Ar9av/PaperOrchestra / skills/paper-orchestra
paper-orchestra
VerifiedOrchestrate the full PaperOrchestra (Song et al., 2026, arXiv:2604.05018) five-agent pipeline to turn unstructured research materials (idea, experimental log, LaTeX template, conference guidelines, optional figures) into a submission-ready LaTeX manuscript and compiled PDF. TRIGGER when the user asks to "write a paper from my experiments", "turn this idea and these results into a paper", "generate a conference submission", "run paper-orchestra on X", or otherwise wants the end-to-end paper-writing pipeline. Coordinates the outline-agent, plotting-agent, literature-review-agent, section-writing-agent, and content-refinement-agent skills.
| File | Kind | Size |
|---|---|---|
| SKILL.md | skill md | 13.8 KB |
| references/anti-leakage-prompt.md | markdown | 2.5 KB |
| references/host-integration.md | markdown | 7.8 KB |
| references/io-contract.md | markdown | 7.1 KB |
| references/outline-reconciliation.md | markdown | 4.7 KB |
| references/paper-summary.md | markdown | 4.1 KB |
| references/pipeline.md | markdown | 7.2 KB |
| scripts/anti_leakage_check.py | script | 3.3 KB |
| scripts/build_pdf.py | script | 28.3 KB |
| scripts/check_idea_density.py | script | 5.6 KB |
| scripts/check_tex_packages.py | script | 5.9 KB |
| scripts/claim_evidence_gate.py | script | 16.1 KB |
| scripts/diff_outlines.py | script | 3.5 KB |
| scripts/init_workspace.py | script | 4.5 KB |
| scripts/test_claim_evidence_gate.py | script | 5.1 KB |
| scripts/validate_consistency.py | script | 9.7 KB |
| scripts/validate_inputs.py | script | 4.8 KB |
references/io-contract.md
View on GitHub# I/O Contract
Reference for the orchestrator. Defines the workspace layout and the schema of
every input and intermediate artifact.
## Workspace layout
```
workspace/
├── inputs/ # User-provided
│ ├── idea.md # I — Sparse or Dense idea (markdown)
│ ├── experimental_log.md # E — setup, raw numeric data, observations
│ ├── template.tex # T — conference LaTeX template
│ ├── conference_guidelines.md # G — formatting rules, page limit, sections
│ └── figures/ # F — optional pre-existing figures (PNG/PDF)
├── outline.json # Step 1 output
├── figures/ # Step 2 output (generated)
│ ├── <figure_id>.png
│ ├── ...
│ └── captions.json # {figure_id: caption_text, ...}
├── refs.bib # Step 3 output
├── citation_pool.json # Step 3 output (verified S2 metadata)
├── drafts/
│ ├── intro_relwork.tex # Step 3 output
│ └── paper.tex # Step 4 output (then mutated by Step 5)
├── refinement/ # Step 5 working dir
│ ├── worklog.json
│ ├── iter1/{paper.tex,paper.pdf,review.json,score.json}
│ ├── iter2/...
│ └── iter3/...
├── final/ # Accepted snapshot + compiled PDF
│ ├── paper.tex
│ └── paper.pdf
└── provenance.json # Input/output hashes for reproducibility
```
## Input file schemas
### `idea.md` — Idea Summary (I)
Markdown. Two valid variants (the paper distinguishes them in App. C.3).
**Sparse variant** (high-level concept note, no math):
```markdown
## Problem Statement
(Precise definition of the technical problem.)
## Core Hypothesis
(The proposed solution / intuition.)
## Proposed Methodology (High-Level Technical Approach)
(Conceptual description; describe modules by function, not their math.)
## Expected Contribution
(Intended theoretical or practical value.)
```
**Dense variant** (preserves math, equations, variable definitions):
```markdown
## Problem Statement
## Core Hypothesis
## Proposed Methodology (Detailed Technical Approach)
- includes LaTeX equations, variable definitions, architectural choices
## Expected Contribution
```
The Outline Agent automatically handles both. Dense produces more rigorous
methodology sections; Sparse exercises the system's robustness (per the
paper's ablation, App. E).
### `experimental_log.md` — Experimental Log (E)
Markdown. Strict structure required:
```markdown
# Experimental Log
## 1. Experimental Setup
* **Datasets:** ...
* **Evaluation Metrics:** ...
* **Baselines Compared:** ...
* **Implementation Details:** ...
## 2. Raw Numeric Data
(Tables in markdown format. Section-Writing Agent extracts these into LaTeX
booktabs tables. Use plain markdown table syntax — | col | col | — with no
references to "Table N" or "Figure N".)
## 3. Qualitative Observations
* (factual statements like "training loss converged after 200 epochs",
"method X failed on test case Y", etc.)
```
**Critical rules** (from App. F.2 Experimental Log Generation prompt):
- No references to figure or table numbers ("See Table 1", "as shown in Fig. 5")
- Past-tense persona ("We ran...", "The results were...")
- Self-contained: no citations, no URLs, no author names
- Numeric values must be 100% accurate — they become the ground truth for
the Refinement Agent's hallucination check
### `template.tex` — LaTeX Template (T)
A conference LaTeX template (CVPR, ICLR, NeurIPS, ICML, etc.) with empty
`\section{...}` placeholders. The Section Writing Agent fills the empty
sections in-place; the preamble (`\documentclass`, `\usepackage`, etc.) is
preserved verbatim.
