The Agent Skills Directory

[SAFE]: The skill uses allowed tools Read, Glob, and Grep to access local project files and a vendor-specific tool bitwize-music-mcp for database-like interactions (e.g., get_session, list_albums). All referenced skills (e.g., /bitwize-music:lyric-writer) belong to the author's own namespace and represent legitimate workflow transitions.
[INDIRECT_PROMPT_INJECTION]: The skill has a surface for indirect prompt injection as it ingests untrusted data from external project files to make workflow decisions.
Ingestion points: Reads track frontmatter, SOURCES.md, and track detail tables using Read, Glob, and Grep tools.
Boundary markers: The instructions do not define explicit delimiters for the data being read, instead directing the agent to look for specific keys like instrumental: true or Album Status strings.
Capability inventory: The skill can read files, list files, and invoke specific music-production tools via bitwize-music-mcp. It does not possess direct network access or arbitrary command execution capabilities.
Sanitization: The skill logic relies on pattern matching and status lookups. While it doesn't explicitly sanitize file content, the restricted capability set and the task-specific nature of the tool invocations minimize the risk of malicious exploitation.

next-step