Minimal domain skill
Minimal example Observe-phase domain skill. Checks disk usage and records it. Copy this as a starting point for your own domain.From its SKILL.md
npx -y skills add mataeil/OODA-loop --skill minimal-domain-skillAssembled from the repository path, not quoted from the project. Check it against their README if it does not work.
2 things to look at
- 5 stars5 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.
- runs commandsInstructs the agent to run 2 commands, including `ls agent/safety/HALT 2>/dev/null && echo "HALT_ACTIVE" || echo "HALT_INACTIVE"` and 1 more.
SKILL.md
2.0 KB, 435 tokens by cl100k_base, as published. Nobody here has run it
/scan-disk — Minimal Domain Skill Example
The smallest useful Observe skill: it reads disk usage, records a baseline, and emits an alert when usage crosses a learned threshold. Use it as a 30-line template for your own domain. Every skill starts with a HALT check.
Step 0: Safety
ls agent/safety/HALT 2>/dev/null && echo "HALT_ACTIVE" || echo "HALT_INACTIVE"
If HALT is active, print the reason and stop immediately.
Step 1: Observe
df -P / | tail -1 | awk '{print $5}' | tr -d '%' # → integer percent used
Load the lens (agent/state/disk_usage/lens.json) if present. Use its
learned_thresholds for warn_pct if it has one; otherwise default warn_pct = 85.
Step 2: Record + alert
Write agent/state/disk_usage/state.json:
{
"schema_version": "1.0.0",
"last_run": "<now ISO-8601>",
"status": "healthy", // healthy | degraded | critical
"metrics": { "used_pct": 73 },
"alerts": [] // e.g. ["disk 91% > warn_pct 85"] when crossed
}
Set status: "degraded" and add an alert when used_pct >= warn_pct;
"critical" at >= 95. The evolve engine reads status/alerts during Observe
and factors them into scoring.
Adaptive Lens (optional but recommended)
evolve's Step 5-E initializes and updates agent/state/disk_usage/lens.json
automatically. To participate, just emit consistent metrics/alerts; the engine
learns a per-host warn_pct over time (raising it on repeated false positives,
lowering it after real incidents). You don't write the lens yourself.
What ships with it: 1 file
1.3 KB alongside SKILL.md
- README.md1.3 KB