paper-writing · wanshuiyin — скилл для Claude Code: что делает и как установить

---
name: paper-writing
description: "Workflow 3: Full paper writing pipeline that goes from a narrative report to a polished, submission-ready PDF. Use when user says \"写论文全流程\", \"write paper pipeline\", \"从报告到PDF\", \"paper writing\", or wants the complete paper generation workflow."
argument-hint: "[narrative-report-path-or-topic] [— style-ref: <source>]"
allowed-tools: Bash(*), Read, Write, Edit, Grep, Glob, Skill, mcp__codex__codex, mcp__codex__codex-reply, mcp__antigravity__antigravity, mcp__antigravity__antigravity-reply, mcp__grok__grok, mcp__grok__grok-reply, mcp__claude-review__review, mcp__claude-review__review_reply, mcp__claude-review__review_start, mcp__claude-review__review_reply_start, mcp__claude-review__review_status
---

# Workflow 3: Paper Writing Pipeline

> **Before each reviewer slot**, follow [Selectable Reviewers](../shared-references/reviewer-backends.md):
> resolve invocation → `.aris/reviewer.json` → Claude Code default (Codex MCP)
> with `reviewer_route.py` through the canonical helper resolver, preserving this
> slot's tier, fresh contexts, source/image inputs and deterministic gates.
> Codex examples/pins below apply only to `backend=codex`; other routes use their
> own first-call/reply contract. Save the route and actual returned identity;
> unavailable selections are `REVIEW_UNAVAILABLE`. Classify acceptance from actual
> families/completion/scope. Forward invocation reviewer/model/effort overrides
> to reviewer-bearing children. Generation workers are unchanged.


Orchestrate a complete paper writing workflow for: **$ARGUMENTS**

## Overview

This skill chains five sub-skills into a single automated pipeline:

```
/paper-plan → /paper-figure → /paper-write → /paper-compile → /auto-paper-improvement-loop
  (outline)     (plots)        (LaTeX)        (build PDF)       (review & polish ×2)
```

Each phase builds on the previous one's output. The final deliverable is a polished, reviewed `paper/` directory with LaTeX source and compiled PDF.

In this hybrid pack, the pipeline itself is unchanged, but `paper-plan` and `paper-write` use Orchestra-adapted shared references for stronger story framing and prose guidance.

## Constants

- **VENUE = `ICLR`** — Target venue. Options: `ICLR`, `NeurIPS`, `ICML`, `CVPR`, `ACL`, `AAAI`, `ACM`, `IEEE_JOURNAL` (IEEE Transactions / Letters), `IEEE_CONF` (IEEE conferences). Affects style file, page limit, citation format.
- **MAX_IMPROVEMENT_ROUNDS = 2** — Number of review→fix→recompile rounds in the improvement loop.
- **REVIEWER_MODEL = `gpt-6-astra`** — Model used via Codex MCP for plan review, figure review, writing review, and improvement loop.
- **AUTO_PROCEED = true** — Auto-continue between phases. Set `false` to pause and wait for user approval after each phase.
- **HUMAN_CHECKPOINT = false** — When `true`, the improvement loop (Phase 5) pauses after each round's review to let you see the score and provide custom modification instructions. When `false` (default), the loop runs fully autonomously. Passed through to `/auto-paper-improvement-loop`.
- **ILLUSTRATION = `figurespec`** — Architecture/illustration generator for Phase 2b: `figurespec` (default, deterministic JSON→SVG via `/figure-spec`, best for architecture/workflow/topology), `gemini` (AI-generated via `/paper-illustration`, best for qualitative method illustrations; needs `GEMINI_API_KEY`), `codex-image2` (AI-generated via `/paper-illustration-image2` through the local Codex native image bridge — no external API key, uses your ChatGPT Plus/Pro quota; experimental), `mermaid` (Mermaid syntax via `/mermaid-diagram`, free, best for flowcharts), or `false` (skip Phase 2b, manual only).

> Override inline: `/paper-writing "NARRATIVE_REPORT.md" — venue: NeurIPS, illustration: gemini, human checkpoint: true`
> IEEE example: `/paper-writing "NARRATIVE_REPORT.md" — venue: IEEE_JOURNAL`

## Inputs

This pipeline accepts one of:

1. **`NARRATIVE_REPORT.md`** (best) — structured research narrative with claims, experiments, results, figures
2. **Research direction + experiment results** — the skill will help draft the narrative first
3. **Existing `PAPER_PLAN.md`** — skip Phase 1, start from **Phase 1.5** (the contract negotiation still runs; only resuming a genuine pre-1.5 legacy run may skip it, and then Phase 6.0's row 0 records "no contract")

The more detailed the input (especially figure descriptions and quantitative results), the better the output.

