scientific-figure

KO | EN

gitlite — search

#css #cli #html #pytorch #automation #electron #gpt #agent #terminal #gemini

scientific-figure

(★ 9)

No description available.

파일 탐색기

ZIP 다운로드

- - marketplace.json
- - ci.yml
  - release.yml
- settings.json
- banner.svg
- example_compound.png
- example_heatmap.png
- example_line_plot.png
- example_multipanel.png
- gui-credentials.png
- gui-model-routes.png
- gui-providers.png
- workflow.svg
- user-guide.md
- user-guide.zh-CN.md
- - .mcp.json
- 0.6.0.json
- README.md
- 0.6.0.md
- 0.6.1.md
- 0.6.2.md
- 0.6.3.md
- 0.6.4.md
- - openai.yaml
- - __init__.py
  - __main__.py
  - _resources.py
  - activation.py
  - codex_plugin.py
  - components.py
  - config.py
  - config_editor.py
  - connection_test.py
  - convergence.py
  - execution.py
  - figure_graph.py
  - figure_layout.py
  - generation_conditions.py
  - generation_intent.py
  - install_paths.py
  - install_transaction.py
  - lifecycle_contracts.py
  - lifecycle_prompts.py
  - local_status.py
  - orchestrator.py
  - phase_operation.py
  - phase_workers.py
  - provenance.py
  - provider_configuration.py
  - publication_profiles.py
  - qml_controller.py
  - qml_gui.py
  - release_bundle.py
  - release_source.py
  - report.py
  - run_invalidator.py
  - run_store.py
  - runtime_context.py
  - server.py
  - state.py
  - style_spec.py
- - auth_cleanup.py
  - configure_codex.py
  - install_delivery.py
  - provider_environment.py
  - uninstall_delivery.py
- - asset-manifest.schema.json
  - execution-result.schema.json
  - export-result.schema.json
  - figure-brief.schema.json
  - figure-graph.schema.json
  - figure-plan.schema.json
  - generation-conditions.schema.json
  - layout-analysis.schema.json
  - layout-manifest.schema.json
  - panel-request.schema.json
  - phase-operation.schema.json
  - planning-advice.schema.json
  - plot-spec.schema.json
  - repair-plan.schema.json
  - root-cause-report.schema.json
  - run-state.schema.json
  - solved-layout.schema.json
  - structure-questions.schema.json
  - style-bible.schema.json
  - style-spec.schema.json
  - validation-report.schema.json
- - default-project.yaml
  - default-style-bible.json
  - publication.mplstyle
- - __init__.py
  - support.py
- .gitignore
- install.sh
- LICENSE
- pyproject.toml
- SKILL.md
- uv.lock
- capture_readme_screenshots.py
- generate_readme_examples.py
- record_release_acceptance.py
- release.py
- sync_plugin_bundle.py
- verify_release_candidate.py
.gitattributes
.gitignore
CITATION.cff
CONTRIBUTING.md
install.sh
LICENSE
README.md
README.zh-CN.md
SECURITY.md
uninstall.sh

# CDN으로 사용하기

jsDelivr

jsDelivr는 공개 GitHub 리포지토리를 별도 설정 없이 CDN으로 즉시 서빙합니다. 버전과 파일을 고르면 웹페이지에 바로 붙일 수 있는 링크와 예시 코드가 만들어집니다.

버전

파일

링크

예시

// repository documentation

Was this content helpful?

★ 0 (0 ratings)

<a href="./README.md">English</a> &nbsp;|&nbsp; <a href="./README.zh-CN.md">简体中文</a>

From a clarified scientific request to a validated, publication-ready export.

<img src="assets/example_compound.png" alt="Publication-ready compound scientific figure" width="820"> 
 Illustrative composite output from synthetic demonstration data.

Scientific Figure Builder helps an agent turn a clarified request into source-backed
plots and assets, an assembled figure, validation evidence, and exportable outputs.
Measured data stays on deterministic Python/SVG paths; configured model providers
handle eligible image generation and visual analysis.

## Requirements

- Python 3.11 or newer.
- [`uv`](https://docs.astral.sh/uv/).
- Codex for the native agent workflow. The Core runtime can also be installed alone.

## What it delivers

- Reproducible line, scatter, bar, heatmap, error-bar, and multipanel plots.
- Structure-first mechanism figures with addressable nodes and connectors.
- Explicit planning and approval before paid image generation.
- Deterministic assembly, layered validation, targeted repair, and resumable runs.
- PNG, SVG, and PDF export, with optional PowerPoint-friendly SVG/PPTX output.
- A native Codex plugin, local runtime, CLI, and optional configuration app.

