Agent skill

Figure Composer

by aipoch in aipoch/open-science

Compose one publication-grade multi-panel figure. An agent skill from aipoch/open-science.

Apache-2.0Auto-check passedResearch & Science

Install Figure Composer

skills CLI
$ npx skills add aipoch/open-science --skill figure-composer -a claude-code

Project install by default; add -g for ~/.claude/skills/.

GitHub CLI
$ gh skill install aipoch/open-science figure-composer --agent claude-code

Project scope by default; add --scope user for a personal install. Needs GitHub CLI 2.90.0 or later (public preview).

Manual copy
$ git clone --depth 1 https://github.com/aipoch/open-science.git skills-src && mkdir -p .claude/skills && cp -r skills-src/resources/skills/figure-composer .claude/skills/figure-composer && rm -rf skills-src

Use ~/.claude/skills/ instead of .claude/skills for a personal install. The folder must contain SKILL.md.

Claude Code skills documentation · loads skills from .claude/skills/

Facts

Skill name
figure-composer
GitHub stars
5.5k
Token cost
~2.9k tokens
SKILL.md length
1,314 words
Files
3
Skills in repo
8
Repo updated
First seen
Licence
Apache-2.0

At a glance

Compose one publication-grade multi-panel figure. An agent skill from aipoch/open-science.

  • Works in 4 steps: Narrative → panel outline → Fan out panel workers → Compose and bind the producer Run → …
  • Tasks that involve Reproducible research
  • SKILL.md covers Open-Science Notebook call, Inputs, Entry points and 1. Narrative → panel outline, plus 5 more sections
  • Runs Python scripts from its folder

What it does

Figure Composer is an agent skill from aipoch/open-science. Compose one publication-grade multi-panel figure. Start from a one-line claim plus immutable data Artifact Version references, or inspect an existing figure and draft its outline directly. Plan a 12-column panel outline, delegate one worker per panel, compose and inspect the result, then run at most three adversarial review rounds while regenerating only affected panels. For a standalone plot use figure-style; for whole-paper figure ordering use paper-narrative.

Its SKILL.md is about 2.9k tokens, which your agent loads only when the skill is triggered. The skill folder holds 2 other files (for example `kernel.py` and `open-science.json`).

It sits in Research & Science, covering Reproducible research. The repository describes itself as: The open-source AI research workbench for scientific research and agent workflows. Local-first, model-agnostic desktop app with extensible skills, MCP tools and connectors… The licence is Apache-2.0.

When your agent uses it

  • Tasks that involve Reproducible research

Example prompts

  • “/figure-composer”

Requirements

  • Python 3

Workflow steps

4 steps, taken from the step headings in SKILL.md.

  1. Narrative → panel outline
  2. Fan out panel workers
  3. Compose and bind the producer Run
  4. Adversarial review loop

What it can do on your machine

Read from SKILL.md and the folder at commit 51d7079. It shows what the files ask for, not the result of running them.

  • Tool permissions

    Pre-approves nothing: there is no allowed-tools line, so your agent's usual permission prompts apply.

    From allowed-tools in the SKILL.md frontmatter.

  • Runs code

    Ships script files (Python), which the agent can run.

    From the folder's file list and the shell code blocks in SKILL.md.

  • Network

    No URLs in SKILL.md.

    From URLs in SKILL.md, links to its own repository left out.

  • Credentials

    Names no API keys, tokens, secrets or passwords.

    From names ending in _API_KEY, _TOKEN, _SECRET, _KEY or _PASSWORD in SKILL.md.

Context cost

Figure Composer loads about 2.9k tokens when it runs. Until then it costs about 122 tokens; SKILL.md has 1,314 words of instructions outside code blocks.

Always · name and description, kept in context so the agent knows when to use it
~122
When it runs · the whole SKILL.md, loaded when a task matches
~2.9k

Estimates: characters ÷ 4, the usual rule of thumb; real counts depend on the model's tokenizer. Scripts and assets cost tokens only if the agent reads them.

Safety

Auto-check passed

The automated check found no risky patterns in SKILL.md.

Automated static check — not a guarantee. Review scripts before installing. It scans the text of SKILL.md for risky patterns (piping downloads into a shell, reading credential files, hidden Unicode, destructive commands); files beside SKILL.md are not scanned.

SKILL.md

The full file from aipoch/open-science at commit 51d7079, republished under its Apache-2.0 licence (© aipoch). 1,314 words, ~2,904 tokens.

