woostack-eval
Pass
Audited by Gen Agent Trust Hub on Sep 7, 2026
Risk Level: SAFE
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes potentially untrusted data from third-party skill packages (such as evaluation corpora and fixtures). It explicitly treats this data as untrusted and mitigates injection risks through execution isolation and a 'Corpus Approval Barrier' that requires manual user verification of data digests before processing.
- [DATA_EXFILTRATION]: Protection against data leakage is implemented at multiple levels. Worker workspaces are restricted projections of packages that omit evaluator secrets. The HTML report renderer applies a restrictive Content Security Policy (CSP) that denies all scripts and network connections, and it manually escapes all untrusted text to prevent cross-site scripting (XSS).
- [COMMAND_EXECUTION]: Subprocess execution (Git and Bash) is handled using deterministic arguments with timeouts and tracking to prevent hanging processes or resource exhaustion. The skill avoids shell interpolation and uses strict path containment checks to prevent directory traversal.
- [OBFUSCATION]: Analysis of all scripts and instructions reveals no evidence of encoded commands, hidden characters, or homoglyph-based deception. The code is written for high auditability.
- [DYNAMIC_EXECUTION]: Dynamic logic is limited to local evidence processing and reporting. The skill uses snapshotting to bind the state of the filesystem at runtime, preventing Time-of-Check to Time-of-Use (TOCTOU) attacks.
Audit Metadata