krosebrook.app.n8n.cloud
Chain 1 — Execution log monitoring via MCP	❌ UNAVAILABLE	Check manually: n8n.cloud → Executions
Chain 1 — Webhook trigger (direct curl)	✅ OPERATIONAL	curl to CHAIN1_WEBHOOK_URL still works; see Chain 1 section
Chain 2 — Agent Council (agent_council.py)	✅ OPERATIONAL	Runs locally via Python; no n8n dependency
Chain 2 — Drift Check (--drift-check)	✅ OPERATIONAL	Runs locally; reads Drive via API, writes to Supabase
Supabase audit log (file_audit, council_memos)	✅ OPERATIONAL	Supabase MCP and direct SQL both available
Google Drive reads (classify, list, fetch)	✅ OPERATIONAL	Google Drive MCP unaffected
Smoke Tests (partial)	⚠ PARTIAL	`[PASS] Chain 1 webhook reachable` will FAIL if n8n is down; all other checks pass

In DEGRADED MODE, the recommended workflow is:

Trigger Chain 1 manually (see manual path in FAIL_IF above), OR trigger via direct webhook curl (see Chain 1 section)
Run Chain 2 Council as normal (fully operational)
Check audit results directly in Supabase
Run smoke tests and expect exactly 1 FAIL (Chain 1 webhook reachable); treat this as expected, not a blocker for Chain 2 operations

Decision Tree

What do you need?
  ├─ Route a batch of new files
  │    ├─ n8n MCP operational ──────────────► Chain 1 — see §Golden Path: Chain 1
  │    └─ n8n MCP CONNECTION_ISSUE ──────────► Chain 1 MANUAL — see §Manual Trigger
  │
  ├─ Deep evaluation on files ────────────────► Chain 2 Council — FULLY OPERATIONAL
  │                                             see §Golden Path: Council
  │
  ├─ Check for file drift ────────────────────► Drift Watcher — FULLY OPERATIONAL
  │                                             see §Golden Path: Drift
  │
  ├─ Verify pipeline health ──────────────────► Smoke Tests — PARTIAL (1 expected FAIL)
  │                                             see §Smoke Test Protocol
  │
  ├─ View what happened to a file ────────────► Supabase audit — FULLY OPERATIONAL
  │                                             see references/supabase-schema.md
  │
  ├─ Understand crew roles/behavior ──────────► FULLY OPERATIONAL
  │                                             see references/crews.md
  │
  ├─ Configure n8n Chain 1 ───────────────────► Manual only while MCP down
  │                                             see references/n8n-config.md
  │                                             and §Manual Trigger below
  │
  └─ Apply asyncio throughput fix ────────────► FULLY OPERATIONAL
                                               see references/asyncio-fix.md

System Map

Component	Location	Purpose	MCP Status
Chain 1 — Drive Router	`krosebrook.app.n8n.cloud` → `chain1_drive_router.n8n.json`	Watches Drive, classifies files, routes to destination folders	❌ MCP UNAVAILABLE — manual trigger available
Chain 2 — Agent Council	`personal/drive-automation/agent_council.py`	10 CrewAI crews evaluate each file; produces structured memo	✅ OPERATIONAL
Supabase Audit	`personal/drive-automation/supabase_migration.sql`	Schema: `file_audit` + `council_memos` tables	✅ OPERATIONAL
Smoke Tests	`personal/drive-automation/smoke_test.py`	End-to-end validation; 1 FAIL expected in degraded mode	⚠ PARTIAL
Drift Watcher	Embedded in `agent_council.py`	Detects files that moved or changed classification since last run	✅ OPERATIONAL

⚠ Lane rule: Personal drive automation uses a dedicated Supabase project. Never route through oorstbwcydxwiojrsoim (Loxihart). Verify .env SUPABASE_URL before any Supabase op.

⚠ Large-file rule: Files >3KB (agent_council.py 22K, chain1_drive_router.n8n.json 32K, supabase_migration.sql 7.6K, smoke_test.py 7.9K) must be pushed via git push locally — Zapier's github_create_or_update_file times out above 3KB.

Golden Path: Chain 1 (Route New Files)

CONNECTION CHECK — n8n MCP: CONNECTION_ISSUE Claude cannot trigger or monitor Chain 1 via MCP. Use direct webhook curl (path A, below) OR the manual UI trigger (path B). Both paths work regardless of MCP status.

