Siril Astro Processor — Claude Code plugin

Process deep sky astrophotography from your terminal using natural language. Hand Claude Code a folder of light frames and get back a finished, ready-to-share image — no Siril GUI, no .ssf scripts, no menus to learn.

Built around the Siril CLI and StarNet2, with a curated pipeline that handles convert → register → stack → background-extract → denoise → star removal → stretch → recombine. Pick a named recipe (natural, starless, or punchy) for the star treatment, and the plugin generates and runs the right script for your data.

Demo

Talk to Claude Code:

"Process the photos in ~/AstroData/2025-LagoonNebula/"

The plugin will:

1. Run a preflight check (siril-cli present, StarNet configured, dataset writable).
2. Read a sample FITS header to detect OSC vs mono and confirm the Bayer pattern.
3. Generate a tailored .ssf script and run it via siril-cli.
4. Save before.png (raw stack autostretched) and after.png (final processed) side-by-side so you can see the result.

If you want a different look, say so:

"Make it the punchy version with stars at 50%"

The plugin re-runs only the recombine step and produces a new after_v2.png.

Install

In Claude Code, run these two commands:

/plugin marketplace add portkeys/siril-astro-processor
/plugin install siril-astro-processor@siril-plugins

The first command registers this repo as a plugin marketplace; the second installs the siril-astro-processor plugin from it.

Prerequisites

The plugin handles environment setup automatically on first use, but you'll need:

• macOS or Linux with Claude Code installed
• Siril 1.4+ (download)

  # macOS via Homebrew (cask):
  brew install --cask siril

• StarNet2 (only required for star removal)
  • On Apple Silicon Macs the Homebrew cask installer needs sudo. Ask Claude Code to "install StarNet2" — it will use the workaround documented in the skill (extract the binary to ~/Applications/, patch the rpath, configure Siril) automatically.
• Disk space: stacking 100 frames produces ~1 GB of intermediate FITS files. Plan accordingly.

How recipes work

Star treatment is the most personal post-processing decision. The plugin offers three named recipes; the rest of the pipeline is identical.

Recipe	What it does	When to pick it
natural (default)	Processed nebula + stars exactly as they appear in a single autostretch of the raw stack. Stars look real-sized.	Most cases. Closest to "what the photo would look like, but better."
starless	The processed nebula only — stars removed entirely.	When the nebula structure is the whole point and stars distract from it.
punchy	Recombined with an autostretched starmask, scale factor adjustable (0.3–1.0).	When you want more dramatic, contrasted stars over the nebula.

You can iterate just on the recipe — the expensive convert + stack steps run once and are cached.

What gets produced

In your dataset directory, alongside LIGHTS/:

result.fit                      # raw stack (linear)
result_bg_extracted.fit         # background-extracted
result_denoised.fit             # noise-reduced
result_starless.fit             # starnet output
result_enhanced.fit             # processed starless (autostretch + clahe + satu)
final.fit                       # final composite
before.png / before.jpg         # raw stack autostretched (the "before" preview)
after.png / after.jpg / after.tif  # final result, multiple formats

Every intermediate is preserved, so you can re-run from any point if you want to try a different recipe or stretch.

Troubleshooting

The skill includes a thorough troubleshooting + macOS gotchas section that the agent reads automatically. Common issues:

• "No images found" — your frames must be inside a LIGHTS/ subfolder.
• Heavily green image — happens with OSC data without color calibration. Ask: "remove the green cast" — it'll add rmgreen.
• Stars dominate the image — you ran the punchy recipe with too high a scale factor. Ask for natural or lower the factor.

Contributing

Issues and PRs welcome. The skill itself is a single SKILL.md — start there if you want to add a recipe or fix a workflow.

Acknowledgments

The processing workflow this plugin automates was originally taught by Paolo Nicosia at the Chabot Space & Science Center astrophotography workshop. His step-by-step Siril tutorial is the source of every pipeline choice here — convert → register → stack → background extraction → denoise → starnet → stretch → enhance. The plugin packages that workflow so others can run it with a single sentence to Claude Code, but the recipe is his.

License

MIT