## Optional: Style reference (`— style-ref: <source>`, opt-in)

Lets the user steer **structural** style (section ordering, theorem density, sentence cadence, figure density, bibliography style) of the generated paper toward a reference paper they admire. **Default OFF — when the user does not pass `— style-ref`, do nothing differently from before.**

When `— style-ref: <source>` is in `$ARGUMENTS`, run the helper FIRST, before Phase 1 (paper-plan):

```bash
# Resolve $STYLE_HELPER via the canonical strict-safe chain (see
# shared-references/integration-contract.md §2). Policy A — gate:
# unresolved helper means --style-ref cannot be satisfied, so abort.
cd "$(git rev-parse --show-toplevel 2>/dev/null || pwd)" || exit 1
if [ -z "${ARIS_REPO:-}" ] && [ -f .aris/installed-skills.txt ]; then
    ARIS_REPO=$(awk -F'\t' '$1=="repo_root"{print $2; exit}' .aris/installed-skills.txt 2>/dev/null) || true
fi
if [ -z "${ARIS_REPO:-}" ] && [ -f "$HOME/.aris/repo" ]; then
    ARIS_REPO=$(cat "$HOME/.aris/repo" 2>/dev/null) || true
fi
STYLE_HELPER=".aris/tools/extract_paper_style.py"
[ -f "$STYLE_HELPER" ] || STYLE_HELPER="tools/extract_paper_style.py"
[ -f "$STYLE_HELPER" ] || { [ -n "${ARIS_REPO:-}" ] && STYLE_HELPER="$ARIS_REPO/tools/extract_paper_style.py"; }
[ -f "$STYLE_HELPER" ] || {
  echo "ERROR: extract_paper_style.py not resolved at .aris/tools/, tools/, \$ARIS_REPO/tools/, or via ~/.aris/repo." >&2
  echo "       Fix: rerun bash tools/install_aris.sh or smart_update.sh (refreshes ~/.aris/repo), export ARIS_REPO, or copy the helper to tools/." >&2
  echo "       --style-ref cannot be satisfied; aborting." >&2
  exit 1
}
STYLE_STATUS=0
CACHE=$(python3 "$STYLE_HELPER" --source "<source>") || STYLE_STATUS=$?
case "$STYLE_STATUS" in
  0) ;;                                       # share $CACHE/style_profile.md with downstream WRITER phases only
  2) echo "warning: style-ref skipped (missing optional dep)" >&2 ;;
  3) echo "error: --style-ref source failed; aborting pipeline" >&2 ; exit 1 ;;
  *) echo "error: helper failed unexpectedly; aborting pipeline" >&2 ; exit 1 ;;
esac
```

Then forward `— style-ref: <source>` only to the **writer-side** sub-skills:
- `/paper-plan` (Phase 1) — outline structure
- `/paper-write` (Phase 3) — section-by-section prose
- `/paper-illustration` (Phase 2b) — figure structural matching, optional

Sources accepted: local TeX dir / file, local PDF, arXiv id, http(s) URL. Overleaf URLs/IDs are rejected — clone via `/overleaf-sync setup <id>` first and pass the local clone path.

**Strict rules** (full contract in `tools/extract_paper_style.py` docstring):

- Use `style_profile.md` as **structural** guidance only. Match section-count tendency, theorem density, caption-length distribution, sentence cadence, math display ratio, citation style.
- **Never copy prose, claims, examples, or terminology** from anything reachable through the cache.
- **Never pass `— style-ref` (or the cache contents) to reviewer / auditor sub-skills** — Phase 4.5 (`/proof-checker`), Phase 4.7 / 5.5 (`/paper-claim-audit`), Phase 5 (`/auto-paper-improvement-loop` reviewer), Phase 5.8 (`/citation-audit`) MUST run on the artifact alone. Cross-model review independence (`../shared-references/reviewer-independence.md`).

