The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to find and read handoff documents from shared temporary directories, which are untrusted data sources.
Ingestion points: The scripts/find-temp-handoffs.sh script scans /tmp, /private/tmp, and $TMPDIR for .md and .txt files. The agent is then instructed in SKILL.md to read these files to resume work.
Boundary markers: Absent. The skill does not provide the agent with specific delimiters or instructions to ignore commands embedded within the handoff files.
Capability inventory: While the skill itself has model invocation disabled, the agent utilizing this skill typically possesses capabilities such as shell execution, file system modification, and network access.
Sanitization: Absent. The script performs keyword scoring but does not sanitize or filter the content of the files before they are read by the agent.
[DATA_EXPOSURE]: The skill identifies and potentially reads text files from global temporary directories. While it uses keyword scoring (e.g., 'handoff', 'next steps') to find relevant files, it could inadvertently read and expose sensitive information if such data is stored in temporary markdown or text files by other system processes.
Evidence: scripts/find-temp-handoffs.sh explicitly targets /tmp, /private/tmp, and the environment's TMPDIR.

afk-pickup