remember
Pass
Audited by Gen Agent Trust Hub on Sep 18, 2026
Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The execution instructions in
SKILL.mddirect the agent to run a shell command where user-supplied content and types are interpolated directly into the command string.\n - Evidence:
uv run python scripts/core/store_learning.py --type <TYPE or WORKING_SOLUTION> --content \"<ARGS>\"inSKILL.md.\n - Risk: If metacharacters like semicolons or quotes are present in the
<ARGS>or<TYPE>placeholders without sanitization by the executing agent, it could lead to arbitrary command execution.\n- [INDIRECT_PROMPT_INJECTION]: The skill facilitates the ingestion of arbitrary data into a long-term memory system, presenting a risk of persistent indirect prompt injection.\n - Ingestion points: Input captured via the
/remembercommand.\n - Boundary markers: Absent. The content is passed directly to the storage script.\n
- Capability inventory: Shell execution and database interactions.\n
- Sanitization: Not described; untrusted input is stored for future retrieval, which may influence later agent decisions.
Audit Metadata