Download SKILL.mdSave it as .claude/skills/figure-composer/SKILL.md (or your agent's skills folder). This skill also uses 2 other files; get the full folder from GitHub.
name
figure-composer
description
Compose one publication-grade multi-panel figure. Start from a one-line claim plus immutable data Artifact Version references, or inspect an existing figure and draft its outline directly. Plan a 12-column panel outline, delegate one worker per panel, compose and inspect the result, then run at most three adversarial review rounds while regenerating only affected panels. For a standalone plot use `figure-style`; for whole-paper figure ordering use `paper-narrative`.
license
Apache-2.0

Figure Composer — narrative → panels → compose → adversarial loop

figure-composer is the outer workflow for one multi-panel figure. Use the figure-style rules while planning and reviewing; every panel worker uses those rules independently. Run paper-narrative first when the paper-level figure sequence is still undecided.

Open-Science Notebook call

Every notebook_execute request whose code uses a function named in this skill includes this skill ID:

json
{ "kernelSkillIds": ["figure-composer"], "code": "print(figure_outline_schema())" }

kernelSkillIds contains the skill ID; function calls belong in code. Call the named functions directly without an import or discovery step.

Inputs

  • claim: the one sentence the figure makes true without surrounding prose.
  • dataVersionIds: immutable Upload or Artifact Version identities grounding the panels.
  • width_mm: venue column width, commonly 85–89 mm single or 174–183 mm double.
  • rulesVersionId: immutable Artifact Version containing the design rules used by the composite reviewer.
  • delegatePrefix: short branch-unique prefix for panel and reviewer child names.

Run this workflow only in the Main/root agent. Delegated children cannot call host.delegate, so the whole composer cannot itself be delegated.

Entry points

  • From a claim: Main writes the outline in step 1 from the claim, data, and figure-style rules.
  • From an existing figure: inspect it with host.viewImage, then have Main draft and review the outline directly. Current host.llm calls do not accept images, so do not add a second hidden inference step. Pixels cannot supply Artifact Version identities; fill data_vid from the provided data.

1. Narrative → panel outline

Main produces a panel_outline matching figure_outline_schema():

json
{
  "claim": "…",
  "width_mm": 180,
  "ncol": 12,
  "row_heights_mm": [40, 60, 46, 52],
  "panels": [
    {
      "letter": "a",
      "role": "schematic",
      "row": 0,
      "col": 0,
      "colspan": 12,
      "chart_family": "schematic overview",
      "message": "…",
      "data_vid": null,
      "ask": "…"
    },
    {
      "letter": "b",
      "role": "primary",
      "row": 1,
      "col": 0,
      "colspan": 7,
      "chart_family": "scatter + trend",
      "message": "…",
      "data_vid": "…",
      "ask": "…"
    }
  ]
}

Outline rules:

  • A is the context-free hook: schematic or hero, normally full width.
  • B carries the claim: it should make the sentence true on its own.
  • Remaining panels add evidence in descending importance.
  • Use one row per sub-claim, normally 5–10 panels, and a 12-column grid.
  • Every non-schematic data_vid must be one of the supplied immutable Version identities. Do not invent or rewrite Version IDs.
  • Set fixed_panel_set: true only when the user explicitly requires the exact listed panels.

Geometry helpers reject duplicate panel letters (case-insensitive), overlapping grid spans, panels outside the grid, and invalid or subpixel grid dimensions. Use unique panel identifiers and non-overlapping positive spans within the grid.

Review the outline before fan-out. Use the schema as a contract; Main does the reasoning and does not call host.llm to generate the outline again.

2. Fan out panel workers

Generate each task in Python with panel_task(outline, letter, fig_label). The returned task contains the complete panel procedure. Pass it unchanged on the first render and supply the panel's data Version in inputs.

Dispatch from repl_execute. host.delegate accepts at most four children per atomic call, so send ordered waves of no more than four. Each request uses this output schema:

javascript
const panelOutputSchema = {
  type: 'object',
  additionalProperties: false,
  required: ['panelVersionId', 'labelsUsed'],
  properties: {
    panelVersionId: { type: 'string', minLength: 1 },
    labelsUsed: { type: 'array', items: { type: 'string' } }
  }
}

Use wait: false, then collect the exact { frameId, attemptId } receipt handles. A collect timeout ends observation, not the child Attempt: collect the same handles again while any remain running. Retry only after a terminal failure or an explicitly rejected output, using a fresh child name. Panel workers must submit their structured result with host.submitOutput before finishing. Reject a non-completed/error child, missing or unsatisfied structured output, a missing or duplicate expected panel_<letter>.png, or a mismatch between its Artifact versionId and structuredOutput.panelVersionId. MIME metadata may be absent; the exact filename and Version identity are the binding checks. Return each wave's validated { letter, versionId } values from the repl_execute call instead of relying on local const or let declarations to survive a later call.

