Write Paper

Preamble (run first)

# --- GENERATED PREAMBLE START ---
# Resolve project root
_PROJECT_ROOT=$(git rev-parse --show-toplevel 2>/dev/null || pwd)
if ! git rev-parse --show-toplevel >/dev/null 2>&1; then
  echo "WARNING: Not inside a git repository. Files will be written to $(pwd)."
fi
mkdir -p ~/.rstack/sessions ~/.rstack/analytics "$_PROJECT_ROOT/.rstack"
touch ~/.rstack/sessions/"$PPID"
find ~/.rstack/sessions -mmin +120 -type f -delete 2>/dev/null || true
_SESSIONS=$(find ~/.rstack/sessions -mmin -120 -type f 2>/dev/null | wc -l | tr -d ' ')
_RSTACK_DIR="$(cd "$(dirname "$0")/.." 2>/dev/null && pwd || echo "$HOME/.claude/skills/rstack")"
_RSTACK_CONFIG="$_RSTACK_DIR/bin/rstack-config"

# Update check
_UPD=$("$_RSTACK_DIR/bin/rstack-update-check" 2>/dev/null || true)
[ -n "$_UPD" ] && echo "$_UPD" || true

# Config values
_VENUE=$("$_RSTACK_CONFIG" get venue 2>/dev/null || echo "arxiv")
_COMPUTE=$("$_RSTACK_CONFIG" get compute_preferred 2>/dev/null || echo "modal")
_PROACTIVE=$("$_RSTACK_CONFIG" get proactive 2>/dev/null || echo "true")
_TEL=$("$_RSTACK_CONFIG" get telemetry 2>/dev/null || echo "off")
_TEL_PROMPTED=$([ -f ~/.rstack/.telemetry-prompted ] && echo "yes" || echo "no")

# Detect available state files (work products at root, plumbing in .rstack/)
_HAS_IDEA="no"
_HAS_REFINED="no"
_HAS_LIT_REVIEW="no"
_HAS_LIT_REVIEW_MD="no"
_HAS_NOVELTY="no"
_HAS_EXPERIMENTS="no"
_HAS_ANALYSIS="no"
_HAS_FIGURES="no"

# Work products at project root
[ -f "$_PROJECT_ROOT/idea.md" ] && _HAS_IDEA="yes"
[ -f "$_PROJECT_ROOT/refined-idea.md" ] && _HAS_REFINED="yes"
[ -f "$_PROJECT_ROOT/lit-review.md" ] && _HAS_LIT_REVIEW_MD="yes"
[ -f "$_PROJECT_ROOT/novelty-assessment.md" ] && _HAS_NOVELTY="yes"
[ -d "$_PROJECT_ROOT/analysis" ] && _HAS_ANALYSIS="yes"

# Plumbing in .rstack/
[ -f "$_PROJECT_ROOT/.rstack/lit-review.jsonl" ] && _HAS_LIT_REVIEW="yes"
[ -f "$_PROJECT_ROOT/.rstack/experiments.jsonl" ] && _HAS_EXPERIMENTS="yes"

_FIG_COUNT=0
if [ -d "$_PROJECT_ROOT/analysis/figures" ]; then
  _HAS_FIGURES="yes"
  _FIG_COUNT=$(find "$_PROJECT_ROOT/analysis/figures" -type f \( -name "*.png" -o -name "*.pdf" -o -name "*.svg" \) 2>/dev/null | wc -l | tr -d ' ')
fi

echo "PROJECT_ROOT: $_PROJECT_ROOT"
echo "STATE: idea=$_HAS_IDEA refined=$_HAS_REFINED lit-review=$_HAS_LIT_REVIEW lit-review-md=$_HAS_LIT_REVIEW_MD"
echo "STATE: novelty=$_HAS_NOVELTY experiments=$_HAS_EXPERIMENTS analysis=$_HAS_ANALYSIS"
echo "STATE: figures=$_HAS_FIGURES (count=$_FIG_COUNT)"
echo "VENUE: $_VENUE"
echo "COMPUTE: $_COMPUTE"
echo "PROACTIVE: $_PROACTIVE"

# Count experiment runs if available
if [ "$_HAS_EXPERIMENTS" = "yes" ]; then
  _EXP_COUNT=$(wc -l /dev/null | tr -d ' ')
  echo "EXPERIMENTS: $_EXP_COUNT runs logged"
fi

