The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill implements automated review and reporting features that ingest and process previously logged user content without sanitization or rigorous data boundaries.
Ingestion points: The skill captures raw user utterances in data/inputs_log.md and data/errors_tagged.md. Additionally, it populates the data/methodology.md file (used for behavioral modeling) from external text provided via the /pte feed command.
Boundary markers: During the 'Session-end batch review' and 'Weekly report' generation, the agent interpolates historical data using simple Markdown formatting but lacks robust delimiters or 'ignore' instructions for the processed strings.
Capability inventory: The agent has the ability to modify its own conversational output based on stored rules and to update local state and data files based on these summaries.
Sanitization: There is no evidence of a sanitization or validation layer to filter potential prompt injection patterns from user-provided content before it is processed in the higher-privilege review context.

english-immersion