Path A: Direct Webhook (n8n MCP not required)

# 1. Confirm Chain 1 is active in n8n dashboard
#    https://krosebrook.app.n8n.cloud → Workflows → "Drive Router Chain 1"
#    Status toggle must be: Active

# 2. Trigger via webhook (see references/n8n-config.md for full URL + auth headers)
curl -X POST <CHAIN1_WEBHOOK_URL> \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer <N8N_WEBHOOK_TOKEN>" \
  -d '{"trigger": "manual", "folder": "<DRIVE_INBOX_FOLDER_ID>"}'

# 3. Monitor execution in n8n Executions tab (manual, since MCP unavailable)
#    https://krosebrook.app.n8n.cloud → Executions → filter by "Drive Router Chain 1"

# 4. Verify results in Supabase file_audit table (Supabase MCP: OPERATIONAL)
#    SELECT * FROM file_audit ORDER BY created_at DESC LIMIT 20;

→ Full webhook config, auth, node map: references/n8n-config.md

Path B: Manual UI Trigger (n8n MCP not required)

1. Navigate to https://krosebrook.app.n8n.cloud
2. Workflows → "Drive Router Chain 1"
3. Click "Execute workflow" button (lightning bolt, top-right of editor)
4. Executions tab → watch the live run populate
5. When complete: verify in Supabase (see Path A, step 4)

Golden Path: Chain 2 (Agent Council) — FULLY OPERATIONAL

# Pre-flight: run smoke tests first — NEVER skip
python personal/drive-automation/smoke_test.py
# Expected in DEGRADED MODE:
#   [FAIL] Chain 1 webhook reachable  ← expected, non-blocking in degraded mode
#   [PASS] all other checks
# If any check other than Chain 1 webhook FAILs: stop and debug before proceeding.

# Build manifest
MANIFEST='{
  "run_id": "'$(date +%Y-%m-%d-%H%M)'",
  "items": [
    {"file_id": "...", "name": "...", "mime_type": "...", "path": "..."}
  ]
}'

# POST to council endpoint
curl -X POST https://<COUNCIL_HOST>/council \
  -H "Content-Type: application/json" \
  -d "{\"manifest\": $MANIFEST}"

# Poll Supabase for memo results (Supabase MCP: OPERATIONAL)
# SELECT * FROM council_memos WHERE run_id = '<run_id>';

If batch > 5 files: Apply the asyncio parallel fix first → references/asyncio-fix.md Council crew behavior: → references/crews.md Or use the script wrapper: bash scripts/run_council.sh --manifest <path_to_manifest.json>

Golden Path: Drift Check — FULLY OPERATIONAL

# Drift watcher runs inside agent_council.py when invoked with --drift flag
python personal/drive-automation/agent_council.py --drift-check

# Output: files flagged with [DRIFT] have changed classification or path since last run
# Review each DRIFT entry before re-routing — don't auto-re-route drift without review
# Results written to Supabase file_audit (Supabase MCP: OPERATIONAL for review)

Smoke Test Protocol

Run before every production council batch. In DEGRADED MODE, 1 FAIL is expected.

cd personal/drive-automation/
python smoke_test.py

# Expected output — DEGRADED MODE (n8n MCP CONNECTION_ISSUE):
# [FAIL] Chain 1 webhook reachable  ← EXPECTED in degraded mode; non-blocking for Chain 2
# [PASS] Supabase connection (personal project)
# [PASS] file_audit table exists
# [PASS] council_memos table exists
# [PASS] agent_council.py imports cleanly
# [PASS] Sample crew instantiation
# [PASS] Drift watcher init

# Expected output — FULLY OPERATIONAL (n8n MCP connected):
# [PASS] Chain 1 webhook reachable
# [PASS] Supabase connection (personal project)
# [PASS] file_audit table exists
# [PASS] council_memos table exists
# [PASS] agent_council.py imports cleanly
# [PASS] Sample crew instantiation
# [PASS] Drift watcher init

# STOP conditions:
#   Any FAIL other than "Chain 1 webhook reachable" → debug before proceeding
#   "Chain 1 webhook reachable" FAIL in FULLY OPERATIONAL mode → n8n is actually down

Self-Verification Checklist

Before invoking the council on a real batch:

n8n MCP status checked. If CONNECTION_ISSUE, FAIL_IF block acknowledged. FAIL: Chain 1 triggered via MCP while n8n MCP is CONNECTION_ISSUE — use direct webhook curl or manual UI trigger per §Golden Path: Chain 1.
personal/drive-automation/ files are current in GitHub. FAIL: Any of the 4 large files (agent_council.py, chain1_drive_router.n8n.json, supabase_migration.sql, smoke_test.py) not pushed locally via git push — do NOT use Zapier for files >3KB.
Supabase migration has run on personal project. FAIL: file_audit or council_memos tables missing in personal Supabase project → run supabase_migration.sql against the personal project before proceeding.
.env SUPABASE_URL points to personal project (NOT oorstbwcydxwiojrsoim). FAIL: SUPABASE_URL contains oorstbwcydxwiojrsoim → lane isolation violation; stop immediately and correct the env var before any Supabase write.
smoke_test.py passes with 0 unexpected failures. FAIL: Any PASS→FAIL other than "Chain 1 webhook reachable" in degraded mode → debug the specific failing component before running production batch.
Chain 1 n8n workflow is Active (manual UI check). FAIL: Chain 1 status is Inactive in n8n Cloud UI → activate before triggering or the webhook will be ignored regardless of trigger method.
If batch > 5 files: asyncio parallel fix applied. FAIL: Batch >5 files submitted to Council without asyncio fix → ThreadPoolExecutor default will exceed RPM cap; apply fix per references/asyncio-fix.md.
COUNCIL_HOST env var is set and /council endpoint is reachable. FAIL: curl -I https://<COUNCIL_HOST>/council returns non-200 → Chain 2 cannot receive the manifest; debug Council host before proceeding.

Reference Layer

File	Load when...
`references/crews.md`	Running council; need to understand crew roles, tweak behavior, or debug a specific crew output
`references/n8n-config.md`	Configuring Chain 1, debugging webhook, updating trigger conditions or node logic; required for CHAIN1_WEBHOOK_URL and auth headers
`references/supabase-schema.md`	Querying audit log, running migrations, understanding table structure or RLS
`references/asyncio-fix.md`	Batch > 5 files; applying or reverting parallel throughput improvement

Key Constraints Summary

n8n MCP CONNECTION_ISSUE — Chain 1 auto-trigger via MCP is unavailable. Use direct webhook curl or manual UI trigger at krosebrook.app.n8n.cloud. Chain 2, Supabase, Drive, and drift checks are unaffected.
Supabase lane isolation — personal ≠ Loxihart (oorstbwcydxwiojrsoim). Cross-contamination breaks audit trails.
RPM cap — 10 crews × LLM calls. Default model: Haiku. Cap ThreadPoolExecutor at 5 workers.
Smoke-gate — smoke_test.py must fully PASS before any production run (in degraded mode: all checks except "Chain 1 webhook reachable"). One unexpected FAIL = debug first.
Git-push for large files — 4 files exceed Zapier's 3KB limit. Always push locally.
Drift review before re-route — DRIFT flags require manual review, not automatic re-routing.

CLAIMS / COUNTEREXAMPLES / CONTRADICTIONS

[GENERATED] Chain 1 n8n workflows continue running on their own schedules when MCP is CONNECTION_ISSUE — the MCP connection is a Claude-to-n8n channel, not the workflow execution engine. Scheduled triggers are unaffected.
[GENERATED] Direct webhook curl bypasses n8n MCP entirely and remains operational regardless of MCP status, provided the n8n Cloud instance itself is healthy.
[GENERATED] Supabase MCP operates independently of n8n MCP — Supabase audit queries remain available while n8n is unavailable.
! CONTRADICTION: smoke_test.py uses [FAIL] Chain 1 webhook reachable as a hard-stop signal, but in degraded mode this FAIL is expected and non-blocking for Chain 2. The smoke test protocol in this skill annotates this explicitly, but the smoke_test.py script itself does not distinguish between CONNECTION_ISSUE (n8n MCP down) and n8n Cloud itself being down. A test FAIL on this line requires manual confirmation of n8n Cloud health before proceeding.

GAPS / BLINDSPOTS / UNKNOWN-UNKNOWNS

references/n8n-config.md contains the load-bearing CHAIN1_WEBHOOK_URL and auth headers. If that reference file is absent, the direct webhook curl in §Golden Path: Chain 1 cannot be completed without locating those values in the n8n Cloud UI manually. Consider embedding the webhook URL pattern (without the token) in this SKILL.md directly.
Root cause of n8n MCP CONNECTION_ISSUE not captured here. May be token expiry, MCP server config change, or n8n Cloud outage. Check https://status.n8n.io and re-authenticate in Claude.ai Settings → Integrations as first recovery step.
No automated alert when n8n MCP reconnects. The degraded mode annotation in this skill must be manually updated (compatibility block + description + DEGRADED MODE table) when MCP is restored. No hook or self-detection exists.
Smoke test [PASS] Chain 1 webhook reachable tests the webhook URL, not the n8n MCP connection. If n8n Cloud itself goes down (not just MCP), this test FAIL is no longer "expected and non-blocking" — it indicates Chain 1 cannot run at all, even via manual trigger. Manual confirmation of n8n Cloud health status required when this check fails.

running-drive-automation

Files5

SKILL.mdplugins/loxihart-ai/skills/running-drive-automation/SKILL.md