# Detect LaTeX compiler
_LATEX_COMPILER="none"
if command -v tectonic >/dev/null 2>&1; then
  _LATEX_COMPILER="tectonic"
elif command -v pdflatex >/dev/null 2>&1; then
  _LATEX_COMPILER="pdflatex"
fi
echo "LATEX_COMPILER: $_LATEX_COMPILER"

# Detect rstack skill root for templates
_SKILL_ROOT=""
if [ -d "$HOME/.claude/skills/rstack" ]; then
  _SKILL_ROOT="$HOME/.claude/skills/rstack"
elif [ -d ".claude/skills/rstack" ]; then
  _SKILL_ROOT="$(pwd)/.claude/skills/rstack"
fi
echo "SKILL_ROOT: ${_SKILL_ROOT:-NOT_FOUND}"

_BRANCH=$(git branch --show-current 2>/dev/null || echo "unknown")
echo "BRANCH: $_BRANCH"
if [ ! -f ~/.rstack/.setup-complete ]; then
  echo "NEEDS_SETUP"
fi
_TEL_START=$(date +%s)
_SESSION_ID="$$-$(date +%s)"
echo "TELEMETRY: ${_TEL:-off}"
echo "TEL_PROMPTED: $_TEL_PROMPTED"
if [ "$_TEL" != "off" ]; then
  echo '{"skill":"write-paper","ts":"'$(date -u +%Y-%m-%dT%H:%M:%SZ)'","repo":"'$(basename "$(git rev-parse --show-toplevel 2>/dev/null)" 2>/dev/null || echo "unknown")'"}'  >> ~/.rstack/analytics/skill-usage.jsonl 2>/dev/null || true
fi
if [ "$_TEL" != "off" ]; then
  echo '{"skill":"write-paper","ts":"'$(date -u +%Y-%m-%dT%H:%M:%SZ)'","session_id":"'"$_SESSION_ID"'","rstack_version":"'"$(cat "$_RSTACK_DIR/VERSION" 2>/dev/null | tr -d "[:space:]" || echo "unknown")"'"}'  > ~/.rstack/analytics/.pending-"$_SESSION_ID" 2>/dev/null || true
fi
for _PF in $(find ~/.rstack/analytics -maxdepth 1 -name '.pending-*' 2>/dev/null); do
  if [ -f "$_PF" ]; then
    _PF_BASE="$(basename "$_PF")"
    _PF_SID="${_PF_BASE#.pending-}"
    [ "$_PF_SID" = "$_SESSION_ID" ] && continue
    if [ "$_TEL" != "off" ] && [ -x "$_RSTACK_DIR/bin/rstack-telemetry-log" ]; then
      "$_RSTACK_DIR/bin/rstack-telemetry-log" --event-type skill_run --skill _pending_finalize --outcome unknown --session-id "$_SESSION_ID" 2>/dev/null || true
    fi
    rm -f "$_PF" 2>/dev/null || true
  fi
  break
done
eval "$("$_RSTACK_DIR/bin/rstack-slug" 2>/dev/null)" 2>/dev/null || true
echo "SLUG: ${SLUG:-unknown}"
# --- GENERATED PREAMBLE END ---

After the preamble runs, note the state and proceed.

If output shows UPGRADE_AVAILABLE : read rstack-upgrade/SKILL.md and follow the "Inline Upgrade Flow". Then continue with this skill. If output shows JUST_UPGRADED : tell user "Running RStack v{to} (just updated!)" and continue.

If output shows NEEDS_SETUP: tell user "RStack not configured yet. Run /setup first." Then read setup-skill/SKILL.md and follow it inline before continuing.

If TEL_PROMPTED is no: Ask the user about telemetry. Use AskUserQuestion:

> Help RStack get better! Community mode shares usage data (which skills you use, > how long they take, crash info) with a stable device ID so we can track trends > and fix bugs faster. No code, file paths, or repo names are ever sent. > Change anytime with rstack-config set telemetry off.

Options:

• A) Help RStack get better! (recommended)
• B) No thanks

If A: run rstack-config set telemetry community

If B: ask a follow-up: > How about anonymous mode? Just a counter that helps us know if anyone's out there.

Options:

• A) Sure, anonymous is fine
• B) No thanks, fully off

If B→A: run rstack-config set telemetry anonymous If B→B: run rstack-config set telemetry off

Always run:

touch ~/.rstack/.telemetry-prompted

This only happens once. If TEL_PROMPTED is yes, skip this entirely.

Important: Note the PROJECT_ROOT value from the preamble output. All file paths below are relative to this project root directory. Work products (idea.md, paper.tex, analysis/) are at the project root. Plumbing (.rstack/lit-review.jsonl, .rstack/experiments.jsonl) is in the .rstack/ subdirectory.

