The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill reads data from a local file and interpolates it into the output prompt.
Ingestion points: The version field is read from ${CLAUDE_PLUGIN_ROOT}/.claude-plugin/plugin.json.
Boundary markers: There are no explicit delimiters or instructions to ignore embedded content within the version string.
Capability inventory: The skill is restricted to the Read tool in its frontmatter configuration.
Sanitization: No sanitization or validation is performed on the content retrieved from the JSON file before it is displayed to the user.
[DYNAMIC_EXECUTION]: The documentation in PROJECT_HISTORY.md refers to architectural patterns for self-modifying or dynamically updated instructions.
The document states that "Skills update themselves when they learn new things" and mentions a "Model updater" that keeps instructions current.
While this specific SKILL.md does not implement these features, the design philosophy indicates a risk surface for persistence or logic modification through automated script updates.
[DATA_EXFILTRATION]: The project documentation describes the use of sensitive local file paths for configuration management.
PROJECT_HISTORY.md mentions the use of ~/.bitwize-music/config.yaml for storing user preferences and settings.
Although the current skill only reads from the plugin's internal directory, the project design encourages the storage and use of data within the user's home directory.

about