Keep finalized Version identities in outline order. Temporary paths are never the Agent-to-Agent contract. Child names remain occupied after settlement, so use a unique delegatePrefix and round number.

3. Compose and bind the producer Run

Generate a producer task with composition_task(outline, panelVersions, fig_label). Main's newly written Artifact can remain pending until its turn ends; the producer child publishes a finalized composite that the reviewer can use. Pass the ordered panel Version identities in inputs and require this output schema:

javascript
{
  type: 'object',
  additionalProperties: false,
  required: ['compositeVersionId'],
  properties: { compositeVersionId: { type: 'string', minLength: 1 } }
}

The producer resolves the collected Version identities and places the paths in a small JSON handoff under process.env.OPEN_SCIENCE_HANDOFF_DIR. On its notebook_execute request, it passes the ordered, de-duplicated panel identities as artifactVersionInputs. This registers the delegated immutable panel Versions as the composition Run's provenance inputs; paths remain byte-access implementation details and must never replace Version identities in this field. The producer calls compose_figure, verifies notebook completion, and keeps the actual returned runId. It publishes the final PNG with write_artifact_file({ filename: "figure.png", producerRunId: composeResult.runId }); never substitute a round number or locally invented Run identity. This binds the composite Artifact to the run that last wrote its bytes. Fail the workflow if any panel Version cannot be validated in the active Project; never silently compose with an unregistered provenance input.

Collect the exact producer Attempt and require completed status, satisfied structured output, and exactly one figure.png Artifact whose versionId matches structuredOutput.compositeVersionId. Use that finalized composite Version for inspection and review. The producer submits the structured result with host.submitOutput and finishes normally.

compose_figure requires each input image to match its panel_px dimensions exactly. A mismatch raises before the output is saved; regenerate the panel at the requested size. Images are never stretched to fit. Use the exact figsize expressions generated by panel_task, rather than rounded inch measurements, and verify the saved PNG dimensions.

Show full SKILL.md (465 more words)Show less

3.5 Look before review

Call compose_crops in Python and inspect every crop before formal review. host.viewImage never upscales and caps the output long edge at 1568 pixels; omit maxSize when native pixels are required.

One repl_execute invocation can attach at most four images. Split five or more crops into ordered batches of no more than four, and let each invocation finish successfully before starting the next; a failed enclosing invocation discards every image staged by that invocation. For each cropBatch, use the current camelCase API:

javascript
if (cropBatch.length > 4) throw new Error('viewImage crop batch exceeds four images')
for (const [letter, box] of cropBatch) {
  await host.viewImage(
    { versionId: compositeVersionId },
    { crop: { unit: 'pixels', left: box[0], top: box[1], right: box[2], bottom: box[3] } }
  )
}
return { inspectedPanels: cropBatch.map(([letter]) => letter) }

Check contrast, smallest marks, leader crossings, color identity, legend binding, seams, panel-letter overlap, gutter bleed, and resize artifacts. Fix an obvious defect before formal review.

4. Adversarial review loop

Run at most three rounds. An independent reviewer Attempt is required before returning any composite. Generate the reviewer task with composite_review_task(...) and its outputSchema with review_schema(). Pass the task unchanged to one reviewer; include the composite, optional previous composite, rulesVersionId, and every non-null panel data Version in inputs. Collect the exact receipt and use only validated structuredOutput as the review object. The reviewer submits it with host.submitOutput; do not replace formal review with Main's own inspection.

After each result:

  1. Accept when the verdict is accept or minor_revision, there are no BLOCKERs, and there are at most two MAJORs.
  2. Save previous_outline = copy.deepcopy(outline) before applying outline_revisions explicitly. Then call apply_outline_revisions(outline, revisions, previous_outline=previous_outline). This includes new panels and every panel whose pixel dimensions changed, even when a shared row-height change names only one panel. Pass the same dpi and gutter_mm as composition if overriding their defaults. Removed panels are excluded; drop their entries from the collected panel Versions.
  3. Call group_fixes_by_panel(review) and compute regen = (affected | set(fixb)) & {p["letter"] for p in outline["panels"]}.
  4. Regenerate only regen. Build each retry task as panel_task(outline, letter, fig_label) + fixb.get(letter, "") and add: “Do not over-correct: preserve everything the previous version got right.” Include the prior panel Version when one exists and its data Version in inputs.
  5. Keep every clean panel's exact Version identity. Compose a new revision with a fresh producer child only after every regenerated panel passes the same identity checks. Review only that new composite Version.

