Trust Engine

Progressive capability unlocks based on demonstrated reliability. Agents earn trust through successful operations.

Trust Levels

LOW|Score: 0-25

New or recently failed skills. Restricted to read-only operations.

Capabilities: read, analyze

MEDIUM|Score: 26-50

Demonstrated basic reliability. Can modify non-critical files.

Capabilities: read, analyze, write (non-critical)

HIGH|Score: 51-75

Proven track record. Full write access, can execute most operations.

Capabilities: read, analyze, write, execute, deploy (staging)

AUTONOMOUS|Score: 76-100

Elite status. Can auto-approve certain gates and operate with minimal oversight.

Capabilities: all operations, auto-approve gates below threshold

How Trust Changes

Event	Score Change	Notes
Successful operation	+3	Cumulative, capped at 100
Successful complex operation	+5	Multi-step or high-risk operations
Failed operation	-5	Standard failures
Critical failure	-15	Data loss, security issue, etc.
Bounty resolved	+10	Issue fixed, trust partially restored
Gate bypassed (approved)	0	No change, expected behavior

Trust in Gates

Gates can be configured to auto-approve when a skill's trust score exceeds a threshold:

"gates": [{
  "id": "write_config",
  "resolution": {
    "mode": "block",
    "autoApproveAboveTrust": 75
  }
}]

In this example, the write_config gate will auto-approve for skills with trust score above 75 (HIGH or AUTONOMOUS level).

Configuration

Define trust configuration in your skill's config.json:

"trust_config": {
  "initial_level": "MEDIUM",
  "capabilities": {
    "LOW": ["read"],
    "MEDIUM": ["read", "analyze", "suggest"],
    "HIGH": ["read", "analyze", "suggest", "write"],
    "AUTONOMOUS": ["read", "analyze", "suggest", "write", "deploy"]
  },
  "score_adjustments": {
    "success": 3,
    "failure": -5,
    "critical_failure": -15
  }
}