If LATEX_COMPILER is none: warn the user that no LaTeX compiler is installed. Paper will be written but not compiled. Suggest: "Run /setup to install tectonic, or install manually: cargo install tectonic / brew install tectonic."

Voice

Direct, concrete, no filler. Sound like a research collaborator, not a writing service. Name the file, the section, the citation. When something is missing, say what's missing and what to do about it.

Writing Style — Sound Human

The paper must read like a human researcher wrote it. AI-generated academic prose has tells that reviewers notice. Follow these rules:

Banned words (never use these in the paper text): delve, utilize, leverage, facilitate, comprehensive, multifaceted, moreover, furthermore, additionally, notably, importantly, it is worth noting, pivotal, paradigm, synergy, holistic, nuanced, in the realm of, shed light on, pave the way, it is important to note, a myriad of, in conclusion it can be said

Use instead: simple, direct alternatives. "use" not "utilize". "show" not "shed light on". "many" not "a myriad of". "helps" not "facilitates".

No em dashes. Use commas, periods, or parentheses instead of "—".

Sentence variety: Mix short sentences (5-8 words) with longer ones (15-25 words). Don't start three sentences in a row with the same structure. Avoid the pattern "While X, Y" or "Although X, Y" more than once per section.

Hedging: Real researchers hedge naturally. Use "suggests" not "proves". Use "we observe" not "it is clearly evident". But don't over-hedge either... one hedge per claim is enough.

Active voice preferred. "We trained the model" not "The model was trained". "Our method achieves" not "It can be observed that the method achieves".

No throat-clearing. Don't start paragraphs with "In recent years" or "With the advent of" or "It has been widely recognized that". Start with what matters.

Keep it natural. A good test: read each paragraph out loud. If it sounds like something a person would say at a conference, it's fine. If it sounds like a textbook written by committee, rewrite it.

Rigor Rules

These are non-negotiable. Violating any of these is worse than not writing the paper.

1. NEVER hallucinate results. Every number in the paper must come from

.rstack/experiments.jsonl. If a metric is not in the logs, do not write it.

1. NEVER invent citations. Every \cite{key} must reference a real entry in

paper.bib generated from .rstack/lit-review.jsonl. No phantom references.

1. NEVER fabricate figures. Every \includegraphics{path} must point to a real

file in analysis/figures/. If the file does not exist, skip it and write % [Figure pending: description] as a comment.

1. Say what's real. If results are negative or inconclusive, write them honestly.

Reviewers catch fabricated positive results. Always.

---

Step 0: Gather Context

Read all available state files. The preamble already detected what exists. Read each available file in order:

1. idea.md -- the original research idea (project root)
2. refined-idea.md -- sharpened hypothesis (project root)
3. .rstack/lit-review.jsonl -- structured paper records (plumbing)
4. lit-review.md -- human-readable literature review (project root)
5. novelty-assessment.md -- novelty analysis (project root)
6. .rstack/experiments.jsonl -- experiment log with metrics (plumbing)
7. analysis/ directory -- any analysis outputs, tables, summaries (project root)
8. analysis/figures/ -- publication-ready figures (project root, list all files)

For each file that exists, read it. For each file that does not exist, note it and move on. The paper will be written with whatever is available.

Minimum requirements to proceed:

• At least ONE of: idea.md, refined-idea.md (need to know what the paper is about)
• experiments.jsonl (need real results -- this is a research paper, not a proposal)

If experiments.jsonl is missing, use AskUserQuestion:

> No experiment results found in .rstack/experiments.jsonl (plumbing). > > Options: > A) Provide results manually (paste metrics or point to a file) > B) Run /experiment first to generate results > C) Write a paper draft with placeholder results (will need filling in later)

If the user picks C, proceed but mark every results reference with % PLACEHOLDER: replace with real results comments.

If BOTH idea files are missing, use AskUserQuestion:

> No research idea found. What is this paper about?

Take the user's response and use it as the idea context for the rest of the workflow.

---

Step 1: Venue Selection

Check venue configuration. Use rstack-config get venue or default to arxiv.

_RSTACK_CONFIG="$HOME/.claude/skills/rstack/bin/rstack-config"
"$_RSTACK_CONFIG" get venue 2>/dev/null || echo "venue: arxiv"

Read the formatting rules from _shared/venues.md if the file exists in the skill root. If _shared/venues.md does not exist, use built-in knowledge of arXiv formatting.

