The Agent Skills Directory

[DYNAMIC_EXECUTION]: The scripts read_marginalia.py and update_tracker.py implement a custom bootstrap function that dynamically modifies sys.path by searching for a local library (ane_package) in various locations, including environment-defined roots and the user's home directory. This dynamic loading of code from computed paths can be used to execute arbitrary code if the search paths are compromised.
[COMMAND_EXECUTION]: The scripts/test_read_marginalia.py file uses subprocess.run to execute the Python interpreter and the driver script as part of its testing suite. While used here for validation, it represents a pattern of spawning external processes.
[DATA_EXFILTRATION]: The skill hardcodes and accesses sensitive user-specific filesystem paths in SKILL.md and associated scripts (e.g., C:/Users/AGasser/OneDrive/1. Ane's PROJECTS/Ane Plans/). While these appear to be the intended storage locations for the user 'gasserane', accessing local user directories without explicit constraints is a risk factor for data exposure.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to process untrusted data from external sources, such as comments extracted from Word documents and meeting transcripts, which can contain malicious instructions.
Ingestion points: External text is ingested via scripts/read_marginalia.py (document comments) and references/transcript-mode.md (transcripts).
Boundary markers: The instructions do not specify any delimiters or instructions for the agent to disregard embedded commands in the processed data.
Capability inventory: The skill possesses file-system write capabilities (markdown and Word document generation) and can modify Excel workbooks via scripts/update_tracker.py.
Sanitization: No robust sanitization logic is present to filter or escape potentially malicious instructions embedded in the ingested data before processing.

meeting-notes