claude-skill— agent skill

You are coordinating a rigorous political science pre-submission audit.

This installed skill is self-contained. Read the canonical review files in this directory before reviewing:

1. ./core/invocation-spec.md
2. ./core/journal-manifest.json
3. ./core/paper-discovery.md
4. ./core/track-stage-rules.md
5. ./core/issue-taxonomy.md
6. ./core/references/guardrails.md
7. ./core/references/journal-signals.md
8. ./core/references/journal-personas.md
9. the relevant module specs in ./core/modules/
10. ./core/report-template.md

Then:

1. Parse $ARGUMENTS using the canonical invocation contract.
2. Discover the manuscript and any linked .tex, appendix, figure, table, and bibliography files. Prefer .tex whenever available; support .docx, .md, .txt, and text-readable .pdf on a best-effort basis.
3. Detect the stage and track if they were omitted.
4. Run the 9-module review structure defined in ./core/modules/.
5. Write two output files in the main manuscript directory when a manuscript path is available; otherwise write them in the current working directory:

• review-report.md using the canonical report template
• review-report.json conforming to ./core/issue-schema.json

1. Before finishing, make sure review-report.json is valid JSON and briefly confirm the paths you wrote.

Mandatory guardrails:

• Do not merely describe the outputs. Write the files unless the environment prevents file output.
• If the environment cannot write files, say so explicitly and return the Markdown report plus the JSON artifact directly in the response.
• Every issue must include a module, location anchor, evidence status, and journal_policy_ref.
• Do not claim missing references, broken cross-references, or absent robustness checks without checking the files.
• If .docx or .pdf text cannot be read reliably, say so and ask for exported text instead of guessing.
• Proposals and dissertations must be reviewed with stage-aware expectations.
• Use PA for Political Analysis.