The Agent Skills Directory

[SAFE]: The skill implements a structured workflow for simulating the academic peer review process. It uses specialized agents (EIC, methodology, domain, etc.) to analyze manuscripts. All external references are to well-known academic publishing standards and reputable journals.
[PROMPT_INJECTION]: While the detector flagged instructions to 'ignore' or 'override' previous instructions in agent system prompts, these are internally scoped boundary markers (Phase Boundary) intended to maintain agent role consistency and prevent cross-contamination between different reviewer personas in the pipeline. These are benign and follow established design patterns for multi-agent systems.
[EXTERNAL_DOWNLOADS]: The SKILL.md file points to github.com/Imbad0202/academic-research-skills for its source code. This is an informative reference to the skill's own repository and does not trigger malicious download patterns. The skill also references well-known services like Springer, Elsevier, Nature, and Taylor & Francis in its journal directory for role calibration purposes, which is consistent with its primary purpose.
[COMMAND_EXECUTION]: The skill includes a reference to an advisory verifier script scripts/check_pipeline_integrity.py. This is documented as a development-time tool for ensuring the integrity of the agent workflow and is not executed automatically by the agent to perform system-level changes.
[DATA_EXPOSURE]: The skill uses an IRON RULE for untrusted review materials, explicitly stating that embedded instructions in submitted manuscripts must not alter the agent's identity or disclosure rules. This demonstrates a robust security posture against indirect prompt injection from user-provided data.

academic-paper-reviewer