v0.3 — Two-Stage Manuscript Review Protocol

What's new

Adds a specialized Two-Stage Manuscript Review Protocol for peer review of academic manuscripts.

Pre-flight

Characterize the target journal (tier, indexing, audience, typical methodological standard) before reviewing. Critique severity is calibrated to the journal's published bar, not to top general-science venues.

Stage 1 — Summary, Logic Model, Questions

• Plain-language summary of the contribution
• Verbatim research questions / hypotheses
• A Mermaid logic model: theory → predictions → design → measures → analysis → findings → claims, with explicit "inferential leap" branches
• Evidence map (claim → evidence cited → strength)
• 5–10 categorized provocations (logical / evidential / framing / structural)
• 3–6 targeted questions for the human reviewer

Stops and waits for the reviewer's answers. Stage 1 is neutral-to-mildly-skeptical inventory, not a verdict.

Stage 2 — Review in the Reviewer's Voice

Two-letter format:

• To Editor: short, compresses the paper's contribution into two sentences, conveys recommendation through tone
• To Authors: opens with what's genuinely strong, 3–5 named-header sections with numbered points (location pointer + why it matters + recommended literature where useful), closes constructively

Also in v0.3

• New manuscript_review_protocol.md defining the protocol
• New examples/manuscript-review-stage2.md worked example (redacted from a real review)
• Routing from the Academic Manuscript query type in Skill.md to the new protocol
• CLAUDE.md technique table caught up with three rows already in Skill.md (Manuscript Reviewer, Critique Taxonomy, Survivorship Bias Check)
• README and docs site updated with a dedicated Manuscript Review section

Installation

Drop be-critical-v0.3.zip at claude.ai/settings/capabilities, or unzip into ~/.claude/skills/be-critical for Claude Code.