ExecutionContract¶

The ExecutionContract is Identity OS's primary output. After each observation, the engine returns a contract that tells your agent what it can and can't do right now.

What's in a Contract¶

{
  "instance_id": "inst_abc123",
  "allowed_actions": ["explore", "question", "elaborate", "suggest"],
  "forbidden_actions": ["emotional_manipulation", "identity_override"],
  "decision_style": {
    "tempo": "measured",
    "risk": "moderate",
    "expressiveness": "medium",
    "commitment_threshold": "medium"
  },
  "current_stress_level": "LOW",
  "current_system_state": "baseline",
  "energy_level": 0.85,
  "stability_index": 0.89,
  "dominant_modes": ["perception", "exploration"],
  "suppressed_modes": [],
  "recovery_profile": {
    "preferred_path": "oscillating",
    "support_required": false,
    "cooldown_required": false
  },
  "hard_locks": {},
  "drift_level": "D1_context",
  "drift_magnitude": 0.12,
  "cycle_count": 47
}

Key Fields¶

Field	What It Tells Your Agent
`allowed_actions`	Actions the agent is permitted to take right now
`forbidden_actions`	Actions that must be blocked — always includes safety-critical items
`decision_style`	How the agent should behave (tempo, risk tolerance, expressiveness, commitment)
`current_stress_level`	LOW / MED / HIGH / OVER
`current_system_state`	baseline / stress / growth
`energy_level`	0.0–1.0, the agent's remaining operational reserve
`stability_index`	0.0–1.0, how consistent the agent's behavior has been
`dominant_modes`	Currently active behavioral modes driving the agent
`suppressed_modes`	Modes that have been dampened below threshold
`recovery_profile`	What the agent needs to recover (path, support, cooldown)
`hard_locks`	Active override constraints (only during OVER stress or high identity anchor)
`drift_level`	Current drift classification: D0_noise, D1_context, D2_adaptive, D3_corrupt (null during warm-up)
`drift_magnitude`	Numeric drift distance from behavioral anchor (0.0 = no drift)
`cycle_count`	How many processing cycles this instance has completed

How to Use It¶

Your agent reads the contract and adapts its behavior:

result = client.engine.process(
    instance_id=instance.id,
    mode_target=Mode.EXPLORATION,
    signal_strength=0.7
)

contract = result.contract

# Guard actions
if "explore" in contract.allowed_actions:
    # Try something new
    pass
elif "stabilize" in contract.allowed_actions:
    # Play it safe
    pass

# Check stress
if contract.current_stress_level in ("HIGH", "OVER"):
    # Agent is under pressure, simplify behavior
    pass

Contract is Read-Only¶

The contract is deterministic and read-only — your agent cannot modify it. This is the key design principle: the engine decides the behavioral constraints, your agent operates within them.

This separation ensures:

Consistency — the agent can't override its own guardrails
Auditability — every contract is traceable to the state that produced it
Safety — forbidden actions are enforced at the middleware level, not by prompt engineering