skill-dispatcher
Skill1 starsChanged 35 days agoMITView on GitHub
Skill-library analytics: who used which skill, when, with which model, and which skills have gone stale. Regenerates the usage wallboard, rebuilds the skill registry, runs the staleness audit, and on request suggests the best-fitting skill for a task from registry metadata. Usage is logged automatically by harness hooks; this skill never has to be invoked to log anything. Use when the user asks for skill usage statistics, the wallboard, a staleness report, a registry rebuild, or explicitly asks which skill should handle something.
We split this skill into the rules it gives an agent
3 rules. About 0 are common, 1 is distinctive, and 2 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈1
Distinctive · estimate
This author's own choices. Worth a look.
2
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Very specific: line 25
Run scripts from the installed copy (~/.agents/skills/skill-dispatcher/).
- Distinctive: line 62
When the user explicitly asks which skill should handle a task:
- Very specific: line 87
Keep list values on one line (a, b, c).
≈ “Common” is an estimate: at least 1 in 5 of the 308,232 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
Author: jovd83 | Version: 4.0.2 | License: MIT
Skill Dispatcher
Analytics for the skill library. Since 4.0.0 the dispatcher no longer sits in front of every task: harness hooks log each skill use on their own (see hooks/README.md), and this skill turns that log into answers.
Run scripts from the installed copy (~/.agents/skills/skill-dispatcher/). Data lives outside the skill, in ~/.agents/dispatcher-data/, so reinstalls never touch it:
| Data | Path |
| Usage log (one JSON event per line) | ~/.agents/dispatcher-data/logs/dispatch_events.jsonl |
| Wallboard | ~/.agents/dispatcher-data/reports/wallboard.html |
| Skill registry | ~/.agents/dispatcher-data/registry/SKILL_REGISTRY.json (+ .md) |
| Hook script, state, debug | ~/.agents/dispatcher-data/hooks/ |
Tasks
Show usage / regenerate the wallboard
python scripts/generate_wallboard.pyReads the usage log and writes the wallboard: totals, most-used skills, models, recent activity, chains and staleness. Point the user to file:///<home>/.agents/dispatcher-data/reports/wallboard.html. For a quick answer without the page, read the log directly: rows written by hooks have intent = skill used (via <harness> hook).
Rebuild the skill registry
python scripts/build_registry.pyIndexes every installed SKILL.md (name, description, dispatcher-* metadata) into the registry. Its preflight also reports name drift and oversized frontmatter. Rebuild after installing, removing or renaming skills.
Staleness audit
python scripts/staleness_audit.py --days 90Lists installed skills with no logged use in the lookback window. Treat the result as a prompt for a decision (keep, make on-demand, archive), not as a verdict: a skill that was installed recently, or that is loaded in a harness without hooks, can look unused.
Recommend a skill (only when asked)
When the user explicitly asks which skill should handle a task:
python scripts/match_candidates.py --intent <normalized_intent> [--keywords k1,k2] [--stack s1,s2] [--max-risk medium] --format textAnswer with the top candidate, the runner-up, and one line on why, grounded in the matched metadata fields. Do not perform the task yourself as part of the recommendation, and do not route tasks the user did not ask to have routed: harnesses already pick skills from their descriptions.
Logging
Nothing to do. The hooks call scripts/dispatch_logger.py for every skill use in Claude Code, Codex, Grok and Antigravity (Copilot and Hermes are supported once registered). Do not run log-dispatch for normal skill use; it remains available for manual or scripted events only.
Skill metadata schema
build_registry.py and match_candidates.py read these namespaced keys from each skill's metadata: block:
| Key | Meaning |
dispatcher-category, dispatcher-capabilities, dispatcher-accepted-intents | What the skill does, as matchable terms |
dispatcher-input-artifacts, dispatcher-output-artifacts | What it consumes and produces |
dispatcher-stack-tags | Technologies it covers |
dispatcher-risk, dispatcher-writes-files | How careful a caller must be |
dispatcher-layer | information (reads), execution (changes things) or feedback (reviews) |
dispatcher-lifecycle | active, sunset (warn) or archived (ignore) |
Keep list values on one line (a, b, c); keep everything else in the body of the SKILL.md.
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.