> ## Documentation Index
> Fetch the complete documentation index at: https://mintlify.com/getcompanion-ai/feynman/llms.txt
> Use this file to discover all available pages before exploring further.

# Paper Draft

> Turn research findings into a polished paper-style draft with equations, sections, figures, and explicit claims.

The paper draft workflow turns research findings into a structured academic document. It uses the `writer` subagent to produce well-organized prose from collected notes, with inline citations, LaTeX equations where appropriate, and figures with captions.

## Invocation

<Tabs>
  <Tab title="CLI">
    ```bash theme={null}
    feynman draft "<topic>"
    ```
  </Tab>

  <Tab title="REPL">
    ```
    /draft <topic>
    ```
  </Tab>
</Tabs>

**Examples**

```bash theme={null}
feynman draft "a survey of retrieval-augmented generation techniques"
feynman draft "efficient fine-tuning methods for large language models"
```

```
/draft sparse attention mechanisms: a comparative analysis
/draft "on the scaling properties of mixture-of-experts models"
```

## Workflow stages

<Steps>
  <Step title="Outline">
    Before writing, the lead agent outlines the draft structure: proposed title, sections, key claims to make, source material to draw from, and a verification log for critical claims, figures, and calculations. The outline is written to `outputs/.plans/<slug>.md` and presented to you for confirmation before proceeding.
  </Step>

  <Step title="Write">
    The `writer` subagent produces the draft from the collected notes and outline. The draft includes at minimum:

    * Title and abstract
    * Problem statement
    * Related work
    * Method or synthesis
    * Evidence or experiments
    * Limitations
    * Conclusion
    * Sources appendix with direct URLs for all primary references

    Clean Markdown is used throughout, with LaTeX where equations materially help the exposition. Quantitative data, benchmarks, and comparisons are rendered as charts using `pi-charts`; architectures and pipelines use Mermaid diagrams. Every figure includes a caption.
  </Step>

  <Step title="Cite and verify">
    The `verifier` subagent adds inline citations to the draft and verifies every source URL.
  </Step>

  <Step title="Claim sweep">
    Before delivery, the draft is swept for any claim that sounds stronger than its supporting evidence. Tentative results are labeled as tentative; unsupported numerics are removed rather than left for the verifier to discover.
  </Step>

  <Step title="Deliver">
    Exactly one draft is saved to `papers/<slug>.md`.
  </Step>
</Steps>

## Outputs

| Artifact          | Path                       |
| ----------------- | -------------------------- |
| Draft outline     | `outputs/.plans/<slug>.md` |
| Final paper draft | `papers/<slug>.md`         |

<Note>
  Paper-style drafts are saved to `papers/` rather than `outputs/`. This follows the workspace output convention defined in `AGENTS.md`.
</Note>

## Draft structure

The produced draft follows standard academic structure:

* **Abstract** — concise summary of scope and key findings
* **Introduction** — motivation, context, and contribution statement
* **Related work** — survey of prior work with inline citations
* **Method or synthesis** — the core technical content
* **Evidence or experiments** — results, benchmarks, or analysis
* **Limitations** — honest assessment of scope and gaps
* **Conclusion** — summary and future directions
* **Sources** — complete bibliography with direct URLs

## Subagents used

| Subagent   | Role                                                                         |
| ---------- | ---------------------------------------------------------------------------- |
| `writer`   | Produces the structured draft from collected notes and the confirmed outline |
| `verifier` | Adds inline citations and verifies source URLs                               |

## Working from existing research

`/draft` is a natural follow-up to [Deep Research](/workflows/deep-research) or [Literature Review](/workflows/literature-review). When you have already gathered source material in a session, the writer draws from those findings directly rather than re-gathering sources.

```
/deepresearch "efficient attention mechanisms"
# ... review the brief ...
/draft "efficient attention mechanisms: a survey"
```

<Tip>
  Run `/draft` after `/deepresearch` or `/lit` to skip the research phase. The writer uses the session's already-vetted findings to produce a more focused draft.
</Tip>

## Related

* [Deep Research](/workflows/deep-research) — gather source material before drafting
* [Literature Review](/workflows/literature-review) — structured academic survey to draw from
* [Peer Review](/workflows/review) — review the draft once written