vipin wiki

PaperOrchestra

5 min read

PaperOrchestra

Role In The Wiki

PaperOrchestra is a project-local D-drive skill pack for turning research ideas, experiment logs, templates, and guidelines into LaTeX research-paper drafts through a multi-agent workflow.

Current Claims

  • EXTRACTED: The source repository is https://github.com/Ar9av/PaperOrchestra.git.
  • EXTRACTED: The inspected remote HEAD was 5eda989cc284a32e27dfd08f881e958a3b317406.
  • EXTRACTED: The repository was mirrored locally under skill/paper-orchestra/ inside vipin wiki.
  • EXTRACTED: The repository contains 9 installable SKILL.md folders plus a non-skill shared reference folder.
  • EXTRACTED: The local Codex installation copied the 9 skill folders into .codex/skills/ and copied shared into .codex/skills/shared/.
  • EXTRACTED: The skill pack implements the PaperOrchestra paper-writing pipeline: outline generation, plotting, literature review, section writing, and content refinement.
  • EXTRACTED: The project also includes support skills for agent-log aggregation, PaperWritingBench-style benchmark construction, and PaperOrchestra-style autorating.
  • EXTRACTED: The repository’s design uses instruction files and deterministic helpers rather than embedded LLM or API clients.
  • INFERRED: This is especially relevant for Vipin’s paper digestion, literature mapping, and research-project write-up workflows because it provides a structured path from messy experimental material to reviewable paper artifacts.

Local Routing

  • Source mirror: D:/Research/vipin's knowledgebase/skill/paper-orchestra/.
  • Installed skill root: D:/Research/vipin's knowledgebase/.codex/skills/.
  • Main installed skill: .codex/skills/paper-orchestra/SKILL.md.
  • Supporting installed skills:
    • .codex/skills/agent-research-aggregator/
    • .codex/skills/outline-agent/
    • .codex/skills/plotting-agent/
    • .codex/skills/literature-review-agent/
    • .codex/skills/section-writing-agent/
    • .codex/skills/content-refinement-agent/
    • .codex/skills/paper-writing-bench/
    • .codex/skills/paper-autoraters/
    • .codex/skills/shared/

When To Use

  • Use paper-orchestra when the goal is to produce an end-to-end LaTeX manuscript from research materials.
  • Use agent-research-aggregator first when the user has scattered coding-agent logs or project directories rather than clean idea.md and experimental_log.md inputs.
  • Use literature-review-agent alone when the immediate task is bibliography construction, citation verification, or Intro/Related Work drafting.
  • Use paper-autoraters when comparing paper drafts or scoring literature-review quality with the PaperOrchestra evaluation framing.
  • Use paper-writing-bench when turning an existing paper into benchmark raw materials for testing a paper-writing pipeline.

Official Usage Pattern

Normal usage starts from the orchestrator, not from manually invoking all 9 skills.

Prepare A Workspace

Create or scaffold a workspace with the required input tuple:

workspace/
  inputs/
    idea.md
    experimental_log.md
    template.tex
    conference_guidelines.md
    figures/                 optional existing figures
  • idea.md: the paper idea, method summary, and target contribution.
  • experimental_log.md: experiment setup, raw numeric results, observations, ablations, and constraints.
  • template.tex: the target conference or journal LaTeX template.
  • conference_guidelines.md: page limit, formatting rules, mandatory sections, deadline notes, and submission constraints.
  • figures/: optional pre-existing figures; if absent, plotting-agent attempts to create the needed visuals.

The source repo provides a scaffold helper:

python .codex/skills/paper-orchestra/scripts/init_workspace.py --out workspace/
python .codex/skills/paper-orchestra/scripts/validate_inputs.py --workspace workspace/

Main Invocation

Once the four required files are present, ask Codex or another host agent:

Run the paper-orchestra pipeline on ./workspace.

or, with an absolute path:

Use paper-orchestra to turn D:/path/to/workspace into a complete LaTeX paper.

The orchestrator then loads and coordinates the sub-skills:

  1. outline-agent writes workspace/outline.json.
  2. plotting-agent writes workspace/figures/*.png and workspace/figures/captions.json.
  3. literature-review-agent writes workspace/refs.bib and workspace/drafts/intro_relwork.tex.
  4. section-writing-agent writes a complete workspace/drafts/paper.tex.
  5. content-refinement-agent iterates and promotes the best version to workspace/final/paper.tex, then the workflow compiles workspace/final/paper.pdf when LaTeX tooling is available.

Steps 2 and 3 are independent; official guidance says to run them in parallel when the host supports parallel agents. If not, run literature review first because citation discovery and verification are slower.

If Inputs Are Messy

If the project has scattered notes, coding-agent logs, or experiment folders instead of clean idea.md and experimental_log.md, use the optional aggregator first:

Use agent-research-aggregator on D:/path/to/project, create PaperOrchestra inputs, then run paper-orchestra on the resulting workspace.

agent-research-aggregator should produce:

  • workspace/inputs/idea.md
  • workspace/inputs/experimental_log.md
  • workspace/ara/aggregation_report.md

The user still needs to supply template.tex and conference_guidelines.md.

Targeted Invocation

The support skills can also be used individually:

  • literature-review-agent: use when only citations, BibTeX, Introduction, or Related Work are needed.
  • plotting-agent: use when only paper figures and captions are needed from an existing outline or experiment log.
  • section-writing-agent: use when outline, figures, references, and Intro/Related Work already exist and the task is to fill the rest of the paper.
  • content-refinement-agent: use when a draft already exists and the task is review-style revision.
  • paper-autoraters: use to score or compare paper drafts.
  • paper-writing-bench: use to reverse-engineer benchmark raw materials from an existing paper.

Output Convention

Official output paths are workspace-relative:

  • workspace/outline.json
  • workspace/figures/
  • workspace/refs.bib
  • workspace/drafts/intro_relwork.tex
  • workspace/drafts/paper.tex
  • workspace/refinement/worklog.json
  • workspace/final/paper.tex
  • workspace/final/paper.pdf
  • workspace/provenance.json

For this wiki’s D-drive setup, prefer placing the workspace under the relevant research project folder or another explicit D-drive path, then invoke the installed project-local skills from .codex/skills/.

Contribution Summary

  • EXTRACTED: PaperOrchestra contributes a host-agent-pluggable implementation of a five-agent research-paper writing process.
  • EXTRACTED: It separates LLM-dependent reasoning from deterministic checks, which makes the skills portable across coding-agent hosts.
  • EXTRACTED: It encodes paper-fidelity mappings, schemas, prompts, gates, and refinement rules in a reusable local skill structure.
  • INFERRED: Its practical value is strongest when paired with a disciplined evidence/provenance workflow: agent logs become structured inputs, citations are verified, figures and tables are grounded in experiment logs, and revisions are constrained by explicit review gates.

Counterpoints And Gaps

  • UNVERIFIED: No full local manuscript-generation run has been completed yet in this wiki.
  • AMBIGUOUS: The best long-term local workspace convention for PaperOrchestra outputs across multiple projects has not been standardized yet.
  • AMBIGUOUS: Large literature-review runs may need explicit rate-limit handling or API keys depending on Semantic Scholar availability.
  • INFERRED: For serious use, the wiki should pair this skill with project-specific evidence pages so generated paper claims do not outrun verified results.

Graph View

Backlinks