garage
Generate agent performance visuals — watch faces, API gauges, failure states, livery presets. SVG and Unicode renderers for fleet status displays.
SKILL.mdskills/garage/SKILL.md
GitHubHUMMBL Garage
Agent Performance Index (API), livery presets, watch faces, and failure aesthetics. Implements automotive + horology + kintsugi metaphors for agent status visualization.
When to Use
- Rendering agent status as a watch face (analog clock metaphor)
- Generating API (Agent Performance Index) gauges (100-999 score)
- Applying racing livery presets (Martini, Gulf, JPS, Rothmans, Marlboro, Castrol)
- Visualizing failure states (degraded=wabi-sabi, broken=kintsugi, dead=death screen)
- Generating sparklines for activity history
Usage
[garage] watch "devin" --state executing
[garage] gauge "devin" --score 750 --class B
[garage] livery "devin" --preset gulf
[garage] failure "devin" --state degraded
[garage] sparkline "devin" --data "1,3,2,5,4,6,5,7"
Python API
from hummbl_garage import (
AgentPerformanceIndex,
LiveryPresets,
WatchFace,
FailureState,
)
# API score (100-999, 6 sub-ratings)
api = AgentPerformanceIndex(
agent_name="devin",
score=750,
sub_ratings={"speed": 80, "accuracy": 85, "coordination": 75, ...}
)
# Watch face (4-layer: analog hands, complications, dial finish, cockpit)
watch = WatchFace(agent_name="devin", state="executing")
# SVG rendering
from hummbl_garage.svg import render_watch_svg, render_api_gauge_svg
watch_svg = render_watch_svg(watch)
gauge_svg = render_api_gauge_svg(api)
# Unicode terminal rendering
from hummbl_garage.text import (
render_watch_unicode,
render_api_gauge_unicode,
render_failure_state_unicode,
render_sparkline,
)
watch_str = render_watch_unicode(watch)
gauge_str = render_api_gauge_unicode(api)
sparkline = render_sparkline([1, 3, 2, 5, 4, 6, 5, 7])
Key Concepts
- API (Agent Performance Index): 100-999 composite score with 6 sub-ratings (speed, accuracy, coordination, autonomy, reliability, trust)
- Watch face states: idle, planning, executing, verifying, blocked, offline
- Livery presets: Racing-inspired color schemes — Martini (red/blue/silver), Gulf (blue/orange), JPS (black/gold), etc.
- Failure aesthetics: degraded (wabi-sabi patina), broken (kintsugi gold seams), dead (death screen blue)
- Goodhart mitigation: Built-in gaming detection and held-out evaluation
Install
cd /work/active/oss/packages/python/hummbl-garage
pip install -e ".[test]"
Package
- Repo:
hummbl-io/oss - Path:
packages/python/hummbl-garage/ - License: Apache 2.0
- Dependencies: stdlib only