## Pipeline

### Phase 0: Assurance Setup

Resolve the active `assurance` level and persist it so Phase 6's external
verifier reads the same value. **Run once at pipeline start, before Phase 1.**

**Resolution order** (first match wins):

1. Explicit `— assurance: draft | submission` in `$ARGUMENTS`
2. Derived from `— effort:`
   - `lite` / `balanced` → `draft` (default, **zero change from current behavior**)
   - `max` / `beast` → `submission`
3. Default: `draft`

**Action:**

```bash
mkdir -p paper/.aris
echo "<resolved-level>" > paper/.aris/assurance.txt   # draft or submission
```

**What each level does downstream:**

- **`draft`** — Existing behavior. Audits run only when their content detector
  matches (Phase 4.5 / 4.7 / 5.5 / 5.8). Missing artifacts are non-blocking.
  Silent-skip allowed.
- **`submission`** — The four mandatory audits (proof-checker,
  paper-claim-audit, citation-audit, kill-argument) are treated as load-bearing gates. Each
  sub-audit must emit its JSON artifact (PASS / WARN / FAIL / NOT_APPLICABLE /
  BLOCKED / ERROR) — never silent-skip. Phase 6 runs
  `verify_paper_audits.sh` (canonical name; resolved per
  [`shared-references/integration-contract.md`](../shared-references/integration-contract.md) §2);
  a non-zero exit blocks the Final Report.

**Escape hatch:** a user wanting the old "beast = depth-only, no audit gate"
can pass `— effort: beast, assurance: draft` explicitly. Legal but
discouraged for actual submissions. See
`shared-references/assurance-contract.md` for the full contract.

**Announce the resolved level in-line before Phase 1:**

```
📋 Assurance: <level> (derived from effort: <effort>)
   <either "current behavior, no audit gate" OR "mandatory audits gated by verify_paper_audits.sh (resolved per integration-contract §2)">
```

### Phase 1: Paper Plan

Invoke `/paper-plan` to create the structural outline:

```
/paper-plan "$ARGUMENTS"
```

If `— style-ref: <source>` was passed in `$ARGUMENTS` and the helper succeeded above, append `— style-ref: <source>` to the invocation: `/paper-plan "<topic> — style-ref: <source>"`. (Writer-side phase — forwarding is allowed; reviewer/auditor phases below must not see the style ref.)

**What this does:**
- Parse NARRATIVE_REPORT.md for claims, evidence, and figure descriptions
- Build a **Claims-Evidence Matrix** — every claim maps to evidence, every experiment supports a claim
- Design section structure (5-8 sections depending on paper type)
- Plan figure/table placement with data sources
- Scaffold citation structure
- GPT-6-Astra reviews the plan for completeness

**Output:** `PAPER_PLAN.md` with section plan, figure plan, citation scaffolding.

**Checkpoint:** Present the plan summary to the user.

```
📐 Paper plan complete:
- Title: [proposed title]
- Sections: [N] ([list])
- Figures: [N] auto-generated + [M] manual
- Target: [VENUE], [PAGE_LIMIT] pages

Shall I proceed with figure generation?
```

- **User approves** (or AUTO_PROCEED=true) → proceed to Phase 1.5.
- **User requests changes** → adjust plan and re-present.

### Phase 1.5: Negotiated Acceptance Contract (before any writing)

The plan says what the paper will contain; the CONTRACT says what "done" means
— a checklist of testable assertions, negotiated ADVERSARIALLY before the first
section is written, and graded at the end. (The plan is the boundary; the
contract is what gets graded.)

1. **Executor proposes.** From `PAPER_PLAN.md` + `NARRATIVE_REPORT.md`, draft
   `PAPER_ACCEPTANCE_CONTRACT.md`: **10–20 testable assertions** covering —
   every headline claim has a named evidence source (file/table); every
   number in the abstract traces to a results file; scope qualifiers the title
   /abstract must carry; figures that must exist and what each must show;
   section-level completeness (e.g. "limitations names ≥2 real limitations,
   not hedges"); venue constraints (page limit, anonymization). Each assertion
   must be CHECKABLE by reading the final PDF + results files — no vibes
   ("writing is clear" is not an assertion; "every acronym is defined at first
   use" is). Fewer than ~10 assertions and the gate rubber-stamps; beyond ~20
   the negotiation stalls on trivia.

2. **Reviewer pushes back** (fresh thread — the negotiation is adversarial):

   > Resolve this reviewer slot first; the following is the `backend=codex` example.

