The Agent Skills Directory

[SAFE]: The skill is designed for complex document management within the scientific peer-review process. Its behavior is consistent with its stated purpose of parsing reviews and drafting grounded responses.
[SAFE]: File system operations are restricted to the local rebuttal/ directory and designated input paths. No attempts to access sensitive system files or credentials (e.g., .ssh, .aws, .env) were found.
[PROMPT_INJECTION]: The skill is designed to ingest and process untrusted data in the form of external reviewer comments. While this creates an inherent surface for indirect prompt injection, the risk is mitigated by the skill's multi-stage safety model and the primary purpose of the task.
Ingestion points: External reviewer comments are parsed and normalized into rebuttal/REVIEWS_RAW.md during Phase 1.
Boundary markers: The skill does not explicitly mention the use of delimiters (e.g., XML tags) to isolate untrusted reviewer text, but it atomizes concerns into structured issue cards in ISSUE_BOARD.md to maintain control over the processing logic.
Capability inventory: The skill possesses high-privilege capabilities including Bash(*), Write, Edit, and the ability to call other Agent and Skill components (e.g., /experiment-bridge).
Sanitization: The skill employs three 'hard gates' (Provenance, Commitment, and Coverage) to validate that every statement in the draft is grounded in the paper or user-confirmed evidence before finalization.

rebuttal