The Agent Skills Directory

[PROMPT_INJECTION]: The skill exhibits an indirect prompt injection surface. It processes untrusted data from feedback recordings, including transcripts and event logs, and incorporates this content into markdown scaffolds such as analysis.md and problem-analysis.md. These scaffolds are subsequently used to guide agent behavior during brainstorming and GitHub issue creation. Malicious instructions embedded within the feedback data could attempt to manipulate the agent's downstream actions.
Ingestion points: scripts/analyze_riffrec_zip.py extracts data from user-provided .zip bundles and media files.
Boundary markers: The generated markdown output does not consistently use explicit boundary markers or "ignore instructions" delimiters to isolate untrusted content from agent instructions.
Capability inventory: The agent has access to gh issue create, local file system writes, and execution of local scripts across the provided references.
Sanitization: The skill performs basic slugification and includes robust path traversal checks (ZipSlip protection) in its zip extraction logic.
[COMMAND_EXECUTION]: The skill requires the execution of multiple system commands to process feedback. The agent is instructed to use the GitHub CLI (gh) for repository authentication and issue management. The included Python script (scripts/analyze_riffrec_zip.py) invokes system utilities including ffmpeg, ffprobe, and curl. While these invocations use list-based arguments to mitigate shell injection risks, they operate on paths and data derived from untrusted user inputs.
[DATA_EXFILTRATION]: The skill transmits media data to the OpenAI API (https://api.openai.com/v1/audio/transcriptions) for transcription purposes using curl. This operation utilizes the OPENAI_API_KEY from the user's environment. While this is the intended functionality for transcribing feedback, it constitutes a deliberate transfer of potentially sensitive audio and video data to a well-known third-party service.

riffrec-feedback-analysis