The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill establishes an attack surface for indirect prompt injection by processing external data as a core part of its workflow.
Ingestion points: The skill ingests untrusted data from the user-provided source_path (articles, transcripts, notes) and repo_path (git history archaeology) in SKILL.md.
Boundary markers: The orchestrator does not define explicit delimiters or instructions to ignore potential commands embedded within the source materials, which could influence the agent's behavior during the extraction and synthesis stages.
Capability inventory: The skill leverages the Read and Write tools to process these inputs and generate various summary, analysis, and script files, allowing untrusted input to directly affect the content and structure of the generated files.
Sanitization: There is no evidence of sanitization, filtering, or validation logic applied to the external content before it is processed by the agent or the sub-skills it invokes.

talk-pipeline