Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →navishachiku/agentskill-config
Stores and retrieves per-user, per-project preferences for AI skills via a unified JSON-backed CLI. Loaded by consumer skills that declare a '## Skill Config' section with keys to persist across invocations.
Passed 19 of 21 test cases with the skill, and 8 of 21 without it.
and 14 more in the full report
12 fixed8 passed either way2 still failing0 broke
Tokens, turns and time describe resource use, not a dollar price.
One of 12 test cases the skill fixed. Same model, same prompt, with and without the skill.
What is the relative file path to the Node.js configuration script within the skill directory structure?
| Checked | Without skill | With skill |
|---|---|---|
| Specifies scripts/agentskill-config.js as the script path. | Missed | Passed |
The instructions your model reads when the skill loads.
Two-layer JSON files with per-key override priority: project overrides global; within the same layer, <skill-name> overrides _shared.
| Scope | Path |
|---|---|
global | ~/.agents/skill_config.json |
project | <project-root>/.agents/skill_config.json |
<project-root> is resolved via git rev-parse --show-toplevel. The project layer path should be gitignored — personal preferences are not for team version control.
_shared is a reserved top-level key used when multiple skills share the same conceptual preference. Read applies a 4-layer automatic fallback:
1. project.<skill-name>.<key>
2. project._shared.<key>
3. global.<skill-name>.<key>
4. global._shared.<key>
The first non-null value is used.
<skill>= the agentskill-config directory (resolved from the load path by the agent);<X>= consumer skill name (e.g.git-open-pr) or_shared.
SC=<skill>/scripts/agentskill-config.js
# Read (auto-fallback; outputs empty string and exits 0 if not found)
value=$(node "$SC" read <X> <key>)
# Write (writes to the corresponding file by scope; mkdir -p automatically; idempotent merge)
# When writing to _shared, the CLI prints an impact-scope warning to stderr — agent should confirm with user before proceeding
node "$SC" write <X> <key> "<value>" --scope <global|project>
test cases passed, vs 8 / 21 without
More skills from the same source and nearby in the registry, with their headline benchmark lift.