### `conference_guidelines.md` — Conference Guidelines (G)
Markdown describing:
- Page limit (in pages, integer)
- Mandatory sections (e.g., "must have an Abstract, Introduction, Methods,
Experiments, Conclusion")
- Formatting requirements (single-column vs two-column, font, margins)
- Submission deadline date (used to derive `cutoff_date` for the literature
review and section writing agents)
### `inputs/figures/` — Pre-existing Figures (F)
Optional. PNG or PDF files. If present, the Plotting Agent will reuse them
where the outline plan permits and only generate the missing ones. If empty,
the Plotting Agent generates everything from scratch (the paper calls this
mode `PlotOff`; the GT-figure mode is `PlotOn`).
## Intermediate artifact schemas
### `outline.json`
See `skills/outline-agent/references/outline-schema.md` and the JSON Schema at
`skills/outline-agent/references/outline_schema.json`.
### `figures/captions.json`
```json
{
"fig_framework_overview": "Plain-text caption here, no markdown, no 'Figure N:' prefix.",
"fig_main_results": "..."
}
```
The Section Writing Agent splices these into `\caption{...}` commands.
### `citation_pool.json`
Internal record produced by the Literature Review Agent of every verified
citation, with full Semantic Scholar metadata. Schema:
```json
{
"papers": [
{
"paperId": "abc123...", // S2 unique ID
"bibtex_key": "vaswani2017attention",
"title": "Attention Is All You Need",
"authors": [{"name": "A. Vaswani"}, ...],
"year": 2017,
"venue": "NeurIPS",
"abstract": "...",
"externalIds": {"DOI": "...", "ArXiv": "1706.03762"},
"verified_via": "semantic_scholar",
"match_score": 100, // Levenshtein ratio
"discovered_for": ["intro", "related_work"]
}
],
"cutoff_date": "2024-11-01",
"min_cite_paper_count": 27 // 90% of len(papers), rounded down
}
```
### `refinement/worklog.json`
```json
{
"iterations": [
{
"iter": 1,
"timestamp": "2026-04-09T12:34:56Z",
"review": { "strengths": [...], "weaknesses": [...], "questions": [...] },
"score": { "overall": 67, "axes": {...} },
"actions_taken": ["Rewrote Section 3.2 for clarity", ...],
"decision": "accept"
},
...
],
"halted_because": "iteration_cap_reached" | "overall_decreased" | "tie_with_negative_subaxis_delta" | "no_new_weaknesses",
"best_iter": 2
}
```
### `provenance.json`
```json
{
"created_at": "2026-04-09T12:34:56Z",
"inputs": {
"idea.md": {"sha256": "...", "bytes": 1234},
"experimental_log.md": {"sha256": "...", "bytes": 5678},
"template.tex": {"sha256": "...", "bytes": 9012},
"conference_guidelines.md": {"sha256": "...", "bytes": 345}
},
"outline.json": {"sha256": "..."},
"refs.bib": {"sha256": "...", "n_entries": 59},
"figures": {
"fig_framework_overview.png": {"sha256": "..."},
...
},
"final": {
"paper.tex": {"sha256": "..."},
"paper.pdf": {"sha256": "..."}
},
"skill_versions": {
"paper-orchestra": "0.1.0"
}
}
```
This is an out-of-paper improvement for reproducibility. Optional but
recommended.
Related skills
| Skill | Stars | Updated | Works with | Status | Install |
|---|---|---|---|---|---|
| writing-skillsobra/superpowersUse when creating new skills, editing existing skills, or verifying skills work before deployment | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |
| xlsxanthropics/skillsUse this skill any time a spreadsheet file is the primary input or output. This means any task where the user wants to: open, read, edit, or fix an existing .xlsx, .xlsm, .xltx, .csv, or .tsv file (e.g., adding columns, computing formulas, formatting, charting, cleaning messy data); create a new spreadsheet from scratch or from other data sources; or convert between tabular file formats. Trigger especially when the user references a spreadsheet file by name or path — even casually (like "the xlsx in my downloads") — and wants something done to it or produced from it. Also trigger for cleaning or restructuring messy tabular data files (malformed rows, misplaced headers, junk data) into proper spreadsheets. The deliverable must be a spreadsheet file. Do NOT trigger when the primary deliverable is a Word document, HTML report, standalone Python script, database pipeline, or Google Sheets API integration, even if tabular data is involved. | Stars: 198k | Updated | Claude CodeCodex | Verified | Install |
| shipping-and-launchaddyosmani/agent-skillsPrepares production launches. Use when preparing to deploy to production, or when asking what needs to be in place before shipping. Use when you need a pre-launch checklist, when setting up monitoring, when planning a staged rollout, or when you need a rollback strategy. | Stars: 119k | Updated | Claude CodeCodex | Verified | Install |
| childhoodthedotmack/claude-memBabysit a shipped release after merge, publication, or deployment. Verify published packages, production behavior, installed plugins, and late CI or user reports until the observation window is clean; use babysit for PR readiness before release. | Stars: 118k | Updated | Claude CodeCodex | Verified | Install |