Use AskUserQuestion:

> Target venue: {detected venue} > > Options: > A) Proceed with {detected venue} > B) Change venue (specify)

v1 limitation: Only arXiv format is currently supported. If the user picks a different venue (NeurIPS, ICML, ACL, etc.), tell them:

> arXiv is the only supported template in v0.1. NeurIPS/ICML templates are coming > in v1.1. For now, writing in arXiv format -- you can reformat later with the > venue-specific style files.

Always use the arXiv template regardless of selection in v0.1.

---

Step 2: Generate BibTeX

If .rstack/lit-review.jsonl exists, read it and generate BibTeX entries.

For each paper record in the JSONL file, generate a BibTeX entry:

• Cite key format: {firstauthorlastname}{year} (lowercase, no spaces)
• Example: author "Vaswani et al." year 2017 becomes vaswani2017
• If duplicate keys, append a letter: vaswani2017a, vaswani2017b
• Entry type: @article for journal papers, @inproceedings for conferences,

@misc for arXiv preprints

• Fields: author, title, year, journal/booktitle, url (if available)

Write the complete BibTeX file to paper.bib at the project root.

_PROJECT_ROOT=$(git rev-parse --show-toplevel 2>/dev/null || pwd)
wc -l "$_PROJECT_ROOT/paper.bib" 2>/dev/null || echo "paper.bib not yet created"

If .rstack/lit-review.jsonl does not exist, use AskUserQuestion:

> No literature review found (.rstack/lit-review.jsonl missing). > > Options: > A) Provide references manually (paste BibTeX or paper titles) > B) Run /lit-review first to find relevant papers > C) Proceed without citations (Related Work section will be thin)

If the user provides references manually, parse them and write paper.bib at the project root. If they pick C, create an empty paper.bib and note that Related Work will need references added later.

---

Step 3: Paper Planning

Generate a paper plan inspired by the Ignis paper-planner-agent pattern. Based on all the context gathered in Step 0, produce:

3a. Title Candidates

Generate exactly 3 title candidates:

1. A descriptive title (clear, specific, says what the paper does)
2. A punchy title (shorter, memorable, could be a good workshop title)
3. A question title (frames the contribution as answering a research question)

3b. Section Outline

Generate a bullet-point outline for each section:

• Abstract (3-4 bullet points: problem, method, results, conclusion)
• Introduction (motivation, gap, contribution, outline of paper)
• Related Work (group by theme from lit-review.md, note how this work differs)
• Method (approach, architecture/algorithm, key design decisions)
• Experiments (setup, baselines, datasets, metrics, results summary)
• Results and Discussion (main findings, ablations, analysis, limitations)
• Conclusion (summary, implications, future work)

3c. Figure Plan

List which figures go in which section, with descriptions:

• Match against actual files in analysis/figures/
• If figures exist: map each to its appropriate section
• If no figures: note which figures SHOULD exist and where they would go

3d. Human Review

Present the plan to the user via AskUserQuestion:

> Paper Plan > > Title candidates: > 1. {descriptive title} > 2. {punchy title} > 3. {question title} > > Outline: {summary of sections and key points} > > Figures: {N} figures mapped to sections > > Options: > A) Use title 1 > B) Use title 2 > C) Use title 3 > D) Suggest a different title > E) Modify the outline (describe changes)

Apply the user's feedback before proceeding to Step 4.

---

Step 4: Write LaTeX

This is the main writing step. Load the arXiv template and write the full paper.

4a. Load Template

Read the template from the skill root:

TEMPLATE_PATH=""
if [ -n "$_SKILL_ROOT" ] && [ -f "$_SKILL_ROOT/templates/arxiv/template.tex" ]; then
  TEMPLATE_PATH="$_SKILL_ROOT/templates/arxiv/template.tex"
elif [ -f "templates/arxiv/template.tex" ]; then
  TEMPLATE_PATH="templates/arxiv/template.tex"
fi
echo "TEMPLATE: ${TEMPLATE_PATH:-NOT_FOUND}"

If the template is found, read it and use it as the structural skeleton. If not found, use the standard arXiv article format from built-in knowledge: \documentclass[11pt]{article} with standard packages.

4b. Write Document

Write the complete LaTeX document from \documentclass through \end{document}.

Mandatory rules during writing:

• Use REAL citation keys from paper.bib with

…

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: sunnnybala
• Source: sunnnybala/Rstack
• License: MIT

Install and usage instructions live in the source repository linked above.