   ```
   mcp__codex__codex:
     model: gpt-6-astra
     config: {"model_reasoning_effort": "xhigh"}
     prompt: |
       You are negotiating the acceptance contract for a paper BEFORE it is
       written. Read these files directly:
       - Plan: <abs path>/PAPER_PLAN.md
       - Proposed contract: <abs path>/PAPER_ACCEPTANCE_CONTRACT.md
       - Evidence inventory: <abs path>/NARRATIVE_REPORT.md (+ results/ paths)

       Push back on the contract, not the plan: (a) assertions that are
       untestable or vibes — demand a checkable rewrite; (b) missing
       assertions — claims in the plan that no assertion covers, numbers with
       no traceability assertion, foreseeable overclaim risks with no scope
       assertion; (c) assertions the evidence inventory cannot possibly
       satisfy — flag now, not after writing. End with exactly one line:
       CONTRACT_ACCEPTED: yes    or    CONTRACT_ACCEPTED: no
       followed by your numbered revision demands if no.

       === SCOPE LIMITS (these bound what you PROPOSE, never what you look for) ===
       Report anything that is actually wrong here — including a rare-looking case, if
       this repo actually produces it. Then keep the fix in scope:
       1. This is a RESEARCH-WORKFLOW tool, not a security paper. Verification is
          welcome; over-defense is not. Assume a cooperating operator on their own
          machine — a malicious local user is NOT in the threat model.
       2. Do NOT propose SHA / hash / content-fingerprint / digest-binding schemes.
          Reporting a real defect in hashing code that already exists is fine.
       3. NO speculative machinery: do not add feature flags, migration frameworks,
          compat layers, wrappers, pins, or similar mechanisms unless evidence shows
          a current repo defect they fix or an explicit existing invariant they must
          preserve. "Load-bearing", "compatibility", and "not scaffolding" are labels,
          not evidence. Point to the failing path/artifact or invariant, and check the
          proposal's factual premises, such as whether a named package version exists.
       4. NO corner-case obsession: exotic encodings, symlink races, RTL text and
          millisecond races are out of scope unless you can show the case arises here.
       5. Where a rubric or checklist is genuinely needed, do not over-mechanize
          judgement. A clear sentence a human reads beats a scored table nobody
          maintains.
       Exception: code that runs remote commands, starts a network service, or installs
       an MCP server runs on the user's machine with their credentials — trust-boundary
       findings there are in scope and the default is strict.
       Say plainly when something is correct. Do not manufacture findings.
   ```

   A reply with a missing or malformed `CONTRACT_ACCEPTED:` line is treated as
   `no`; request a corrected verdict via `codex-reply` — that correction
   exchange does not consume a negotiation round.

3. **Iterate.** On `no`, revise the contract per the demands and resubmit via
   `mcp__codex__codex-reply` (same thread — the negotiation is one
   conversation). **Max 3 rounds.**

4. **Fallback — never stall the pipeline.** If round 3 still ends in `no`:
   record the unresolved demands verbatim in a "## Disputed" section of the
   contract and mark it `status: contested`. A contested contract is precisely
   the "insert a human when the CONTRACT is wrong" trigger, so it OVERRIDES
   `AUTO_PROCEED` for this one decision: pause and present the dispute for a
   human tie-break. If no human responds (unattended/overnight run), proceed —
   but a contested contract caps the outcome: Phase 6.0 gates on the UNDISPUTED
   assertions and the Final Report must set `Submission-ready: no` with the
   dispute reproduced verbatim; the tie-break happens when the human returns.

**Output:** `PAPER_ACCEPTANCE_CONTRACT.md` (`status: accepted | contested`,
round count, reviewer thread id). The contract is FROZEN once accepted —
Phases 2–5 implement against it; they do not edit it. If writing reveals an
assertion is genuinely wrong (not merely inconvenient), that is a contract
question: surface it at a checkpoint, don't silently rewrite.

**Boundary:** `/paper-claim-audit` (Phases 4.7, 5.5) stays zero-context — the
contract is a WRITER-side gate and is never passed to the claim auditor (its
independence is the point; two different nets).

### Phase 2: Figure Generation

If `— style-ref: <source>` was passed in `$ARGUMENTS` and the helper succeeded above, append `— style-ref: <source>` to every writer-side sub-skill invocation in this pipeline (Phases 1, 2b, 3, 5). Do **not** append it to reviewer/auditor invocations (Phases 4.5, 4.7, 5.5, 5.8).

