The Agent Skills Directory

[DYNAMIC_CONTEXT_INJECTION]: The skill utilizes shell execution at load time via the ! syntax in Phase 0 to run a local script (yaml-helper.sh). The script is targeted using relative path traversal (../../), which accesses hooks outside the skill's own directory.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted design documents and asset lists, passing the distilled content to sub-agents, creating a vulnerability surface for embedded instructions. 1. Ingestion points: SKILL.md (Step 2) reads design docs from design/gdd/ and asset lists from assets/audio/. 2. Boundary markers: Absent; the skill does not use delimiters or instructions to ignore embedded commands in the ingested files. 3. Capability inventory: Orchestrator and sub-agents have access to Write, Edit, and Bash tools. 4. Sanitization: No sanitization or validation of the ingested document content is performed before interpolation.
[COMMAND_EXECUTION]: The skill requires the Bash tool for configuration and technical implementation tasks. It explicitly directs sub-agents to bypass standard Collaboration Protocol write confirmations for artifacts in production/, docs/, and tests/ directories, increasing the potential for autonomous unauthorized modifications.

team-audio