The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted user-provided proposals and summaries in Step 0 and interpolates this data into persona-based evaluations and a generated HTML presentation deck. There are no explicit instructions to sanitize the input or use strict boundary markers to prevent malicious instructions embedded in a proposal from influencing the agent's review logic or the contents of the generated files.
Ingestion points: Technical proposal summaries and file paths provided by the user in Step 0.
Boundary markers: Absent; the skill directly incorporates content into slides and evaluation prompts.
Capability inventory: The skill utilizes the Write tool to create HTML artifacts on disk and the Agent tool to trigger external validation modules (/sim-conductor).
Sanitization: No specific sanitization, escaping, or filtering of user-provided content is mentioned before interpolation.

apex-review