Dashboardinstitutional-transition-labPerform read-only adversarial audits wi…

Perform read-only adversarial audits with deterministic evidence

Category: preference
Confidence: 0.80
ID: 01a086ec-dc5e-75b1-a691-0071c8a2ef6f
Project ID: 1fa3712f-997f-4884-b7ec-8abb79f9e342
Cross-project: No
Recalled in other projects: 1
Source session: 0yGAnpNlrOvBQqxq2
Created: 2026-09-09 02:51:59
Updated: 2026-09-09 16:07:45

Cross-Project Recalls

ProjectHitsLast recalled
opencode-lore 3 2h ago

Content

When auditing security- or governance-critical code, keep the worktree unchanged and remain within the explicitly authorized files, symbols, tests, data, and concerns. Inspect the current workspace and use only narrow deterministic read-only tests or synthetic probes. Verify actual fail-closed behavior, malformed-input handling, provenance integrity, path and ZIP safety, races, cleanup, resource bounds, output consistency, and whether tests exercise each guard. Report substantive findings first in severity order with exact current file:line evidence and PASS, CONCERN, or MUST-FIX labels; include commands and results, hashes or worktree-integrity confirmation, residual risks, and a deterministic fail-first regression for every defect. If blocked, name the exact tool, path, and error. Never return empty output; honor required MERGE or DO-NOT-MERGE endings and avoid overlapping active reviews.

Move to: