The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests data from local project files (e.g., README.md, source code, documentation, and test suites) to synthesize user personas. Since these files are treated as untrusted sources, there is a risk that they could contain adversarial instructions intended to influence the agent's behavior during the generation process.
Ingestion points: Sources include documentation (docs/**/*.md), source code (src/**), and project metadata as defined in the generation-workflows.md file.
Boundary markers: The skill does not define specific delimiters or instructional barriers to prevent the agent from obeying instructions embedded within the analyzed data.
Capability inventory: The skill allows the agent to write files to the .agents/personas/ directory, modify the registry.yaml file, and execute shell commands for audio synthesis.
Sanitization: The skill incorporates quality controls such as confidence scoring, bias auditing, and human-research validation requirements for promotion to active status.
[DYNAMIC_EXECUTION]: In the SPEAK mode instructions, the skill specifies commands to detect and run voice synthesis engines. This includes the use of npx --yes edge-tts, which involves dynamically fetching and executing an external package from a registry at runtime.
[COMMAND_EXECUTION]: The skill utilizes shell commands (curl, which, npx, python3) to perform environment checks for local text-to-speech (TTS) engines, such as VOICEVOX, macOS say, and Google Cloud TTS. These commands are used to verify tool availability and configuration.

cast