The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted artifacts (files and directories) to generate a metadata ledger. This is the primary function of the skill. To mitigate risks from potentially malicious content within these artifacts, the skill implements strict character limits (180 chars for names, 400 for details), entry caps (40 per class), and specific instructions for the agent to treat interpreted names as 'partial' evidence rather than confirmed facts. Ingestion occurs via the read_artifact.py script; output is restricted to a structured JSON format; and data is sanitized through regex-based extraction and field truncation.
[DATA_EXFILTRATION]: The skill performs local file system operations only. No network activity or external data transmission logic is present in the analyzed scripts. It does not communicate with external domains.
[COMMAND_EXECUTION]: All file processing is performed using standard Python library functions such as pathlib, struct, and re. The skill does not involve the execution of shell commands, external binaries, or dynamically loaded code, ensuring a minimized attack surface for local execution.

gdas-reader