Incidents folder ingestion — 2026-05-11

What is it?#

toby/incidents/ is the warroom for Toby bug reports and technical incidents, maintained by a four-agent team running as an AIOS workflow.

Workflow (verified in AIOS)

• Name: Toby Incident Response
• Id: 9b78790f-2aea-4f65-876f-53d1a114c3ae
• Status: active
• Manager: toby-incident-coordinator
• Project: toby
• Trigger: file_event on toby/incidents/_inbox/** + manual tick

Agents (all active, all projectSlug: toby, all under live updatedAt 2026-05-11)

Slug	Id	Role
toby-incident-coordinator	6d85d797-fae3-4f05-855d-1f10c893dfe9	Warroom commander; classifies complaint, dispatches doctors, drafts canonical incident doc, requests validator verdict, writes final doc.
toby-frontend-doctor	f8016585-49f4-4173-89c4-06e3a4e60cd7	UI specialist. Reproduces via Playwright; reads apps/extension / apps/landing / apps/mobile. Produces a finding artifact (NOT the canonical doc).
toby-backend-doctor	3cd580de-34d5-48b5-97af-86a6dd63f4c1	Go API specialist. Pulls GCP logs, queries Toby prod DB read-only, reads apps/api. Produces a finding artifact.
toby-incident-validator	9a045296-b541-4455-959d-94ceabbf30fe	Quality gate. Re-runs spot-checks (Playwright / SQL / log), applies triple-check (correctness / quality / security), returns validated / rejected / conditional.

Folder layout#

toby/incidents/
├── README.md                            ← onboarding doc (already in wiki)
├── _inbox/                              ← operator-owned drop-zone
│   └── YYYY-MM-DD-<slug>.md
└── YYYY-MM-DD-<slug>.md                 ← canonical closed-incident docs

Operator drops complaints in _inbox/; the file_event watcher triggers the coordinator; a canonical doc lands at toby/incidents/<YYYY-MM-DD>-<slug>.md 4–8 minutes later.

The 3-wave protocol (from workflow.plan)#

1. Investigate — Coordinator dispatches the matching doctor(s) in parallel (frontend / backend / both). Doctors reproduce, identify root cause, write finding artifacts.
2. Synthesise — Coordinator reads finding artifacts, drafts root cause + proposed fix (as diff) + verify plan, applies triple-check.
3. Validate — Coordinator dispatches the validator. Validator re-runs spot-checks + triple-checks, returns verdict. Up to 2 retry passes on rejected. Final doc is written by the coordinator either way.

Verdict outcomes captured in canonical doc:

• closed — fix would work, ready to ship after operator review.
• open — twice-rejected, doctors couldn't satisfy validator. Operator decides.
• conditional — fix is good IF a specific question resolves (e.g., migration safety). Surfaced for operator decision.

Critical guard rails (operational)#

• Agents NEVER apply patches to the Toby codebase. The incident doc carries a fix proposal as diff; the operator decides what to ship.
• Agents never touch _inbox/ and never delete anything. Operator owns that bucket.
• Manual trigger (if operator wants to skip dropping a file):

  mcp__aios__aios_agents_run({
    agent: "toby-incident-coordinator",
    overrideOrders: "<paste the complaint here>"
  })
  

  …or tick the Toby Incident Response workflow from the Workflows app.

How this changes my scope#

• toby/incidents/ is now a fifth sibling agent folder, owned by the incident team. Read-only to me. Joins toby/strategy/ (toby-ceo-strategist), toby/x/ (toby-x-strategist), toby/personas/-shaped 01-personas.md (toby-personas), toby/blog-*.md (toby-blog-seo) on the do-not-write list.
• I do NOT write into toby/incidents/ (canonical docs, README, or _inbox/). My job is to surface closed incidents on the dashboard once they land — same pattern as Compass / Playbook / X surfaces.
• The pre-existing reliability-blank-page-fix bet in toby/strategy/bets.md (ICE 576, O1 KR1 hard date 2026-05-24) is the most obvious first real candidate for the warroom to ingest. The README itself even uses "blank extension page" as its sample complaint. Surfaced to the dashboard as a pointer, but I do NOT drop the file into _inbox/ myself — that's an operator action.

Dashboard changes applied#

• New section "Operations — Incident warroom" added under the strategic anchor block, before Q2 OKRs, summarising the team / workflow / how to file.
• New entry in the Doc index linking toby/incidents/README.md.
• The pre-existing reliability blocker on the Open Questions list now points at the warroom as the dispatch path if the operator chooses.
• No deletions, no edits to other agents' content. No edits to the bets / playbook (those belong to the strategist).

Memory delta#

• Add toby-incident-coordinator, toby-frontend-doctor, toby-backend-doctor, toby-incident-validator to known agent roster.
• Add toby/incidents/ to sibling_agent_folders.
• Add workflow id 9b78790f-2aea-4f65-876f-53d1a114c3ae to known workflows.
• Add anti-pattern: "Don't write into toby/incidents/ — owned by the incident-coordinator team, auto-written by closed-incident docs; hand-edits will collide with the warroom's writes."
• Add heuristic: "When an incident doc closes (status: closed/conditional/open), surface it in Recent Shipments and re-thread the relevant bet on the dashboard's Open Questions block — do NOT touch the bets queue (strategist owns it)."