<table>
 <tr>
 <td align="center"><img src="assets/example_line_plot.png" width="250"> Error-aware line plot</td>
 <td align="center"><img src="assets/example_heatmap.png" width="250"> Efficiency landscape</td>
 <td align="center"><img src="assets/example_multipanel.png" width="330"> Multipanel composition</td>
 </tr>
 </table>

Illustrative outputs generated from synthetic demonstration data; they are not scientific evidence.

## Quick start

The working directory matters for the installation and source commands below.
Every command is shown in the directory where it should be run.

### 1. Install a published release

Run these commands from the repository root after cloning:

```bash
git clone https://github.com/assle/scientific-figure.git
cd scientific-figure          # repository root; contains ./install.sh
./install.sh --codex --release latest --with-gui
```

`./install.sh` is a repository script, so run it from the folder that contains it.
Omit `--with-gui` on a headless machine. To install only the Core runtime and CLI:

```bash
./install.sh --runtime-only --release latest
```

### 2. Use the installed CLI from any directory

After installation, the `scientific-figure` launcher is independent of the
repository. These commands work from any project directory:

```bash
scientific-figure status
scientific-figure gui
scientific-figure update --latest
```

If the launcher is not on your shell path, invoke `~/.local/bin/scientific-figure`
or add `~/.local/bin` to `PATH`. You do not need to `cd` into the repository to
run these installed commands.

### 3. Run the GUI from a source checkout

For development, run these commands inside `scientific-figure-builder/`, the folder
that contains `pyproject.toml`:

```bash
cd scientific-figure-builder
uv sync --extra gui
uv run --extra gui python -m figure_tools gui
```

A source checkout does not install the `scientific-figure` launcher by itself.
Use the command above for the current source tree, or complete step 1 and use
`scientific-figure gui` after installation.

## Configuration app

The native Qt Quick app manages Providers, Model routes, and system credentials.
It does not open a browser, start a local web server, or contact a Provider while
opening or saving configuration.

<table>
 <tr>
 <td width="50%"><img src="assets/gui-providers.png" alt="Provider endpoint and capability configuration"></td>
 <td width="50%"><img src="assets/gui-credentials.png" alt="Keyring credential and connection testing"></td>
 </tr>
 <tr>
 <td align="center">Endpoints, protocols, and declared capabilities</td>
 <td align="center">System credentials and explicit connection testing</td>
 </tr>
</table>

For first-time setup, use Providers → Model routes → Credentials and connection → Save configuration.
The app edits global configuration; changes remain in the window draft until saved.

### Model routes: choose a model for each task

- **Provider / Model ID**: select an existing endpoint and enter a model ID, Endpoint ID, or gateway alias accepted by that endpoint, rather than a display name.
- **Phase reasoning**: run separate structured reasoning for intake, planning, and review and repair; it does not draw measured-data curves.
- **Reference analysis**: extract layout, structure, and semantics from reference images.
- **Image generation**: produce non-quantitative assets or Generation units; measured-data plots still use Python/SVG.
- **Image editing / Inherit generation**: revise assets using reference images. Inheritance reuses the image-generation route rather than adding another model.
- **Visual validation**: inspect generated assets and the final assembled figure.

### Providers: endpoints, protocols, and image capabilities

- **Add / Delete**: manage Provider IDs referenced by Model routes. This local identifier is neither an API key nor a model ID.
- **Type**: choose `openai`, `anthropic`, or `dashscope` by the service's actual protocol, not its vendor name. DashScope Native only serves image-generation and editing roles.
- **Base URL**: the API base address, matching the selected protocol.
- **Authentication**: send the key in an `x-api-key` header or as a Bearer Token, as required by the endpoint.
- **Messages Path / Anthropic Version**: the request path and version header for Anthropic Compatible endpoints. Keep the displayed defaults unless the service documents another value.
- **Reference-image editing**: declare that the endpoint can edit an existing image.
- **Generation reference images**: allow reference images as input when generating a new image; this does not create references automatically.
- **Multiple references**: allow several reference images in one request.
- **Mask editing**: allow a mask to constrain the edited region.
- **Structure control**: allow structural control information in the request.
- **Native alpha**: declare that the model can directly return an image with transparency.
- **Fixed seed**: allow a random seed parameter; identical output across service versions or environments is not guaranteed.
- **Candidate batches**: declare support for batch candidate output. Enabling the switch does not generate images immediately.

These switches declare existing service capabilities; they do not add unsupported features.
Available switches depend on Provider type. DashScope does not display mask editing,
structure control, or native-alpha switches.

### Waiting and retries: control how a model call waits

Provider settings supply defaults for its Model routes; each role can override them.
Leave a field empty to inherit. Durations use seconds; attempt counts use positive integers.