Invoke `/paper-figure` to generate data-driven plots and tables:

```
/paper-figure "PAPER_PLAN.md"
```

**What this does:**
- Read figure plan from PAPER_PLAN.md
- Generate matplotlib/seaborn plots from JSON/CSV data
- Generate LaTeX comparison tables
- Create `figures/latex_includes.tex` for easy insertion
- GPT-6-Astra reviews figure quality and captions

**Output:** `figures/` directory with PDFs, generation scripts, and LaTeX snippets.

> **Scope:** `paper-figure` covers data plots and comparison tables. Architecture diagrams, pipeline figures, and method illustrations are handled in Phase 2b below.

#### Phase 2b: Architecture & Illustration Generation

**Skip this step entirely if `illustration: false`.**

If the paper plan includes architecture diagrams, pipeline figures, audit cascades, or method illustrations, invoke the appropriate generator based on the `illustration` parameter:

**When `illustration: figurespec`** (default) — invoke `/figure-spec`:
```
/figure-spec "[architecture/workflow description from PAPER_PLAN.md]"
```
- Deterministic JSON → SVG vector rendering (editable, reproducible)
- Best for: system architecture, workflow pipelines, audit cascades, layered topology
- Output: `figures/*.svg` + `figures/*.pdf` (via rsvg-convert) + `figures/specs/*.json`
- No external API, runs fully local

If `— style-ref: <source>` was passed and the helper succeeded above, append `— style-ref: <source>` to the invocation below as well.

**When `illustration: gemini`** — invoke `/paper-illustration`:
```
/paper-illustration "[method description from PAPER_PLAN.md or NARRATIVE_REPORT.md]"
```
- Claude plans → Gemini optimizes → Nano Banana Pro renders → Claude reviews (score ≥ 9)
- Best for: qualitative method illustrations, natural-style diagrams, result grids
- Output: `figures/ai_generated/*.png`
- Requires `GEMINI_API_KEY` environment variable

**When `illustration: mermaid`** — invoke `/mermaid-diagram`:
```
/mermaid-diagram "[method description from PAPER_PLAN.md]"
```
- Generates Mermaid syntax diagrams (flowchart, sequence, class, state, etc.)
- Best for: lightweight flowcharts, state machines, simple sequence diagrams
- Output: `figures/*.mmd` + `figures/*.png`
- Free, no API key needed

**When `illustration: codex-image2`** — invoke `/paper-illustration-image2`:
```
/paper-illustration-image2 "[method description from PAPER_PLAN.md or NARRATIVE_REPORT.md]"
```
- Claude plans → Codex native image generation renders → Claude reviews (same multi-stage workflow as `gemini`, different renderer)
- Best for: users who want a GPT-image-style renderer without needing `GEMINI_API_KEY`; uses your existing Codex / ChatGPT Plus/Pro quota
- Output: `figures/ai_generated/figure_final.png` + `latex_include.tex` + `review_log.json` (emitted via the `/paper-illustration-image2` SKILL's `finalize` step, which delegates to the canonical `paper_illustration_image2.py` helper resolved per [integration-contract §2](../shared-references/integration-contract.md#2-canonical-helper--one-implementation-not-copy-pasted))
- **Prerequisites** (beyond ARIS's standard Claude Code + Codex coexistence): the local Codex app-server must be signed in (`codex debug app-server send-message-v2 "ping"` succeeds), and the dedicated MCP bridge must be registered — see `mcp-servers/codex-image2/README.md` for the one-time `claude mcp add` command. Delegate the preflight to `/paper-illustration-image2` (which resolves the helper via the canonical chain), or invoke the helper directly via the shim at `tools/paper_illustration_image2.py preflight --workspace .` to confirm before relying on this path.
- **Experimental**: this renderer shells through the Codex debug app-server, which Codex documents as an unstable surface. Prefer `figurespec` or `gemini` for production submission flows until `codex-image2` stabilizes.

**When `illustration: false`** — skip entirely. All non-data figures must be created manually (draw.io, Figma, TikZ) and placed in `figures/` before Phase 3.

**Choosing the right mode:**
- Formal architecture / workflow / topology figures → `figurespec` (default)
- Method concept illustrations with natural style, have `GEMINI_API_KEY` → `gemini`
- Method concept illustrations, prefer ChatGPT Plus/Pro quota over Gemini key → `codex-image2`
- Quick flowchart / state machine → `mermaid`
- Full manual control → `false`

These are complementary, not mutually exclusive: you can run multiple generators for different figures in the same paper by re-invoking with different `illustration` overrides.

