dsh-pixinsight-pipeline

A DeepSeek Harness (DSH) plugin that packages the pixinsight-pipeline skill — automated deep-sky astrophotography processing driven from DSH through a file-based IPC bridge into a running PixInsight 1.9.4.

It bundles three things so it works as a normal DSH plugin:

1. the skill (skills/pixinsight-pipeline/SKILL.md + reference/ knowledge base),
2. the bridge CLI (lib/pixinsight-mcp/scripts/pi-tool.mjs) and its ops modules,
3. the PJSR watcher (lib/pixinsight-mcp/pjsr/pixinsight-mcp-watcher.js) plus the headless WBPP wrapper (lib/pixinsight-mcp/run-wbpp.ps1).

DSH agent ── pi-tool.mjs ──> ~/.pixinsight-mcp/bridge/commands/{id}.json
                                   │
       PixInsight ── pixinsight-mcp-watcher.js (PJSR poll loop)
                                   │  reads command, runs PJSR, writes result
                                   ▼
                           ~/.pixinsight-mcp/bridge/results/{id}.json

For AI agents (read me first)

This package is only the director side. Before calling pi-tool, PixInsight must be running with the bundled watcher script loaded, otherwise every call fails with PixInsight process not found. The exact launch / verify / drive / shutdown steps are in AGENTS.md — start there, then read skills/pixinsight-pipeline/SKILL.md for the processing order.

Requirements

Install

A. As a DSH plugin (recommended)

powershell -ExecutionPolicy Bypass -File scripts\install-as-plugin.ps1 -Profile desktop

It backs up the profile package.json, copies the package into the profile node_modules, deploys the bridge toolkit to %DSH_HOME%\pixinsight-mcp, and appends the package to dsh.profile.bundles. Then fully restart DSH.

Uninstall: ... -Uninstall.

B. As a plain skill-root install

powershell -ExecutionPolicy Bypass -File scripts\install-into-dsh.ps1

Copies the skill to %DSH_HOME%\skills\pixinsight-pipeline and the toolkit to %DSH_HOME%\pixinsight-mcp. Use one of A / B, not both (otherwise the skill appears twice).

First run

1. Open PixInsight and load the watcher: Script > Run Script File… → %DSH_HOME%\pixinsight-mcp\pjsr\pixinsight-mcp-watcher.js. Keep PixInsight open. The watcher logs PixInsight MCP Watcher v0.1.0 and polls every ~1 s.
2. Verify the bridge:

   node "%DSH_HOME%\pixinsight-mcp\scripts\pi-tool.mjs" list_open_images '{}'
   

   [] means alive.

Usage

The agent reads skills/pixinsight-pipeline/SKILL.md for the processing order and parameter guide, then drives each step. Prefer the @ JSON form on Windows (PowerShell mangles inner double quotes):

$args = Join-Path $env:TEMP 'piargs.json'
'{"view_id":"MyTarget","denoise":0.25,"detail":0.15}' | Set-Content $args -Encoding UTF8
node "%DSH_HOME%\pixinsight-mcp\scripts\pi-tool.mjs" run_nxt "@$args"

Headless WBPP (preprocessing) does not need the watcher — run the official automation mode:

powershell -ExecutionPolicy Bypass -File "%DSH_HOME%\pixinsight-mcp\run-wbpp.ps1" `
  -InputDir "" -OutputDir ""

Layout

package.json                 # dsh.bundle.patch -> cordis.patch.yml
cordis.patch.yml             # - insert: id: pixinsight-pipeline, name: dsh-pixinsight-pipeline
AGENTS.md                    # step-by-step PixInsight watcher setup for AI agents
lib/index.js                 # registers the bundled skill via ctx.skills.registerProvider()
lib/pixinsight-mcp/          # bridge toolkit (pi-tool CLI, PJSR watcher, WBPP wrapper)
skills/pixinsight-pipeline/  # SKILL.md + reference/*.md
scripts/                     # install-as-plugin.ps1, install-into-dsh.ps1

Notes / gotchas

• pi-tool’s save_preview / save_variant use forward-slash paths; Windows backslash paths get their \U etc. eaten inside PJSR strings.
• ImageWindow.open() returns an array; use wins[0].mainView.
• PixInsight is memory-hungry — call w.purge() between heavy steps and close intermediates.
• The watcher runs in PixInsight’s PJSR engine; Ctrl+F11 (or a shutdown sentinel file in the bridge dir) stops it.
• Data processed with this plugin is local-only; never upload the user’s raw frames.

Credits

• Bridge design adapted from aescaffre/pixinsight-mcp (MIT).
• Star-stretch / Seti method: Seti Astro (Bill Blanshan).
• GHS: GHS Script (Mike Cranfield & Mark Shelley).
• BXT / NXT / SXT are commercial PixInsight modules by RC Astro (not redistributed).

License

MIT — see LICENSE.