- **Connection timeout**: the time allowed to establish a network connection.
- **Status interval**: how often request status is checked; a status check does not resubmit the model request.
- **Inactivity timeout**: the longest interval without a new message; receiving a message resets this timer.
- **Total timeout**: the overall time budget for the call; incoming messages do not reset it.
- **Maximum transient-error attempts**: the maximum number of attempts, including the initial request, for retryable transient errors.
- **Backoff base / cap**: the starting delay and upper limit for waits before retrying, avoiding immediate repeated submissions.

### Phase output allowances: budget structured reasoning output

- **Intake / Planning / Review and repair initial allowance**: set the initial token allowance independently for each phase. Empty fields inherit configuration; the built-in default is 8192 per phase.
- **Maximum output tokens**: cap the allowance requested for one output. An empty field uses inherited settings or a recognized model limit.
- Allowances cover model reasoning and final output, not a fixed actual consumption. Resuming the same phase retains the allowance already issued. Token allowances and timeouts are separate limits.

### Credentials and connection: store keys and explicitly test endpoints

- **Select Provider**: manage credentials independently for each endpoint.
- **API Key**: written to the operating-system Keyring when configuration is saved, never to YAML. Leaving the field empty preserves an existing credential.
- **Credential ID**: the identifier used to reference a system credential, not the secret itself.
- **Environment fallback**: enter a variable name such as `PROVIDER_API_KEY`. Provider `key_env` fields support credentials supplied to headless environments.
- **Test connection**: actively contact the service using the current unsaved draft, preferring a vision role. If only image generation is available, the app first asks about the potentially billable test.
- **Cancel / Remove credential**: cancel the connection test or remove the selected Provider's system credential. Existing keys are not displayed back in the input field.

### Saving and configuration scope

- **Save configuration**: commit the window draft and credential changes. Ordinary saves do not make network requests.
- **Discard changes**: discard unsaved edits.
- **About**: describe the app's purpose; the sidebar shows the current global configuration path.
- **Project overrides**: run `scientific-figure init /path/to/project` and edit the project configuration; this global configuration window does not switch projects.

Configuration precedence is: packaged defaults → global configuration → project
configuration → per-run overrides. See the [user guide](docs/user-guide.md) for
provider types, model roles, file locations, and a minimal YAML example.

## Workflow

Ask Codex to use the installed skill and describe the source data, intended figure,
language, dimensions or publication target, and required export formats:

```text
Use scientific-figure-builder to create a publication-ready multipanel figure
from data.csv. Export PNG, SVG, and PDF, and keep the SVG editable in PowerPoint.
```

Measured data and quantitative plots stay on deterministic Python/SVG paths.
Non-quantitative figure units can use a configured Provider route. Final
composition, validation, repair, and export remain deterministic and local.
The runtime pauses only when it needs clarification, plan approval, a repair
choice, or acknowledgement of a generation summary.

## Update, status, and uninstall

Use installed commands from any directory:

```bash
scientific-figure update --latest
scientific-figure status
```

An update preserves provider configuration, credential references, project data,
and run artifacts. If `status` reports `restart_required`, save active work and
fully restart Codex so new MCP/GUI processes load the activated runtime. Use
`scientific-figure status --remote` only when you explicitly want to check GitHub
for a newer release.

Preview the source uninstaller from the repository root, then choose the intended
scope:

```bash
./uninstall.sh --dry-run
codex plugin remove scientific-figure-builder@scientific-figure
./uninstall.sh              # Core runtime and CLI; keep configuration
./uninstall.sh --all        # Also remove global config and referenced credentials
```

The native plugin and the Core runtime have separate lifecycles. Remove both for a
complete uninstall. The uninstaller preserves global configuration and credentials
unless `--all` or `--config` is explicit.

## Documentation

- [User guide](docs/user-guide.md): installation, configuration, workflow, updates,
  and removal.
- [Contributing](CONTRIBUTING.md) and [security policy](SECURITY.md).

## Development

Run development commands from `scientific-figure-builder/`:

```bash
cd scientific-figure-builder
uv sync --extra gui
uv run --extra gui pytest -q
uvx pyright --pythonpath .venv/bin/python figure_tools install
```

The authoritative Skill lives in `scientific-figure-builder/SKILL.md`. After changing
it, run the synchronization script from the repository root:

```bash
python3 scripts/sync_plugin_bundle.py
```

After a configuration-app visual change, regenerate its README screenshots from the
repository root:

```bash
uv run --frozen --directory scientific-figure-builder --extra gui \
  python ../scripts/capture_readme_screenshots.py
```

The example figures can be regenerated independently with the same synthetic demo
data:

```bash
uv run --frozen --directory scientific-figure-builder \
  python ../scripts/generate_readme_examples.py
```

## License

[MIT](./LICENSE)

scientific-figure

파일 탐색기

# CDN으로 사용하기

Was this content helpful?

Recent Feedback

Download README