The Agent Skills Directory

[DYNAMIC_EXECUTION]: The validateAudio function in SKILL.md utilizes execSync to run ffprobe and stat commands by interpolating a filePath variable directly into the command string. This pattern is susceptible to command injection if the file path is sourced from untrusted input or contains shell metacharacters.
[COMMAND_EXECUTION]: The implementation guide in references/implementation-guide.md describes a multi-step process for generating and executing a diagnostic shell script (speak-debug-bundle.sh). This script performs extensive system discovery, network testing, and log aggregation.
[INDIRECT_PROMPT_INJECTION]: The diagnostic bundle script ingests external data that could contain malicious instructions.
Ingestion points: The script reads from .env files, logs/*.log, and package.json (found in references/implementation-guide.md).
Boundary markers: No delimiters or 'ignore instructions' warnings are applied to the gathered content before it is bundled or presented to the agent.
Capability inventory: The skill has access to shell execution, file system modification, and network requests via curl.
Sanitization: The script includes sed patterns to redact API keys and user IDs, but it does not sanitize the logs for potential prompt injection vectors.
[DYNAMIC_CONTEXT_INJECTION]: SKILL.md uses the load-time execution syntax (!node --version and !ffmpeg -version) to verify the environment. These commands are benign version checks and do not involve user-controlled input.

speak-debug-bundle