---

name: section-logic-polisher description: | Logic coherence pass for per-H3 section files: enforce an explicit thesis statement + inter-paragraph logical connectors before merging. Trigger: logic polisher, section logic, thesis statement, connectors, 段落逻辑, 连接词, 论证主线, 润色逻辑. Use when: sections/S*.md exist but read like paragraph islands; you want a targeted, debuggable self-loop before section-merger. Skip if: sections are missing/thin (fix subsection-writer first) or evidence packs/briefs are scaffolded (fix C3/C4 first). Network: none. Guardrail: do not add new citations; do not invent facts; do not change citation keys; do not move citations across subsections.

Section Logic Polisher (thesis + connectors)

Purpose: close the main “paper feel” gap that remains even when a subsection is long and citation-dense:

• missing/weak thesis (no central claim)
• weak inter-paragraph flow (paragraph islands; few logical connectors)

This is a local, per-H3 polish step that happens after drafting and before merging.

Role prompt: Logic Editor (argument flow)

CopyYou are the logic editor for one survey subsection.

Your job is to make the subsection read like a single argument:
- paragraph 1 ends with a clear thesis (content claim)
- each paragraph has an explicit logical relation to the previous one
- connectors are semantic (contrast/causal/extension), not slide narration

Constraints:
- do not add new citations
- do not change citation keys
- do not invent facts

Editing lens:
- if a paragraph does not advance the argument (claim/contrast/eval/limitation), compress or delete it
- if a transition is empty, rewrite it as a content-bearing bridge

Inputs

• sections/ (expects H3 body files like S_.md)
• outline/subsection_briefs.jsonl (uses thesis + paragraph_plan[].connector_phrase)
• Optional: outline/writer_context_packs.jsonl (preferred; has trimmed anchors/comparisons + must_use)

Outputs

• output/SECTION_LOGIC_REPORT.md (PASS/FAIL report for thesis + connector density)

Manual / LLM-first (in place):

• Update the H3 files under sections/ to fix thesis/connectors (no new citations; keep keys stable)

Workflow (self-loop)

1. Run the checker script to surface the exact failing files and why.
2. For each failing H3 file:
   • Thesis: ensure the first paragraph ends with a clear thesis sentence.
     • Prefer a conclusion-first takeaway (content claim, not meta-prose). Avoid repetitive openers like This subsection argues/surveys ....
     • Minimal shape (3 sentences): 1) viewpoint 2) why it matters 3) how the subsection is organized.
     • Use the brief’s thesis as the source of truth (from outline/subsection_briefs.jsonl; paraphrase is OK, same meaning).
   • Flow: ensure the subsection contains explicit logical connectors across paragraph boundaries (they can be paragraph-initial or mid-sentence).
     • Treat paragraph_plan[].connector_phrase as semantic intent, not a literal template; paraphrase and vary wording.
     • Avoid “PPT narration” connectors (Next, we ...); prefer argument ties (In contrast, ..., This suggests ..., As a result, ...).
     • If available, prefer the merged pack (outline/writer_context_packs.jsonl) so your edits stay aligned with must_use (anchors/comparisons/limitations).
     • Aim to include causal + contrast + extension signals, but don’t start every paragraph with the same stem (e.g., “However”).
   • Guardrails: do not add/remove citation keys; do not introduce new factual claims.
3. Re-run the checker until output/SECTION_LOGIC_REPORT.md is PASS, then proceed to transition-weaver and section-merger.

Examples (thesis + connectors)

Thesis signal (paragraph 1)

Bad (topic setup only):

• Tool interfaces vary across agent systems, and many recent works explore different designs.

Better (conclusion-first claim):

• A central tension in tool interfaces is balancing expressivity with verifiability; as a result, interface contracts often determine which evaluation claims transfer across environments.

Bad (meta narration):

• This subsection argues that memory is important for agents.

Better (content claim):

• Memory designs trade off retrieval reliability against write-time contamination, and this trade-off shows up as distinct failure modes under fixed evaluation protocols.

Connectors (avoid paragraph islands)

Bad (paragraphs do not relate):

• X does ... (para 2)
• Y does ... (para 3)

Better (explicit logical tie):

• Whereas X optimizes for <axis>, Y shifts the bottleneck to <axis>; as a result, their reported gains are not comparable unless the protocol controls for <constraint>.

Done criteria

• output/SECTION_LOGIC_REPORT.md shows - Status: PASS
• No section file contains placeholders (TODO/…/...) or outline meta markers (Intent:/RQ:/Evidence needs:)
• Thesis + connector checks pass for all H3 files

Script

Quick Start

• python .codex/skills/section-logic-polisher/scripts/run.py --help
• python .codex/skills/section-logic-polisher/scripts/run.py --workspace workspaces/<ws>

All Options

• --workspace <dir>
• --unit-id <U###>
• --inputs <semicolon-separated>
• --outputs <semicolon-separated>
• --checkpoint <C#>

Examples

• Default IO:
  • python .codex/skills/section-logic-polisher/scripts/run.py --workspace workspaces/<ws>
• Explicit output path:
  • python .codex/skills/section-logic-polisher/scripts/run.py --workspace workspaces/<ws> --outputs "output/SECTION_LOGIC_REPORT.md"

Troubleshooting

Issue: report says thesis missing but the subsection has an intro sentence

Fix:

• The checker looks for an explicit thesis/takeaway signal in the first paragraph (not just topic setup).
• Add a single takeaway sentence at the end of paragraph 1 (copy from briefs; do not add new facts/citations).

Issue: connector density fails but the prose reads fine

Fix:

• This check is a proxy for “paragraph islands”. If it FAILs, add a few explicit connector signals (contrast/causal/extension/implication).
• Keep it natural: connectors can be mid-sentence (...; however, ...), and you don’t need to start every paragraph with a connector word.
• Prefer short ties; don’t add long filler sentences.