ace-context-to-memory
Pass
Audited by Gen Agent Trust Hub on Mar 6, 2026
Risk Level: SAFEPROMPT_INJECTIONCOMMAND_EXECUTIONEXTERNAL_DOWNLOADSREMOTE_CODE_EXECUTION
Full Analysis
- [PROMPT_INJECTION]: The skill creates a surface for indirect prompt injection by ingesting external documents (PDF, PPTX, DOCX, etc.) and recommending the agent inject the resulting content directly into its responses. Ingestion points include files processed via scripts/convert_to_markdown.py and scripts/index_memory.py. There are no mandatory boundary markers or 'ignore previous instructions' filters specified for the retrieved content. The skill possesses capabilities to read/write files, execute local scripts, and access the OpenAI API. No sanitization is performed on the content of ingested documents.
- [COMMAND_EXECUTION]: Local script execution and system commands are used for internal pipeline management. scripts/index_memory.py uses subprocess.run to call other internal Python scripts. scripts/link_workspace_source.py executes the Windows mklink command to manage directory junctions.
- [EXTERNAL_DOWNLOADS]: The skill's documentation recommends installing several standard libraries from official registries, including markitdown, openai, faiss-cpu, numpy, openpyxl, and pypandoc.
- [REMOTE_CODE_EXECUTION]: The scripts/build.py script performs dynamic loading from a computed path by modifying sys.path to import and execute logic from a sibling directory (ace-shaping).
Audit Metadata