Academic Paper Reviewer
ReviewAudited by ClawScan on May 16, 2026.
Overview
Prompt-injection indicators were detected in the submitted artifacts (system-prompt-override); human review is required before treating this skill as clean.
Before installing, confirm you are allowed to submit the manuscript to an AI provider and remove confidential details that are not needed. Use a dedicated folder for review inputs and outputs, and be aware that some referenced agent filenames appear inconsistent with the shipped manifest, which may affect reliability. ClawScan detected prompt-injection indicators (system-prompt-override), so this skill requires review even though the model response was benign.
Findings (4)
Artifact-based informational review of SKILL.md, metadata, install specs, static scan signals, and capability signals. ClawScan does not execute the skill or run runtime probes.
Ambiguous prompt-hierarchy wording could confuse maintainers or runtimes, but the visible skill behavior remains focused on paper review and does not expose high-impact tools.
This system-prompt wording can look like hierarchy manipulation, but in context it is part of a delegated reviewer protocol rather than an instruction to override the user's agent or this evaluator.
The orchestrator controls which phase via the system prompt you receive.
Keep this wording clearly scoped as subagent prompt-template text, and consider rephrasing 'system prompt' to 'subagent instruction' to reduce false-positive and hierarchy-confusion risk.
The skill may read manuscript files and create or modify review-output files in the workspace.
The skill grants delegated agents file access, which is expected for reading manuscripts and writing review artifacts, and it explicitly excludes broader terminal, web, and system tools.
Subagents are granted only `file` tools for reading/writing review outputs. No terminal, web, or system tools are exposed.
Use a dedicated workspace containing only the paper and related files you want the reviewer agents to access.
Some delegated reviewer tasks may run without the intended context file, reducing reliability or causing incomplete reviews.
The manifest provided does not list these exact referenced files, suggesting a packaging or reference-name mismatch rather than hidden code execution.
Use agents/evidence_reviewer.md ... Use agents/argument_reviewer.md ... Use agents/revision_coach.md
The publisher should align SKILL.md context paths with the shipped agent files before users rely on the workflow.
Unpublished papers, author information, or prior review materials may be processed by the AI provider's delegated-agent system.
The skill explicitly sends paper content and intermediate outputs through a delegated-agent workflow; this is central to its purpose and is disclosed to users.
This skill delegates review tasks across multiple subagents via `delegate_task`. Manuscript content and intermediate review outputs are processed by these agents.
Use it only for manuscripts you are authorized to share with the AI provider, and remove confidential or unnecessary material before review.