recall
Pass
Audited by Gen Agent Trust Hub on Nov 8, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes external memory files, creating a surface for potential instruction injection from stored data.
-
- Ingestion points:
*.mdfiles in the<repo-declared-memory-store>specified in SKILL.md.
- Ingestion points:
-
- Boundary markers: The instructions for the subagent do not define delimiters to separate file content from the subagent's task.
-
- Capability inventory: The subagent is restricted to read-only file access and an explicit model tier as described in SKILL.md.
-
- Sanitization: Memory content is analyzed verbatim without sanitization.
- [COMMAND_EXECUTION]: The skill provides a shell snippet for environment variable resolution and path normalization.
- Evidence:
SLUG=$(printf '%s' "$PWD" | sed 's#[/_.]#-#g')in SKILL.md. This is a benign path transformation operation.
Audit Metadata