sws

Use this agent when /sws:audit-bibliography is invoked. AUDIT agent: validates the manuscript's EXISTING citations before submission. Resolves DOIs, deduplicates entries, flags format deviations vs the resolved refs_style, proposes fixes in fixes.json. Does NOT write to the manuscript .docx — fixes are proposed only (Review-Then-Act D11). Fallback chain: Zotero -> CrossRef -> OpenAlex. NLM deferred (D9). Active in all 9 profiles (bibliography hygiene is universal — D10).

Use this agent when /sws:check-fidelity is invoked or as the middle stage of /sws:review-paper. Renamed from "plagiarism-screener" in roster v0.1 (2026-05-17) because the v0.1 scope is bounded-corpus fidelity (Zotero full-text) not unbounded plagiarism. Runs scripts/sws_bibliography_fidelity.py which detects four code paths: D9 happy (Zotero corpus available — runs verbatim ≥15-word overlap), D9a (Zotero desktop detected but Claude Code zotero skill missing — actionable recommendation), D9b (skill present but library small/unresponsive/permission-denied), D9c (no Zotero anywhere — neutral note). Writes _review/bibliography-fidelity-checker/{report.md, flags.json, status.json}. Diagnose only. Funding-proposal profile is inactive.

Use this agent when the user invokes /sws:draft-section figure_caption (or per-figure caption requests), or when /sws:draft-paper fans out caption work. Reads each figure file via the native Read tool (multimodal image input), reads the supporting outline section's key_claims, and writes caption text directly into the outline.md frontmatter under each figure entry. Text only — no docx editing, no alt-text.

Use this agent when /sws:verify-claims is invoked or when /sws:review-paper starts its pipeline. Runs scripts/sws_claim_extract.py to harvest citation-bearing claims from _drafts/*-revised.md, then verifies each claim against (1) the user's Zotero library via the existing zotero skill, (2) Semantic Scholar (WebFetch), (3) PubMed (claude_ai_PubMed MCP). NLM consumer wiring deferred to cycle #11; the agent degrades gracefully when notebooklm.enabled=false. Writes _review/claim-verifier/report.md + claims.json. Diagnose only — never writes to the manuscript. Funding-proposal profile is inactive (proposals have forward-looking claims, not verifiable assertions).

Use this agent when a major project step has been completed and needs to be reviewed against the original plan and coding standards. Examples: <example>Context: The user is creating a code-review agent that should be called after a logical chunk of code is written. user: "I've finished implementing the user authentication system as outlined in step 3 of our plan" assistant: "Great work! Now let me use the code-reviewer agent to review the implementation against our plan and coding standards" <commentary>Since a major project step has been completed, use the code-reviewer agent to validate the work against the plan and identify any issues.</commentary></example> <example>Context: User has completed a significant feature implementation. user: "The API endpoints for the task management system are now complete - that covers step 2 from our architecture document" assistant: "Excellent! Let me have the code-reviewer agent examine this implementation to ensure it aligns with our plan and follows best practices" <commentary>A numbered step from the planning document has been completed, so the code-reviewer agent should review the work.</commentary></example>

Build the R&R Traceability Matrix and dispatch the response-to-reviewers agent (Opus 4.7) on _review/round-<N>/. Default round is the highest existing N; pass --round to target a specific round. Two-step pipeline: sws_response_matrix.py (deterministic parse, no LLM) -> response-to-reviewers agent (rebuttal prose + matrix fill). Output mirrored to _submission/ for journal upload. Cost-gated by R1 — runs Opus, so the prompt warns to use only after a finalized revision.

Sequential orchestrator that runs the SWS review pipeline end-to-end: /sws:verify-claims → /sws:check-fidelity → /sws:peer-review. Passes report paths from the first two into peer-reviewer via explicit CLI args (no autoscan). Propagates SWS_FIDELITY_STATUS to peer-reviewer so the peer-review report transparently notes when fidelity was skipped. Inactive in funding-proposal profile (component skills' profile gates also fire individually).

This skill should be used when the user invokes /sws:revise-paper, says 'revise the paper', 'polish the draft', 'run the revision pipeline', 'prepare the final docx', or similar — and the cwd is an SWS project with drafted sections at _drafts/<section>.md. Runs a sequential four-pass pipeline: consistency-checker → reviser (full or fast) → humanizer → style-enforcer, producing Manuscript/<paper-name>.docx.

This skill should be used when the user invokes /sws:revise-section <section-id>, says 'revise the intro', 'polish the results section', 'clean up <section>' — and the cwd is an SWS project with at least one drafted section in _drafts/. Always dispatches reviser-fast (D18 — no --full flag in v0.1). Optionally skips the humanizer pass via --no-humanize.

Master orchestrator. Runs the full SWS pipeline end-to-end: outline → draft → revise → review → cover-letter → ai-disclosure → response-to-reviewers. Each step is skipped when its artifacts already exist (D9 idempotency). Supports --dry-run (plan only) and --only=<csv> (subset). Writes a single passport entry with phase=submit summarising the steps dispatched.

3 hooks across 3 events