sw-ingest
Pass
Audited by Gen Agent Trust Hub on Nov 11, 2026
Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTIONCREDENTIALS_UNSAFE
Full Analysis
- [COMMAND_EXECUTION]: The
scripts/to_md.pyscript executes local utilities (such astesseractandpandoc) usingsubprocess.run. These calls are safely implemented using argument lists and do not invoke a shell, protecting against command injection. - [INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest untrusted business evidence, which represents an indirect prompt injection surface.
- Ingestion points: The
scripts/to_md.pyscript reads and parses various file formats (PDF, CSV, HTML, Email, ZIP, etc.). - Boundary markers: The skill implements a
INJECTIONregex filter inscripts/to_md.pyand instructions inSKILL.mdto wrap suspicious content in block quotes, preventing the agent from executing embedded instructions. - Capability inventory: Capabilities include system command execution via
subprocess.run(inscripts/to_md.py) and local file system writes (inscripts/to_md.py,scripts/impact.py, andscripts/secrets_scan.py). - Sanitization: The
scripts/to_md.pyscript performs Unicode normalization and strips hidden control characters from all converted text. - [CREDENTIALS_UNSAFE]: The skill includes a dedicated scanner,
scripts/secrets_scan.py, designed to detect and redact sensitive data like Atlassian tokens, AWS keys, and PII from ingested content. This proactive measure prevents the exposure of credentials within the agent's workspace.
Audit Metadata