PaperRead-Skills

Analyze uploaded CVPR paper PDFs with a strict two-pass workflow. First pass extracts verifiable facts with page/section/figure/table evidence only. Second p...

Security Scan

The skill is an instruction-only two-pass PDF reviewer that only needs the uploaded PDF and local prompt files; its requirements and instructions are consistent with its stated purpose.

Purpose & Capability

Name/description match the actual behavior: it analyzes uploaded CVPR PDFs via two local prompt files. No unrelated credentials, binaries, or external services are requested.

Instruction Scope

Runtime instructions are narrowly scoped to (1) confirm an uploaded PDF in the chat, (2) run the two local prompts to extract facts and produce a critique, and (3) restrict outputs to PDF evidence. The skill does not instruct reading unrelated files, accessing env vars, or contacting external endpoints.

Install Mechanism

This is an instruction-only skill with no install spec and no code files to execute; nothing is written to disk or downloaded during install.

Credentials

No environment variables, credentials, or config paths are requested. The declared inputs align with the paper-analysis purpose.

Persistence & Privilege

always is false and there is no request to modify other skills or system-wide settings. disable-model-invocation is default (false), which is normal for skills and not, by itself, a concern.

Assessment

This skill is internally consistent and only needs the PDF you upload plus the included prompt files. Before using it, consider: (1) do not upload confidential or unpublished manuscripts you don't want processed by the agent or potentially retained in logs; (2) the skill intentionally forbids external sources, but model behavior may still leak background knowledge—verify outputs against the PDF evidence tags; (3) the second pass requests a long (2000–3000 Chinese-word) critique—ensure you actually want that length. If you need the skill to never run autonomously, disable autonomous invocation in your agent settings.

CVPR PDF Two-Pass Review

Overview

Run a deterministic two-pass paper analysis workflow on an uploaded PDF. Prioritize evidence-grounded outputs and reduce hallucinations by separating fact extraction from interpretation.

Workflow

1. Confirm the user has uploaded the PDF in the current chat. If not, ask them to upload it first.
2. Load and execute prompt in references/round1.md to produce a structured fact sheet.
3. Load and execute prompt in references/round2.md using only round-1 facts plus PDF evidence.
4. Enforce evidence tags for core claims: [Evidence: p.X, Sec.Y, Fig/Table Z].
5. If any required information is missing, explicitly state Not provided in paper or Insufficient evidence.

Output Rules

• Keep analysis strictly based on uploaded PDF content.
• Do not introduce external papers, benchmarks, or assumptions unless user explicitly asks.
• Keep language concise and technical.
• Preserve required length in pass 2: 2000-3000 words or characters as requested by the prompt.

References

• Round 1 prompt: references/round1.md
• Round 2 prompt: references/round2.md