The Agent Skills Directory

[PROMPT_INJECTION]: The skill creates a surface for indirect prompt injection by design, as it ingests and promotes content from historical session data into permanent project files.
Ingestion points: The workflow relies on reading historical data from session_store search and prior session artifacts (SKILL.md).
Boundary markers: There are no explicit instructions for the agent to use delimiters or ignore embedded instructions when processing retrieved session notes.
Capability inventory: The skill facilitates writing and promoting content to sensitive repository configuration files such as .github/copilot-instructions.md and CONTRIBUTING.md (SKILL.md).
Sanitization: The skill provides a manual security guideline to remove secrets and machine-local details before exporting artifacts, which mitigates some data exposure risks.

knowledge-curator