Second Brain

Promote SKILL-FIRST PRINCIPLE from advisory to BLOCKING (PreToolUse exit-2) via three-hook architecture for canonical skills enterprise/vault/sequential-thinking. Test suite 10/10 green. Bypass via SK

1 min read

Promote SKILL-FIRST PRINCIPLE from advisory to BLOCKING (PreToolUse exit-2) via three-hook architecture for canonical skills enterprise/vault/sequential-thinking. Test suite 10/10 green. Bypass via SK

Decision

Promote SKILL-FIRST PRINCIPLE from advisory to BLOCKING (PreToolUse exit-2) via three-hook architecture for canonical skills enterprise/vault/sequential-thinking. Test suite 10/10 green. Bypass via SKILL_FIRST_BYPASS=1 (logged). Audit eradicated 905MB dead weight + restored vault meta/log + added feedback_vault_proactive_workflows.md mandating 5 Vault touchpoints.

Rationale

First-principles decomposition (this session): advisory enforcement loses to concrete tool calls because both compete for the same model attention budget. Inserting a hard gate between suggestion and action — demanding skill invocation as a prerequisite — is the only structural fix. setup-curator-guard.py proved this pattern works for Write/Edit. Extending to Bash|Grep|Glob|Read with AND-of-three guard (suggested + not-invoked + domain-keyword-match) keeps false-positive rate low. Premortem flagged four latent risks (static keyword drift, marker leak, false-positive on incidental keywords, autoloader silent-disable) — three mitigated in-design, one (autoloader-change detection) escalated to NOVA via vault handoff for Hermes Consul cron coverage. Bias detection flagged anchoring on “10/10 tests” — countered by recognising tests cover the mechanism, not real-world keyword coverage; recommend 30-day bypass-log review.

Alternatives Rejected

Outcome

Pending

Graph View

Backlinks