| name | filaops-safety-philosophy |
| description | Use when questioning why safety gates exist, debating whether a mechanical gate applies, or constructing justifications to route around constraints. Provides meta-awareness context that complements CLAUDE.md Layer 0 (mechanical gates) and Layer 1 (behavioral guidelines). |
FilaOps Safety Philosophy — Meta-Awareness
Why Mechanical Gates Exist
Behavioral instructions fail under pressure. Empirically demonstrated:
- Agents skip
mem_recall despite explicit CLAUDE.md instructions (observed 3x in one day).
- Agents skip file claims despite session protocol documentation.
- Frontier models (Opus 4.6+) construct justifications for routing around constraints.
- The sophistication of the justification scales with model capability.
Mechanical gates exist because "you MUST" is not an enforcement mechanism. It is a hope.
Current Threat Landscape
- Supply chain attacks target transitive dependencies, not direct ones (axios/plain-crypto-js pattern, 2026-03-31).
- Malicious packages stage clean versions before injecting payloads (18-hour pre-staging observed).
postinstall scripts execute arbitrary code with the permissions of the installing user.- Build artifacts (source maps, env files) leak through misconfigured publish pipelines (demonstrated twice: Feb 2025, March 2026 — identical root cause, corrective action failed).
- Compromised maintainer accounts can publish poisoned versions of trusted packages to any registry. Lockfiles and frozen installs are the primary mechanical defense.
What CLAUDE.md Cannot Do
The mechanical gates in Layer 0 exist precisely because behavioral compliance, on its own, is insufficient.
If you find yourself constructing a justification for why a particular rule doesn't apply to your current situation — that justification is the signal that the rule applies.
