interlock-ops
Operational rules for Interlock enforcement and receipts
When & Why to Use This Skill
The Interlock Operations skill provides a rigorous framework for managing AI enforcement rules and operational integrity. It ensures system accountability by mandating cryptographically signed receipts for every action—such as refusals or service degradations—while strictly protecting internal heuristics to prevent adversarial gaming. This skill is essential for maintaining high-standard compliance, forensic logging, and fail-safe operational protocols in AI deployments.
Use Cases
- Scenario 1: Automated Audit Trails - Generating structured, signed receipts for every AI decision to maintain a forensic log for compliance and incident investigation.
- Scenario 2: Fail-Safe Enforcement - Implementing 'Refuse-by-Default' logic to ensure that any system failure, such as a validator timeout or schema mismatch, results in a safe state rather than a silent pass.
- Scenario 3: Security & Anti-Gaming - Protecting internal enforcement thresholds and scoring algorithms from public exposure to prevent users from bypassing safety guardrails.
- Scenario 4: Schema Integrity Management - Using automated scripts to audit receipt formats and run negative tests, ensuring that operational logging remains consistent and tamper-proof.
| name | interlock-ops |
|---|---|
| description | Operational rules for Interlock enforcement and receipts |
| user-invocable | false |
Interlock Operations Skill
This skill governs operational behavior for Interlock enforcement.
Core Rules
1. Never Publish Enforcement Heuristics
Enforcement logic is internal. Never commit or document:
- Threshold values for enforcement decisions
- Specific trigger conditions
- Bypass mechanisms or exceptions
- Internal scoring algorithms
Reason: Publishing enforcement heuristics enables gaming.
2. Every Enforcement Action Must Emit a Receipt
No silent enforcement. Every action must produce:
{
"action": "refusal|degradation|allowance",
"timestamp": "2026-01-10T12:00:00Z",
"reason": "quality_floor_breach",
"incident_id": "int-123456",
"confidence": 0.45,
"threshold": 0.80
}
3. Receipt Format Changes Require Validation
Before changing receipt schema:
- Update the schema definition
- Create positive test fixture
- Create negative test fixture
- Run
./scripts/claude/receipt_audit.sh - Verify both positive and negative tests pass
4. Any Failure = Degrade/Refuse
Never allow silent passes:
| Scenario | Action |
|---|---|
| Validator unavailable | REFUSE |
| Schema mismatch | REFUSE |
| Confidence unknown | REFUSE |
| Timeout | DEGRADE |
| Partial data | DEGRADE |
Receipt Lifecycle
1. EMIT → Receipt generated at decision point
2. SIGN → Cryptographic signature applied
3. STORE → Written to forensic log
4. VERIFY → Schema validation on read
5. AUDIT → Periodic integrity check
Validation Commands
# Full receipt audit with negative tests
./scripts/claude/receipt_audit.sh
# Public safety check (no secrets in receipts)
./scripts/claude/public_safety_check.sh
Schema Files
| File | Purpose |
|---|---|
schemas/interlock_event.schema.json |
Event schema definition |
schemas/golden_fixture.jsonl |
Known-good test fixtures |
receipts/examples/operatorpack_example_pass.json |
Valid receipt |
receipts/examples/operatorpack_example_fail.json |
Invalid receipt (for negative test) |
Prohibited Actions
- Publishing enforcement thresholds
- Silent passes on validation failures
- Modifying receipts after signing
- Skipping negative tests
- Committing real receipts to public repo