sarpyx

name	sarpyx
description	Use when asked to run, automate, preflight, validate, or summarize sarpyx SAR processing jobs, including WorldSAR, pipeline recipes, any SNAP GPT operator declared in sarpyx/snapflow/op.py, operator discovery/help, tiling, H5-to-Zarr conversion, mission recipes, validation targets, Docker/SIF workflows, or HPC upload operations.

Sarpyx Processing

Use this skill to autonomously run sarpyx processing after enough concrete inputs are known. The authoritative CLI surface is:

• uv run sarpyx worldsar ...
• uv run sarpyx pipeline ...

Treat older examples such as uv run sarpyx --input ... as stale until uv run sarpyx --help proves otherwise. For arbitrary SNAP GPT operations, use scripts/snap_operator.py; it parses sarpyx/snapflow/op.py and validates the requested operator against the live snap_operators registry.

Preflight First

Before heavy processing or operational Makefile targets, run:

python .agents/skills/sarpyx/scripts/preflight.py --repo /Users/roberto.delprete/Downloads/sarpyx

Add known inputs, for example:

python .agents/skills/sarpyx/scripts/preflight.py \
  --repo /Users/roberto.delprete/Downloads/sarpyx \
  --mode worldsar \
  --input /data/product.SAFE \
  --output /data/out \
  --grid-path /data/grid.geojson \
  --gpt-path /opt/esa-snap/bin/gpt

Stop and ask for missing values when preflight reports blockers for input products, output root, grid, SNAP GPT, master/slave products, or required WKT.

Route Requests

• Use sarpyx worldsar for the normal single-product mission-aware workflow.
• Use sarpyx pipeline when the user names a recipe, wants product-type inference avoided, or needs master/slave inputs.
• Built-in pipeline recipes are s1_tops, s1_strip, tsx, csg, biomass, nisar, and s1_insar.
• s1_insar is double-product and needs --master and --slave; the other built-ins are single-product and need --input.
• Use --param NAME=VALUE for recipe options; values are parsed as JSON when possible.
• Use --h5-to-zarr-only only with sarpyx worldsar.
• Use scripts/snap_operator.py when the user asks for any SNAP GPT operation/operator from sarpyx/snapflow/op.py, including operators not wrapped by sarpyx pipeline or worldsar.

Full GPT Operator Coverage

• Do not maintain a copied operator list in this skill. snap_operator.py --list is the coverage source because it reads snap_operators from sarpyx/snapflow/op.py.
• Use exact operator names from --list; quote names containing spaces, for example "Multi-size Mosaic".
• Use --dry-run to validate registry coverage and command shape without SNAP installed. A dry run can prove the operator is declared and the command can be built; it cannot prove SNAP runtime parameters are sufficient.
• Use --help-operator --gpt-path /path/to/gpt before running unfamiliar operators so SNAP reports required -S sources and -P parameters.
• Use --source NAME=PATH for named source inputs, --input PATH for the default source, repeat --param NAME=VALUE, and append unusual GPT options with --raw-arg.

Autonomy Rules

• You may create output directories and run processing after preflight is clean.
• You may run make validate-*, Docker, SIF, and HPC upload targets only when the user explicitly asks for that operational class.
• Do not delete existing products, prune Docker, publish packages, push images, or push SIFs unless the user explicitly asks for that exact action.
• Prefer explicit --output, --cuts-outdir, --grid-path, --gpt-path, and --snap-userdir values for reproducibility.
• For SNAP-heavy jobs, start with conservative --gpt-memory, --gpt-cache-size, --gpt-parallelism, and --gpt-timeout values supplied by the user or project defaults.
• For generic SNAP operators, first list or inspect the operator unless the user already provided exact -S sources and -P parameters.

Command Shapes

WorldSAR:

uv run sarpyx worldsar \
  --input /data/product.SAFE \
  --output /data/out \
  --cuts-outdir /data/out/tiles \
  --grid-path /data/grid.geojson \
  --gpt-path /opt/esa-snap/bin/gpt

Single-product recipe:

uv run sarpyx pipeline s1_tops \
  --input /data/product.SAFE \
  --output /data/out \
  --grid-path /data/grid.geojson \
  --gpt-path /opt/esa-snap/bin/gpt

Double-product recipe:

uv run sarpyx pipeline s1_insar \
  --master /data/master.SAFE \
  --slave /data/slave.SAFE \
  --output /data/out \
  --gpt-path /opt/esa-snap/bin/gpt

Generic SNAP operator from sarpyx/snapflow/op.py:

python .agents/skills/sarpyx/scripts/snap_operator.py \
  --repo /Users/roberto.delprete/Downloads/sarpyx \
  --operator Calibration \
  --input /data/product.SAFE \
  --target /data/out/product_cal.dim \
  --format BEAM-DIMAP \
  --gpt-path /opt/esa-snap/bin/gpt \
  --param selectedPolarisations=VV

For operator discovery or parameters:

python .agents/skills/sarpyx/scripts/snap_operator.py --repo /Users/roberto.delprete/Downloads/sarpyx --list
python .agents/skills/sarpyx/scripts/snap_operator.py --repo /Users/roberto.delprete/Downloads/sarpyx --operator Terrain-Correction --help-operator --gpt-path /opt/esa-snap/bin/gpt
python .agents/skills/sarpyx/scripts/snap_operator.py --repo /Users/roberto.delprete/Downloads/sarpyx --operator "Multi-size Mosaic" --dry-run

Use --source NAME=PATH for non-default source parameters, repeat --param NAME=VALUE for -P parameters, and use --dry-run before destructive or expensive runs.

After Execution

Summarize the actual artifacts, not just command success:

• Final/intermediate .dim products.
• Tile files such as .h5, .zarr, .tif, .npz, .npy, or .pkl.
• Cut reports, validation PDFs, manifests, JSON summaries, and logs.
• Any failed checks, missing artifacts, skipped validation, or residual runtime risk.

Use narrow searches rooted at the declared output and cuts directories, for example:

rg --files /data/out /data/out/tiles | rg '\.(dim|h5|zarr|tif|tiff|npz|npy|pkl|txt|pdf|json)$'