The Agent Skills Directory

[PROMPT_INJECTION]: The skill instructions specify that the agent should read external context from files such as .agents/product-marketing.md or .claude/product-marketing.md at the start of a task. This creates a surface for indirect prompt injection, where adversarial instructions placed in these project files could influence the agent's behavior.
Ingestion points: SKILL.md (Initial Assessment section) identifies multiple potential file paths for context ingestion.
Boundary markers: The skill does not provide instructions to use delimiters or ignore embedded directives when processing these files.
Capability inventory: The skill is limited to providing UI/UX recommendations and does not possess capabilities for file writing, command execution, or network operations.
Sanitization: There is no requirement or logic provided for sanitizing or validating the content of the ingested files.
[NO_CODE]: The skill consists exclusively of markdown documentation and JSON evaluation files. It does not contain any executable scripts, binaries, or automated tools, which significantly limits the potential for direct malicious impact.

paywalls