Stop when accepted, or when outline_revisions is empty and new findings are only carve-out exceptions to the previous round; that is the over-labeling signal. Otherwise stop after round three. If the current composite was not accepted, report the unresolved findings rather than return an older composite as the final result.

After acceptance, verify the composite's provenance contains the current panel Versions. Return that finalized figure.png Artifact with a user-visible link; do not publish a duplicate root Artifact.

Anti-patterns

  • Do not regenerate clean panels.
  • Do not manufacture findings.
  • Verify review anchors on the composite, not only on isolated panels.
  • Remove labels that a reader with field context would find redundant.

© aipoch, Apache-2.0. Rendered from Markdown: HTML in the file is shown as text, images as links, and headings moved down two levels. Raw file

Files

SKILL.md and 2 other files in resources/skills/figure-composer of aipoch/open-science.

  • SKILL.md
  • kernel.py
  • open-science.json

Open the folder on GitHubat commit 51d7079

Compare with similar skills

Figure Composer next to the 5 skills that share the most tags, products or categories with it. Stars are the repository's; “used in” counts other GitHub owners with a copy.

Figure Composer compared with similar skills
SkillStarsUsed inTokensAuto-checkLicenceRepo updated
Figure Composer this skillaipoch/open-science5.5k—~2.9kAutomated safety check: PassApache-2.0
Peer ReviewK-Dense-AI/claude-scientific-writer2.4k2 repos~3.1kAutomated safety check: NotesMIT
CHARLS Paper Reproduction Guidexjtulyc/MedgeClaw6171 repos~1.8kAutomated safety check: PassNone
Add Bactopia Toolbactopia/bactopia522—~4.1kAutomated safety check: PassMIT
Modeling Code and Result Contractsyushui2022/MathModel-Skill452—~1.4kAutomated safety check: PassMIT
Light Research OrchestratorLight0305/Light-skills641—~3.8kAutomated safety check: PassMIT

Similar skills

  • Peer Review

    K-Dense-AI/claude-scientific-writer

    Prepare evidence-bounded, constructive peer-review drafts and structured manuscript assessments.

    2.4k GitHub starsUsed in 2 repos~3.1k tokens
    Research & ScienceAuto-check: notes
  • Guides an agent through reproducing papers built on the CHARLS health and retirement survey, from variable mapping to cognition, depression and isolation scores.

    617 GitHub starsUsed in 1 repo~1.8k tokens
    Research & ScienceAuto-check passed
  • Add Bactopia Tool

    bactopia/bactopia

    Scaffold a complete Bactopia Tool across all three tiers -- module, subworkflow, and workflow entry point under workflows/bactopia-tools/.

    522 GitHub stars~4.1k tokensUpdated 2 mo ago
    Research & ScienceAuto-check passed
  • Modeling Code and Result Contracts

    yushui2022/MathModel-Skill

    Generates result-evidence contracts, tables and runnable q1 to q3 modeling code scaffolds for a math modeling paper from a model route, a data plan and cleaned data.

    452 GitHub stars~1.4k tokensUpdated yesterday
    Research & ScienceAuto-check passed
  • Light Research Orchestrator

    Light0305/Light-skills

    Coordinates and recovers multi-stage Light research projects from a single passport file, with checkpoints, stale-work tracking and rerouting only when you approve.

    641 GitHub stars~3.8k tokensUpdated 3 mo ago
    Research & ScienceAuto-check passed
  • HypoGeniC Hypothesis Generation

    K-Dense-AI/scientific-agent-skills

    Plans and audits runs of the HypoGeniC and HypoRefine packages, which propose hypotheses from labeled text datasets, with local checks before any model call.

    48k GitHub starsUsed in 1 repo~3.6k tokens
    Research & ScienceAuto-check: notes

More from aipoch/open-science

All 8 skills in this repo
  • Open-Science Skill Creator

    aipoch/open-science

    Creates, revises, evaluates and publishes skills in the Open-Science app through its native host.skills composer, with optional test prompts and benchmarks.

