Skip to main content
GenesisAuditor (bot_id: GenesisAuditor) is the built-in agent for reviewing deployment structure. It produces evidence-backed reports and can propose fixes, which a person applies in Action Review.

Purpose

Review how the deployment’s reusable instructions and agent workflows fit together.

Evidence

Give operators reproducible coordinates and the model revision used for the review.

Safety

Fixes are only proposed; a person who may manage the target applies each one.

​
Availability

Genesis seeds GenesisAuditor as a visible, non-editable system agent with STATUS=ACTIVE. It appears in the customer-facing Agents list without a feature flag. Eve starts an audit when a user asks to review the organization’s health, skills, blueprints, agent activations, or duplicated instructions; the review runs in a hidden background thread and reports back when complete. GenesisAuditor is a built-in agent. Genesis sets its instructions, tools, status, and settings; you cannot change them. They are updated when Genesis is upgraded. You still control who can use GenesisAuditor and where it is deployed. Its private loadout, provenance checks, and tool-by-tool safety boundary are documented in the GenesisAuditor tool reference. Proposing and applying fixes is disabled by default and requires org_health.fix_mode to be active.

​
What It Audits

  • AgentSkills package inventory, including malformed packages that normal summary parsing might omit.
  • Blueprint manifests and phase requests.
  • Live agent activation sets, missing packages, blocked packages, and co-load sizes.
  • Duplicated text and other structural findings defined by the pinned organization model.
See the GenesisAuditor tool reference for the report shape and paging workflow.

​
Typical Workflow

  1. Ask Eve to audit the organization, optionally naming a focus such as a specific agent, skill, blueprint, or suspected duplication.
  2. Genesis starts the hidden GenesisAuditor run and returns the background thread identifiers.
  3. GenesisAuditor runs the complete audit once, then reads the source behind any finding it reports before presenting it.
  4. Use the report’s summary for complete counts and rankings, but present the organization’s scope by default. Discuss Genesis-shipped content or another scope only when the user asks; see the tool reference for scope details. Page only when the complete findings or co-load index is needed.
  5. For a dead skill pin, explicitly ask GenesisAuditor to remove it. For a skill an agent activates but its own access policy blocks, ask whether to admit the skill or remove the activation; GenesisAuditor must not choose between those resolutions. Each proposal is re-proved and waits in Action Review for a person who may manage the agent to apply it. Applying re-proves the target and refuses if the agent changed since the proposal. An applied fix can be reversed the same way: ask GenesisAuditor to reverse it, and apply the reversal in Action Review. A reversal is refused if anything it recorded about the agent (its pins or skill access policy) changed since the fix. See the tool reference for the blocked-skill resolution details.

​
Configuration

GenesisAuditor is seeded during startup and is active by default. Its read-side audit remains available without a fix setting. Fix proposals are disabled unless org_health.fix_mode is set to active; the default is silent, and unset, unreadable, or unrecognized values fail closed to silent. Audit responses honor the existing tools.output_governor.inline_budget_bytes setting when choosing page sizes. For function parameters, paging behavior, and exact response fields, see GenesisAuditor tools.