SuperAPI AI Directory
Partner中文AI newsFavorites
Skillsenvironment-check

environment-check

Deployment & operations

View repository

Updated 10-16 08:02

About this skill

Environment Check

Run the repository's environment checker and present its report. Choose the branch before selecting a model. This deterministic command runs within invocation authority and dispatches no worker.

Run

The checker ships in the shared library, so it is present wherever the skills are installed. Resolve SHARED_DIR exactly as the shared library contract describes (the loaded shared skill's directory; skills/shared/ in a source checkout; shared/ beside the installed skills), then run:

SHARED_DIR="<absolute shared skill directory>"
python3 "$SHARED_DIR/scripts/check_environment.py" --json

In a rar-skills source checkout, bash scripts/check-environment.sh is an equivalent wrapper and stays the documented form for setup guides.

If python3 itself is missing or below 3.11, the script cannot start. Present the Python setup instructions from docs/machine-setup.md (section 1) and stop; no other check ran.

Add flags only when the user's context calls for them:

  • --project /absolute/path/to/repo when the user asks about a specific project.
  • --browser auto|playwright-cli|agent-browser|none when browser work matters. Use an explicit or inherited selection when supplied. Otherwise auto is the safe scoped default. Ask only when a choice changes scope, cost, permitted effects, or a material unresolved decision.
  • --native-models when the user relies on host model access instead of external CLIs.

Present

Parse the JSON and show:

  1. Verdict line from summary.status:
    • local_checks_passed — every required check is OK.
    • needs_verification — nothing missing, but named warnings need the user's action or a live test.
    • missing_requirements — required tools are absent; setup is incomplete.
  2. Check table: one row per entry in checks — status (OK / WARN / MISSING), name, whether it is required, and its detail. Group MISSING first, then WARN, then OK. Do not drop optional checks; label them.
  3. Actions: for every WARN or MISSING row with a non-empty action, list the action verbatim, numbered, in check order. These are instructions for the user — do not execute installs, logins, or browser launches on their behalf.
  4. Limits: show the limits list so the user knows what was not tested (no provider request, no installation, no login; a present credential is not a valid one).

Keep the report to the verdict, the table, the actions, and the limits. Point to docs/machine-setup.md in the rar-skills repository for setup detail.

The load-bearing rule is green only from a fresh run: a check is OK only because this invocation's output says so. Never carry a green forward from memory, an earlier session, or the user's claim; suggest re-running after the user completes an action and report only what the new output shows.