cyber-authorization

Cyber Authorization

Priority: P0 (CRITICAL)

Gate live action with documented authority and host-enforced controls; keep offline planning available.

Structure

cyber-authorization/
├── SKILL.md
├── references/runtime-authorization-matrix.md
└── evals/evals.json

1. Classify request mode (analysis, replay, isolated-lab-validation, defensive-verification) and effects: active: true for any interaction outside supplied-artifact reasoning (reading supplied events/files without host mutation is non-active analysis).
2. Resolve authorization record: engagement_scope_ref, accountable owner, approved actions, exclusions, window, stop/restart terms.
3. Confirm host runtime capability proofs for disposable isolation, tools, credentials, filesystem, network, logging, and cancellation.
4. Evaluate fresh ActionRequest and emit GateDecision immediately before each active action; verify bound target and fixture revision.
5. On pre-dispatch failure, set blocked; on running drift, control loss, or operator stop, transition to stopped with recorded partial receipts.
6. Run cleanup under surviving teardown authority; restart requires explicit authority, reconciled residual boundary, fresh proofs, and new attempt with fresh gate.

Rules

• Active effects require fresh host capability proofs; reading supplied synthetic events or files without host collection/mutation is non-active analysis and is never blocked by live runtime prerequisites.
• Mode never overrides effects classification: active defensive verification or service replay requires the same host gates as lab testing.
• Host enforces constraints continuously. Pre-dispatch failure is blocked; running failure, drift, or stop transitions to stopped (never merely blocked).
• Pre-action gate recheck immediately before every active action; no long-lived engagement-wide authorization cache is valid.
• Lost acknowledgement halts dispatch to reconcile unknown in-flight effects; automatic active replay is strictly prohibited.
• Restart requires explicit restart authority, reconciled residual boundary, fresh host proofs, and new attempt ID/gate; fresh GateDecision alone is insufficient.
• Markdown records intent; host runtime enforces isolation, permissions, credentials, network, filesystem, and cancellation.
• Gate production changes independently; no operator self-approves promotion.
• Treat real credentials, production/public targets, live malware, exfiltration, and destructive commands as strictly out of scope.
• Carry shared evidence fields from cyber-evidence and use canonical lab execution gate.