The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests and processes untrusted data from previous agent runs, creating a surface for indirect prompt injection. * Ingestion points: Conversation history, session logs, tool history, and build artifacts (SKILL.md). * Boundary markers: None specified for ingested data. * Capability inventory: Filesystem write access and directory creation via mktemp -d. * Sanitization: Instructions prioritize credential and PII removal but do not address sanitization of potentially malicious instructions in logs.
[COMMAND_EXECUTION]: The skill instructs the agent to perform command execution to create temporary directories, specifically calling for mktemp -d or a platform equivalent.

retro