**Checkpoint:** List generated vs manual figures.

```
📊 Figures complete:
- Data plots (auto, Phase 2): [list]
- Architecture/illustrations (auto, Phase 2b, mode=<illustration>): [list]
- Manual (need your input): [list]
- LaTeX snippets: figures/latex_includes.tex

[If manual figures needed]: Please add them to figures/ before I proceed.
[If all auto]: Shall I proceed with LaTeX writing?
```

> **Writing invariant (every drafting and revision step):** calibrate each
> claim to its evidence and state it directly; generic caveats live in the
> Limitations section only; writing instructions are never manuscript content
> ("do not mention X" means omit X, not "we do not address X"); tone edits
> never change what the paper knows; the paper is a launch, not a progress
> report — organize around the strongest advantage, give every experiment an
> argumentative duty, keep unfavorable numbers in the tables and explain them
> as tradeoffs only where the evidence supports that, never narrating
> defeats. `/paper-write` carries the full CONFIDENT PROSE, HONEST LIMITS
> contract.

### Phase 3: LaTeX Writing

Invoke `/paper-write` to generate section-by-section LaTeX:

```
/paper-write "PAPER_PLAN.md"
```

If `— style-ref: <source>` was passed in `$ARGUMENTS` and the helper succeeded above, append `— style-ref: <source>` to the invocation: `/paper-write "PAPER_PLAN.md — style-ref: <source>"`.

**What this does:**
- Write each section following the plan, with proper LaTeX formatting
- Insert figure/table references from `figures/latex_includes.tex`
- Build `references.bib` from citation scaffolding
- Clean stale files from previous section structures
- Automated bib cleaning (remove uncited entries)
- De-AI polish (remove "delve", "pivotal", "landscape"...)
- GPT-6-Astra reviews each section for quality

**Output:** `paper/` directory with `main.tex`, `sections/*.tex`, `references.bib`, `math_commands.tex`.

**Checkpoint:** Report section completion.

```
✍️ LaTeX writing complete:
- Sections: [N] written ([list])
- Citations: [N] unique keys in references.bib
- Stale files cleaned: [list, if any]

Shall I proceed with compilation?
```

### Phase 4: Compilation

Invoke `/paper-compile` to build the PDF:

```
/paper-compile "paper/"
```

**What this does:**
- `latexmk -pdf` with automatic multi-pass compilation
- Auto-fix common errors (missing packages, undefined refs, BibTeX syntax)
- Up to 3 compilation attempts
- Post-compilation checks: undefined refs, page count, font embedding
- Precise page verification via `pdftotext`
- Stale file detection

**Output:** `paper/main.pdf`

**Checkpoint:** Report compilation results.

```
🔨 Compilation complete:
- Status: SUCCESS
- Pages: [X] (main body) + [Y] (references) + [Z] (appendix)
- Within page limit: YES/NO
- Undefined references: 0
- Undefined citations: 0

Shall I proceed with the improvement loop?
```

### Phase 4.5: Proof Verification (theory papers only)

**Skip this phase if the paper contains no theorems, lemmas, or proofs.**

```
if paper contains \begin{theorem} or \begin{lemma} or \begin{proof}:
    Run /proof-checker "paper/"
    This invokes GPT-6-Astra xhigh to:
    - Verify all proof steps (hypothesis discharge, interchange justification, etc.)
    - Check for logic gaps, quantifier errors, missing domination conditions
    - Attempt counterexamples on key lemmas
    - Generate PROOF_AUDIT.md with issue list + severity

    If FATAL or CRITICAL issues found:
        Fix before proceeding to improvement loop
    If only MAJOR/MINOR:
        Proceed, improvement loop may address remaining issues
else:
    skip — no proofs, no action
```

### Phase 4.7: Paper Claim Audit

**Skip if no result files exist (e.g., survey/position papers with no experiments).**

```
if results/*.json or results/*.csv or outputs/*.json exist:
    Run /paper-claim-audit "paper/"
    Fresh zero-context reviewer compares every number in the paper
    against raw result files. Catches rounding inflation, best-seed
    cherry-pick, config mismatch, delta errors.

    If FAIL:
        Fix mismatched numbers before improvement loop
    If WARN:
        Proceed, but flag for manual verification
else:
    skip — no experimental results to verify
```

### Phase 5: Auto Improvement Loop

Invoke `/auto-paper-improvement-loop` to polish the paper:

```
/auto-paper-improvement-loop "paper/"
```

If `— style-ref: <source>` was passed in `$ARGUMENTS` and the helper succeeded above, append `— style-ref: <source>` to the invocation: `/auto-paper-improvement-loop "paper/ — style-ref: <source>"`. The improvement loop's reviewer sub-agent will still NOT see the style ref (the loop's own SKILL forbids it); only the fix-implementation phase consumes it.

