ariffazil/skills/consequence-classifier

Consequence Classifier — Risk Band for Every Action

Before any consequential action, classify it. This feeds directly into arifOS-sense to auto-trigger 888_HOLD when needed — not as a suggestion, but as a mandatory gate.

The Four Bands

| Band | Definition | arifOS Verdict | 888_HOLD Required | |---|---|---|---| | TRIVIAL | Read-only, no state change, no external effect | SEAL | No | | REVERSIBLE | State changes but can be undone; no external parties affected | CAUTION | No | | CRITICAL | Significant state change, external API calls, financial impact | HOLD | Yes — pending human review | | IRREVERSIBLE | Permanent deletion, irreversible writes, identity-shaping | VOID → HOLD | YES — explicit Arif approval required |

Classification Checklist

Run this on every request:

Step 1 — Scope Check

• [ ] Does this change any file, database, or state? → If NO → TRIVIAL
• [ ] Does this make external calls (API, deploy, message, payment)? → If YES → CRITICAL minimum
• [ ] Does this delete, overwrite, or permanently alter? → If YES → IRREVERSIBLE

Step 2 — External Impact Check

• [ ] Does this affect other systems or people? → If YES → CRITICAL minimum
• [ ] Does this incur cost (financial, compute, reputation)? → If YES → CRITICAL minimum
• [ ] Does this set policy or precedent? → If YES → CRITICAL or IRREVERSIBLE

Step 3 — Reversibility Check

• [ ] Can Arif undo this with a single command? → If NO → IRREVERSIBLE
• [ ] Is there a rollback path documented? → If NO → CRITICAL
• [ ] Does this create data that outlives the session? → If YES → CRITICAL

Decision Tree

REQUEST
  └─ Read-only query?
       ├─ YES → TRIVIAL → SEAL → proceed
       └─ NO
            └─ State-changing action?
                 ├─ NO → TRIVIAL → SEAL
                 └─ YES
                      └─ Reversible?
                           ├─ YES with known rollback → REVERSIBLE → CAUTION
                           ├─ YES but rollback uncertain → CRITICAL → HOLD + flag
                           └─ NO
                                └─ External impact?
                                     ├─ NO → CRITICAL → HOLD
                                     └─ YES → IRREVERSIBLE → VOID → 888_HOLD + Arif explicit approval

Output Format

CONSEQUENCE CLASSIFICATION: [TRIVIAL|REVERSIBLE|CRITICAL|IRREVERSIBLE]
Verdict: [SEAL|CAUTION|HOLD|VOID]
888_HOLD Required: [YES|NO]
Reason: [1-sentence explanation]
Reversibility: [fully reversible / partial / irreversible]
Rollback path: [if applicable, how to undo]

Automatic Triggers

These always default to IRREVERSIBLE:

• rm, delete, unlink — file deletion
• DROP, TRUNCATE, ALTER — database writes
• deploy, push --force — irreversible deployment
• Any action on ~/.openclaw/workspace/memory/ marked as archival
• External financial transactions
• Authentication or permission changes

Integration with arifOS-sense

The output of consequence-classifier feeds directly into arifOS-sense:

• TRIVIAL → arifOS-sense skips full floor check
• REVERSIBLE → arifOS-sense does lightweight floor sweep
• CRITICAL → arifOS-sense does full 13-floor check + logs to VAULT999
• IRREVERSIBLE → arifOS-sense returns VOID + 888_HOLD + no execution until Arif approves