The Agent Skills Directory

[DATA_EXFILTRATION]: The skill instructions direct the agent to read sensitive local workspace files, such as MEMORY.md and daily activity logs. This content is synthesized and transmitted to an external service (pika.art) to provide context for the virtual meeting agent.
[COMMAND_EXECUTION]: The skill utilizes subprocess.run to execute the ffmpeg binary for audio format conversion during the voice cloning process.
[CREDENTIALS_UNSAFE]: The script accesses the sensitive path ~/.pika/devkey to retrieve API credentials and also checks the PIKA_DEV_KEY environment variable.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from workspace files and meeting links to generate a system prompt for the meeting bot, creating a surface for indirect injection attacks.
Ingestion points: Reads MEMORY.md, daily logs, and identity files from the local workspace.
Boundary markers: None identified; the synthesis process does not use delimiters or instructions to ignore embedded commands.
Capability inventory: The skill can execute shell commands via subprocess.run (ffmpeg) and perform network requests to external APIs.
Sanitization: No sanitization or filtering is applied to the content of the workspace files before they are included in the prompt synthesis.
[EXTERNAL_DOWNLOADS]: The skill downloads user-provided images from remote URLs and installs Python dependencies via pip at runtime.

pikastream-video-meeting