Using hyperproc with an AI assistant¶
The repository carries an agent skill: a set of instructions and helper scripts that teach Claude Code and Codex how this package works. It is distributed with the source, not with the wheel, and installing it is optional.
What it carries¶
The skill is not a copy of this documentation. It holds the things that are easy to get wrong and that fail quietly:
- Which grid a window indexes, per sensor and level. Equal index ranges are not automatically the same ground: EMIT's atmospheric route works on the sensor grid while the default reader maps through the GLT, and PRISMA L1 is a swath where L2D is a mapped UTM grid.
- Work-directory rules for atmospheric retrieval, including the symptoms of a reused run: identical medians, repeated stored runtimes, or a retrieval that finished suspiciously fast on a different region.
- What a diagnostic verdict means.
skip,refuseandinconclusivefrom the topographic gate, and an angular test that reports it cannot tell, are results to report rather than obstacles to work around. - Which products can be searched and which have to be ordered, since the readers cover more than any public archive publishes.
- Helper scripts that read your installed package rather than a stored copy, so the sensor matrix and the API signatures cannot drift from the version you actually have.
Installing it¶
git clone --depth 1 --filter=blob:none --sparse https://github.com/FujiangJi/hyperproc.git
cd hyperproc && git sparse-checkout set skills
./skills/install.sh --claude
The last line takes --claude, --codex or --both.
Both are global: they apply in every project on this machine.
| Option | Installs to |
|---|---|
--claude |
~/.claude/skills/hyperproc |
--codex |
~/.codex/skills/hyperproc, and a routing entry in ~/.codex/AGENTS.md |
Codex reads ~/.codex/AGENTS.md from wherever it happens to be running, so
the routing entry is written with absolute paths to where the skill actually
landed. An existing AGENTS.md is never overwritten: if one is there and
does not already mention hyperproc, the entry is written beside it as
AGENTS.hyperproc.md for you to merge.
Running the installer again updates an earlier install of this skill, so
git pull followed by the same command is how you take an update. Anything
else already at the destination is left alone.
Working inside a clone of this repository needs none of this. The root
AGENTS.md and .claude/skills/hyperproc are already in place, and running
the installer there will say so rather than duplicating them.
Checking it¶
python skills/hyperproc/scripts/env_report.py
python skills/hyperproc/scripts/sensor_table.py
python skills/hyperproc/scripts/validate_runtime.py
The first reports the interpreter, the installed version, which extras are importable and which archives hold a credential, as booleans and never a secret. The second prints the current reader and archive matrices. The third runs deterministic checks against the installed package and names each one.
What it does not do¶
It does not make the scientific decisions, download anything on its own, or read or write your credentials. It reports what a diagnostic found, including when the honest answer is that the data cannot support a conclusion.
The guidance was reviewed against a specific release. A newer hyperproc may behave differently, and the skill's own version maintenance notes say how to check.