The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data in the form of AI session transcripts.
Ingestion points: The skill reads session logs (e.g., .jsonl files) via mcp__frustration-analyzer__list_sessions and Read tools to extract message content.
Boundary markers: There are no specific instructions or delimiters provided to the primary agent or the subagents (frustration-analyzer:batch-detector, frustration-analyzer:context-reconstructor) to ignore instructions embedded within the transcript text.
Capability inventory: The skill possesses the ability to Read and Write files, perform Glob operations, and execute multiple MCP tools that handle sensitive session data.
Sanitization: The instructions do not specify any sanitization, escaping, or filtering of the transcript content before it is processed by LLM subagents, creating a surface where malicious instructions inside a transcript could influence the agent's behavior.
[DATA_EXFLLTRATION]: While the skill does not automatically send data to a remote server, its primary purpose is to extract and reconstruct potentially sensitive transcript data into a PNG format specifically described as "ready for social media." This workflow facilitates the movement of private session data (which may include code, keys, or personal info) from a protected local environment to a format and context intended for external public disclosure.

rtfp