**What this does (2 rounds):**

**Round 1:** GPT-6-Astra xhigh reviews the full paper → identifies CRITICAL/MAJOR/MINOR issues → Claude Code implements fixes → recompile → save `main_round1.pdf`

**Round 2:** GPT-6-Astra xhigh re-reviews the recompiled draft cold (fresh review — no fix summaries, no conversation carry-over) → identifies remaining issues → Claude Code implements fixes → recompile → save `main_round2.pdf`

**Typical improvements (calibration cuts both ways):**
- Fix assumption-model mismatches
- Narrow genuine overclaims to the supported scope — and state supported claims
  directly, removing redundant hedges
- Consolidate scattered generic caveats into Limitations
- Add missing interpretations and notation
- Make Limitations more specific (only when a material limit is missing — never pad)
- Add theory-aligned experiments if needed

**Output:** Three PDFs for comparison + `PAPER_IMPROVEMENT_LOG.md`.

**Format check** (included in improvement loop Step 8): After final recompilation, auto-detect and fix overfull hboxes (content exceeding margins), verify page count vs venue limit, and ensure compact formatting. Location-aware thresholds: any main-body overfull blocks completion regardless of size; appendix overfulls block only if >10pt; bibliography overfulls block only if >20pt.

### Phase 5.5: Final Paper Claim Audit (MANDATORY submission gate)

After `/auto-paper-improvement-loop` finishes, **rerun** `/paper-claim-audit` before the final report whenever the paper contains numeric claims and machine-readable raw result files exist.

Use the same detectors as Phase 4.7:
- numeric-claim regex over `paper/main.tex` and `paper/sections/*.tex`
- raw-evidence file search in `results/`, `outputs/`, `experiments/`, and `figures/` for `.json`, `.jsonl`, `.csv`, `.tsv`, `.yaml`, or `.yml`

This phase is **mandatory** if both detectors are positive. It blocks the final report.
If numeric claims exist but no raw result files are found, stop and warn the user before declaring the paper complete.
If no numeric claims exist, skip.

```bash
NUMERIC_CLAIMS=$(rg -n -e '[0-9]+(\.[0-9]+)?\s*(%|\\%|±|\\pm|x|×)' \
  -e '(accuracy|BLEU|F1|AUC|mAP|top-1|top-5|error|loss|perplexity|speedup|improvement)' \
  paper/main.tex paper/sections 2>/dev/null || true)

RAW_RESULT_FILES=$(find results outputs experiments figures -type f \
  \( -name '*.json' -o -name '*.jsonl' -o -name '*.csv' -o -name '*.tsv' -o -name '*.yaml' -o -name '*.yml' \) 2>/dev/null | head -200)

if [ -n "$NUMERIC_CLAIMS" ] && [ -n "$RAW_RESULT_FILES" ]; then
    Run /paper-claim-audit "paper/"
    If FAIL:
        Fix mismatched numbers before the final report
elif [ -n "$NUMERIC_CLAIMS" ]; then
    Stop and warn: the paper contains numeric claims but no raw evidence files were found
fi
```

**Empirical motivation:** in a real submission run, the final paper claimed a narrower experiment grid than the raw JSON actually contained, and a tolerance value was rounded down past the actual relative error. Both were caught only after manual `paper-claim-audit` invocation in the final round; the improvement loop did not detect them.

### Phase 5.6: Kill-Argument Adversarial Review (theory / scope-heavy papers)

After Phase 5.5 (claim audit) passes, run `/kill-argument` whenever the paper is theory-heavy or makes explicit scope/generality claims in the title or abstract. This is a final adversarial check that complements the claim/citation/proof audits: those verify *local correctness* (numbers match, cites resolve, theorems prove); kill-argument tests *headline-level* survival — whether the paper as a whole answers the worst rejection paragraph a senior area chair would write.

```bash
THEORY_ENV_COUNT=$(rg -c '\\begin\{(theorem|lemma|proposition|corollary)\}' paper/main.tex paper/sections 2>/dev/null...