The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process other skill packages ('targets') to perform simplification. The instructions direct the agent to 'Read the live target' and 'Run existing checks and representative tasks' from these untrusted sources. This creates a surface where a malicious target skill could embed instructions or scripts intended to influence the agent's behavior or execute code during the simplification process.
Ingestion points: SKILL.md Procedure Step 1 (Discovery) requires reading the live target, linked files, and governing source.
Boundary markers: There are no explicit instructions to use delimiters or warnings to ignore embedded instructions when processing target content.
Capability inventory: The skill requires running 'existing checks' and 'representative tasks' (Step 2: Freeze; Step 8: Verify), which implies executing shell commands or scripts provided by the target skill.
Sanitization: The skill lacks explicit sanitization or validation requirements for the content or scripts found within the target skill package before they are processed or executed.

simplify-skill