The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process external source material including notes, transcripts, and rough drafts. This represents a potential vector for indirect prompt injection if the source content contains malicious instructions intended to manipulate the agent's behavior during the drafting or critique stages.
Ingestion points: File references/stages/intake.md specifies that notes, transcripts, and conversations are valid sources for the content pipeline.
Boundary markers: There are no explicit instructions or delimiters defined to isolate the source material or to instruct the agent to ignore instructions embedded within the processed data.
Capability inventory: The skill leverages file-write capabilities in references/stages/approve.md and references/stages/transform.md.
Sanitization: No sanitization, validation, or escaping logic is applied to the input source material before it is used in the pipeline.
[COMMAND_EXECUTION]: In references/stages/approve.md and references/stages/transform.md, the instructions command the agent to automatically write files to the local file system at ~/Desktop or a path defined by the $KERNEL_WRITE_OUTPUT_DIR environment variable. The skill explicitly directs the agent to perform these writes and overwrite existing files of the same version, even if the user has not specifically requested a file creation task.

content-pipeline