    5.5k GitHub stars~1.7k tokensUpdated today
    Auto-check passed
  • Compute Environment Setup

    aipoch/open-science

    Prepares setup instructions and a named activation file for a user-managed software environment on an Open-Science SSH or Slurm compute host.

    5.5k GitHub starsUsed in 1 repo~2.6k tokens
    Auto-check passed
  • Handles /Customize requests by sending Skill work to the internal skill-creator and managing Specialist agents through the JavaScript host.agents SDK.

    5.5k GitHub stars~2.7k tokensUpdated today
    Auto-check passed
  • Figure Style

    aipoch/open-science

    Publication-grade correctness and legibility rules for final-deliverable scientific figures, not exploratory plots.

    5.5k GitHub stars~5.1k tokensUpdated today
    Auto-check passed
  • Paper Narrative

    aipoch/open-science

    Judge and reshape the story told by an entire paper figure deck.

    5.5k GitHub stars~4.4k tokensUpdated today
    Auto-check passed
  • Open-Science Host Inspection

    aipoch/open-science

    Teaches an agent to inspect Open-Science's JavaScript control REPL, check which host.* calls are allowed, and find project files, sessions and agent frames.

    5.5k GitHub stars~3.8k tokensUpdated today
    Auto-check passed

Questions about Figure Composer

What does Figure Composer do?

Compose one publication-grade multi-panel figure. An agent skill from aipoch/open-science. Figure Composer is an agent skill from aipoch/open-science. Compose one publication-grade multi-panel figure.

When should I use Figure Composer?

Figure Composer fits situations like: tasks that involve Reproducible research.

How do I install Figure Composer in Claude Code?

Run `npx skills add aipoch/open-science --skill figure-composer -a claude-code`. Or copy the skill folder (resources/skills/figure-composer in aipoch/open-science) into .claude/skills/figure-composer in your project. Claude Code loads it when a task matches its description.

How do I install Figure Composer in Codex?

Run `npx skills add aipoch/open-science --skill figure-composer -a codex`. Or copy the skill folder (resources/skills/figure-composer in aipoch/open-science) into .agents/skills/figure-composer in your project. Codex loads it when a task matches its description.

Can I use Figure Composer in Cursor, Gemini CLI or GitHub Copilot?

Cursor, Gemini CLI, GitHub Copilot and OpenCode also load SKILL.md folders. With the skills CLI, run `npx skills add aipoch/open-science --skill figure-composer -a cursor` (or -a gemini-cli, github-copilot or opencode for the others). To copy it by hand, put the folder in .cursor/skills/figure-composer, .gemini/skills/figure-composer, .github/skills/figure-composer and .opencode/skills/figure-composer in your project.

What does Figure Composer need to run?

Going by SKILL.md and its folder, Figure Composer needs Python for the scripts in its folder. Our summary lists: Python 3.

Does Figure Composer access the network?

SKILL.md contains no URLs. Any network use would come from the scripts or tools the agent runs. This is read from the text; nothing was executed.

Is Figure Composer safe to install?

Our automated static check of SKILL.md found no risky patterns, such as piping downloads into a shell, reading credential files or hidden Unicode. It is not a guarantee. Review the folder before installing.

What licence does Figure Composer use?

Figure Composer is published under the Apache-2.0 licence (declared in SKILL.md). It allows redistribution, so the full SKILL.md is shown on this page.

How many tokens does Figure Composer use?

About 2.9k tokens (SKILL.md is roughly 12k characters). Agents keep only the skill's name and description in context until a task matches; then they load SKILL.md in full.

What are the alternatives to Figure Composer?

Skills that share tags, products or a category with Figure Composer: Peer Review (K-Dense-AI/claude-scientific-writer, 2.4k stars), CHARLS Paper Reproduction Guide (xjtulyc/MedgeClaw, 617 stars), Add Bactopia Tool (bactopia/bactopia, 522 stars) and Modeling Code and Result Contracts (yushui2022/MathModel-Skill, 452 stars). The comparison table on this page puts their stars, adoption, token cost, safety result and licence side by side.

Who maintains Figure Composer?

aipoch (a GitHub organization) maintains it in aipoch/open-science, which has 5,475 GitHub stars. The repository holds 8 skills in this directory. The repository was last updated on October 8, 2026.

Source: aipoch/open-science on GitHub. Facts on this page come from the repository at the commit we read; the author's words are quoted as theirs.