The Agent Skills Directory

[DATA_EXFILTRATION]: The skill accesses local configuration files and project data, including the pet profile in ~/.claude.json, conversation history, and markdown files within ~/.claude/projects/*/memory/. This data is used to generate personalized lyrics and prompts which are then sent to the vendor's external API via the mmx CLI. While this is the primary purpose of the skill, it involves the transmission of potentially sensitive project-related context to a third-party service.\n- [COMMAND_EXECUTION]: The skill executes multiple system commands to facilitate its workflow, including find for locating memory files, git log for retrieving recent development activity, and various audio players (mpv, ffplay, afplay) for playback of generated music files.\n- [DYNAMIC_EXECUTION]: The skill dynamically constructs shell commands for music generation and playback using variables derived from the pet's name, personality profile, and generated content. It also utilizes computed file paths for storing vocal identity cache files in ~/.claude/skills/buddy-sings/voices/ and generated audio in ~/Music/minimax-gen/.\n- [INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted content from external sources (memory files and git history) to influence the generation of song lyrics and musical styles, creating an attack surface for indirect prompt injection.\n
Ingestion points: ~/.claude/projects/*/memory/*.md, ~/.claude/memory/, and git log output (referenced in SKILL.md).\n
Boundary markers: None. The skill does not employ delimiters or specific instructions to isolate ingested data from the core system prompts during processing.\n
Capability inventory: Subprocess execution for mmx, mpv, ffplay, afplay, git, and find (referenced in SKILL.md).\n
Sanitization: None. Content extracted from local files is interpolated into prompts without validation, filtering, or escaping.

buddy-sings