메뉴
Zouroboros self-healing watchdog that monitors scheduled agent models and automatically switches failing agents to backup models, restoring originals when they recover.
Codex 또는 Claude로 설치 이 Prompt를 복사해 Codex, Claude 또는 다른 어시스턴트에 붙여 넣으면 Skill 페이지를 검토하고 설치를 진행할 수 있습니다.
SOC 직업 분류 기준
Local-only multi-agent orchestrator with token optimization, 6-signal composite routing, and persistent memory. Routes all tasks through local executor bridges (Claude Code, Hermes, Gemini, Codex) with DAG dependencies, auto-episode creation, and cognitive profiles.
Zouroboros self-healing watchdog that monitors scheduled agent models and automatically switches failing agents to backup models, restoring originals when they recover.
Hybrid SQLite + Vector persona memory system for Zo Computer. Episodic memory with temporal queries, graph-boosted search, BFS path finding, knowledge gap analysis, auto-capture pipeline, and per-workload model routing. Default generation workloads use OpenAI `gpt-4o-mini`; embeddings use OpenAI `text-embedding-3-small`.
Security governance gate for AI agent skills, backed by NVIDIA SkillSpector. Statically scans a skill (local path, directory, zip, or Git URL) for 65 vulnerability patterns across 16 categories (prompt injection, data exfiltration, privilege escalation, supply chain, excessive agency, tool/MCP misuse, dangerous code via AST, YARA signatures) and returns a 0-100 risk score with a SAFE / REVIEW / DO_NOT_INSTALL recommendation. Use BEFORE adopting any third-party skill, and for the periodic baseline scan of the Skills/ directory.
Detector for model classifier SOFT-blocks — refusals, filtered-to-empty, truncated-with-disclaimer, content-policy flags — that masquerade as successful empty output to existing infrastructure (agent-model-healer only catches HARD failures: HTTP 4xx/5xx, timeouts, genuine empties). Classifies an output as soft_block / genuine_empty / ok using a per-provider signal catalog, then routes a per-domain fallback (security → broader-tolerance model, bio/chem → human review, distillation → refuse-by-design, never routed around). Every block + fallback is appended to an audit ledger. A quarterly system-card-diff agent flags expanded classifier scope so the catalog stays current.
Rendered-output verification station for the swarm post-flight eval. After a maker subagent produces a visual deliverable (UI/route/site), the station captures a full-page screenshot via agent-browser, then an independent verifier model (not equal to author, per aiewf P0-2) reads the screenshot against seed acceptance criteria + project DESIGN.md tokens + prior screenshots. Match marks the task complete; mismatch emits a structured visual diff for the maker's next iteration. The verifier is the exit condition, not the maker. Recurring visual failure modes graduate to instincts.yaml via the extract-patterns gate (ZOU-452).
| name | agent-model-healer |
| description | Zouroboros self-healing watchdog that monitors scheduled agent models and automatically switches failing agents to backup models, restoring originals when they recover. |
| compatibility | Created for Zo Computer |
| metadata | {"author":"marlandoj.zo.computer","version":"2.0.0","category":"zouroboros-infrastructure"} |
Self-healing model fallback system for Zouroboros scheduled agents.
Zo scheduled agents are configured with a single model. When that model becomes unavailable (insufficient credits, rate limits, provider outage), the entire agent task fails silently with only an email notification. There is no platform-level retry or fallback.
A watchdog agent on zo:fast runs healer.ts auto every 30 minutes. The script is fully autonomous — all orchestration uses direct MCP API calls (zero model cost). The only AI cost is the probe prompts (~5 tokens × ~10 models = negligible).
/zo/asklist_agents (zero cost)edit_agent (zero cost)send_email_to_user only when actions taken.zouroboros/healer-state.json for audit# Full autonomous pipeline (what the agent runs)
bun Skills/agent-model-healer/scripts/healer.ts auto
# Check model health
bun Skills/agent-model-healer/scripts/healer.ts probe
# See current state
bun Skills/agent-model-healer/scripts/healer.ts status
# Legacy: probe + output instructions (manual use)
bun Skills/agent-model-healer/scripts/healer.ts run
The healer agent MUST run on a model that is NOT in any fallback chain it monitors.
If the healer shares a model with the agents it heals, a single provider outage (e.g., OpenRouter 402 insufficient credits) kills both the healer and its patients — a cascade where nothing can self-repair. The healer runs on zo:fast (Zo built-in), which is independent from all external providers it monitors.
This is enforced in healerConfig within assets/fallback-chain.json.
| Concern | v1 (Zo agent on Sonnet) | v2 (zo:fast + direct MCP) |
|---|---|---|
| Agent model cost | 48 Sonnet calls/day | 48 zo:fast calls/day (minimal) |
| Tool call cost | AI interprets JSON → calls tools | Direct HTTP to MCP endpoint (zero) |
| Watchmen safety | Shared provider risk | zo:fast is Zo-native, no external dep |
| Autonomy | Agent interprets instructions | Script executes deterministically |
Edit assets/fallback-chain.json to:
zo:fast or zo:smart to a fallback chain (healer runs on zo:fast)scripts/healer.ts — Main healer engine (auto, probe, diagnose, status, run)assets/fallback-chain.json — Fallback chain config and model labels/home/workspace/.zouroboros/healer-state.json — Runtime state (switches, probe results)/dev/shm/agent-model-healer.log — Operational logThe healer runs as a scheduled Zo agent (every 30 min) on zo:fast. The agent simply executes bun healer.ts auto — all logic is in the script, not the AI.