The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests data from external design and asset files to brief its sub-agents, creating a surface where malicious instructions in those files could be executed by the agents.\n
Ingestion points: Files in design/gdd/, design/audio/, and assets/audio/ are read during the 'Gather context' phase.\n
Boundary markers: Absent; the skill lacks formal delimiters or 'ignore' instructions when interpolating file content into sub-agent prompts.\n
Capability inventory: The skill uses tools like Write, Edit, and Bash, and sub-agents like the gameplay-programmer are authorized to modify source code.\n
Sanitization: Absent; the skill does not escape or validate the content of the design documents before processing.\n- [DYNAMIC_CONTEXT_INJECTION]: The skill uses the !bash syntax to execute a project-local script during the load process to resolve configuration values.\n
Evidence: !bash "${CLAUDE_SKILL_DIR}/../../hooks/yaml-helper.sh" resolve_config --keys review_mode,automation,team.size.\n- [COMMAND_EXECUTION]: The skill instructs sub-agents to perform file system writes without the standard user confirmation step normally required by the platform's Collaboration Protocol.\n
Evidence: The "File Write Protocol" section defines a "bounded exception" where sub-agents can write to production/, docs/, and tests/ directories without triggering a confirmation prompt.\n- [PROMPT_INJECTION]: The skill contains explicit instructions to sub-agents to override the platform's safety guidelines regarding tool confirmation.\n
Evidence: The prompt explicitly states: "A subagent spawned here writes without asking, and that is a deliberate, bounded exception rather than an oversight."

team-audio