ac-init
Pass
Audited by Gen Agent Trust Hub on Sep 15, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill ingests content from external codebase files to determine project structure, metadata, and scripts, exposing a surface for indirect prompt injection.
- Ingestion points: Repository configuration and source files (e.g.,
package.json,pyproject.toml) parsed during Step 1 and Step 3 inSKILL.md. - Boundary markers: Absent. No safety instructions or delimiters are used to ensure that external content is treated purely as data rather than instructions.
- Capability inventory: File-system modification via the Write tool to create or update root-level and module-level
CLAUDE.mdfiles as specified inSKILL.md. - Sanitization: Absent. The skill dynamically interpolates extracted values directly into markdown templates without filtering or validation.
Audit Metadata