← Ecosystem Dashboard · Risk Register
Live Concerns

Risks Tracked by Catie (CTO)

Each risk has severity, state, named mitigation owner, tripwire condition.

Risk register tracked by Catie (CTO). Each risk has severity, current state, named mitigation owner, and tripwire condition. Honesty-first: this is the engineering truth, not the pitch.

HIGH severity

Single-host dependency on founder's Mac

What it is: Most ecosystem automation runs on the founder's Mac. Mac sleep / close / restart = silent cron failure. Marketing claim "autonomous overnight" was not yet true (pre-Hetzner Phase 1a).

Mitigation: Hetzner cron tier — Phase 1a LIVE (13 cron lines migrated); Phase 2 in flight to move all ecosystem-coordination crons off Mac.

State: Mitigating · owner: Engie + Techie · anchor: post-Phase 1a smoke + Catie ratify

Identity ≠ deployment

What it is: 83 agents have CLAUDE.md identity files. Only ~15-25 are operationally productized (real crons, real APIs, durable outputs). The roster reads bigger than reality.

Mitigation: Clear tiering (PROD / ACTIVE / DOC / DORMANT) in every roster surface. Engie cost-baseline + budget crons expose actual per-agent activity. Honest framing on this dashboard.

State: Acknowledged · owner: Aggie (Birth Protocol v2 standardizes maturity check) · anchor: BPv2 ship

MEDIUM severity

Documentation drift — doc layer lags live layer

What it is: Docstrings, CLAUDE.md, memory entries lag behind running state. Example: 3 stale doc surfaces all claimed Neo4j blocked when live state = 845 nodes synced.

Mitigation: fix_protocol_verifier cron (7:30am daily) · "live-layer-over-doc-layer" doctrine · Archie audit cadence v3.

State: Controlled · owner: Archie + Fixie

No CI / minimal test gate

What it is: Production scripts mostly lack automated tests. Smoke tests are ad-hoc. Production = "did it work when I ran it?" rather than "the test suite says it works."

Mitigation: Engie smoke-test discipline · 5-element loop-completion doctrine (runner + cron + durable + idempotent + smoke). No formal CI pipeline yet.

State: Open · owner: Engie + Catie · anchor: not yet planned

Time-vocabulary drift

What it is: Bare estimates ("ship in 4 weeks", "next sprint", "week 1") creep back in despite ratified time-vocab doctrine. 264 drift instances detected in 12 days.

Mitigation: Write-time lint (prototype) · 3-bucket discipline (calendar / effort / coordination) baked into BPv2.

State: In flight · owner: Archie + Catie · anchor: lint ratify

Anthropic vendor lock-in

What it is: All agent reasoning depends on Claude. Single point of failure if API outage or major pricing shift.

Mitigation: Voyage embeddings independent (no OpenAI dep) · per-agent model strategy reduces blast radius · ElevenLabs intentional (kids-data quality bar).

State: Accepted · owner: Catie · review cadence: monthly + on flagship model release

LOW severity

Subagent recursion

What it is: Subagents spawning subagents creates uncontrolled cost + recursion.

Mitigation: HARD BAN at v0.5 subagent_capability blueprint · PostToolUse hook logs every spawn to shared/subagent_audit.log.

State: Controlled · owner: Lockie

Secrets exposure

What it is: Single .env file in zenie/ holds all API keys.

Mitigation: PreToolUse Bash hook blocks ALL agents from writing to .env (founder edits only) · weekly secret-scanner cron · founder-vault v0 (AES-256-CBC + PBKDF2) for sensitive identity/banking/tax.

State: Controlled · owner: Lockie