Skip to content

Agent Governance

Safety rails, monitoring, and operational rules for autonomous agents. Eleven governance rules, session management, and audit trails.

Page What You'll Learn
System Registry Configuration registry, environment variables, service inventory, canonical facts
Email Operations Dual-account management, send checklist, response standards, HTML templates
Cron Scheduling Job management, scheduling patterns, chain jobs, delivery channels
System Monitoring Health checks, drift detection, alerting, self-healing crons, audits

Eleven Governance Rules

  1. Constitution first — load CONSTITUTION.md before any action
  2. Task history — check TASK_HISTORY.md before every action
  3. System registry — verify SYSTEM_REGISTRY.md before creating anything
  4. Profile constraint — single profile, no exceptions
  5. Done-only reporting — never report scheduled/queued/planned work
  6. Execution discipline — action first, reporting second
  7. Session governance — session DB, token optimization, compaction
  8. Content ontology — pre-execution gate for all public content. Six dimensions: Public vs Internal, Human vs AI, Audience, Naming, Help-First, Freshness. Load before every external action.
  9. GitHub repo synchronization — Every new tool, application, architecture change, memory entry, skill, cron job, or system modification must be pushed to the CorpusIQ/corpusiq-docs repository immediately after the change. No change goes undocumented. The repo is the single source of truth for system state.
  10. Platform connection failure — When a platform connection fails (auth error, ban, shadowban, rate limit, token missing): STOP. Do not retry. Do not loop. Log the failure and move on. Crashing by retrying nonstop is worse than being silent on a dead platform. Only retry on explicit user instruction.
  11. Email sender identity — From name is permanently locked. One account, one name. Never change the sender name on outbound emails. Changing names from the same account triggers Gmail impersonation detection and account flags.

Monitoring Stack

Layer 1: Process — agent alive, CPU/memory bounds, no zombies
Layer 2: Operations — crons firing, emails flowing, APIs valid
Layer 3: Quality — outputs meeting thresholds, correct channels

← Infrastructure | Content Ops → ↑ Home