Skill intelligence for Claude Code
Track skill invocation outcomes. Discover skills by community success rate.
claude plugins marketplace add sliday/claude-plugins
claude plugins install alit-
Track — after a skill runs, alit records success/failure/partial outcome using three signals:
- Heuristic: skill produced output without tool errors or exceptions
- LLM self-assessment: agent evaluates whether the skill achieved the user's stated goal
- User feedback: explicit thumbs up/down or verbal confirmation
-
Report — sends one anonymised telemetry row per invocation to the shared hive API. No PII, no user IDs, no correlation between reports.
-
Discover —
/find-skillmerges your local/mnt/skills/registry with community data. Ranks by relevance (40%), success rate (40%), usage volume (20%).
┌─────────────────────────────┐
│ Claude Code Agent │
│ │
│ ┌────────────────────────┐ │ ┌─────────────────────────┐
│ │ alit plugin │────────│ Hive API │
│ │ │ HTTPS │ Cloudflare Worker + D1 │
│ │ registry │◄───────│ │
│ │ tracker │ │ └─────────────────────────┘
│ │ anonymise │ │
│ │ find-skill │ │
│ │ hive-client │ │
│ └────────────────────────┘ │
│ │ │
│ │ scans │
│ ┌──────▼─────────────────┐ │
│ │ /mnt/skills/ │ │
│ │ local registry │ │
│ └────────────────────────┘ │
└─────────────────────────────┘
| Module | File | Role |
|---|---|---|
| registry | src/registry.ts |
Scans /mnt/skills/, parses SKILL.md frontmatter for name, description, tags, tracker config |
| tracker | src/tracker.ts |
Resolves outcome from heuristic + LLM + user signals into success | partial | failure | unknown |
| anonymise | src/anonymise.ts |
Strips PII via 13 regex patterns, caps summaries at 80 chars |
| find-skill | src/find-skill.ts |
Merges local scan with hive community data, ranks results by composite score |
| hive-client | src/hive-client.ts |
HTTP client for the 4 hive API endpoints with timeout and error handling |
Discover and rank skills by relevance and community success rate.
/find-skill "create PDF"
Returns ranked results:
[
{
"skill_name": "pdf",
"description": "Create, read, edit PDF documents",
"tags": ["document-creation"],
"installed": true,
"community_success_rate": 0.87,
"community_usage_count": 1423,
"source": "local+hive"
}
]Ranking formula: relevance (40%) + success_rate (40%) + usage_volume (20%)
Automatically tracks skill outcomes after execution. Resolves outcome priority:
- User feedback (explicit thumbs up/down) — highest weight
- LLM self-assessment (agent evaluates goal achievement)
- Heuristic (output produced without tool errors) — fallback
Base URL: https://skill-tracker-hive.stas6236.workers.dev
Submit anonymised invocation telemetry.
curl -X POST https://skill-tracker-hive.stas6236.workers.dev/report \
-H "Content-Type: application/json" \
-d '{
"skill_name": "pdf",
"outcome": "success",
"tags": ["document-creation"],
"summary": "Generated a formatted report from structured data",
"timestamp": "2026-03-30T10:00:00Z"
}'
# → {"id": "a1b2c3d4-..."}Validation: skill_name required, outcome must be success|partial|failure|unknown, summary max 80 chars.
Search skills by name.
curl -X POST https://skill-tracker-hive.stas6236.workers.dev/query \
-H "Content-Type: application/json" \
-d '{"query": "pdf"}'
# → {"skills": [{"skill_name":"pdf","total":142,"success_rate":0.87,...}]}Get aggregated stats for a single skill.
curl https://skill-tracker-hive.stas6236.workers.dev/stats/pdf
# → {"skill_name":"pdf","total":142,"successes":124,"failures":12,"partials":6,"success_rate":0.87,"last_seen":"2026-03-30 09:58:56"}Full skill catalog with success rates.
curl https://skill-tracker-hive.stas6236.workers.dev/catalog
# → {"skills": [{...}, {...}, ...]}| Field | Type | Description |
|---|---|---|
skill_name |
string | Skill identifier |
total |
integer | Total invocations |
successes |
integer | Success count |
failures |
integer | Failure count |
partials |
integer | Partial success count |
success_rate |
float | 0.0 – 1.0 |
last_seen |
string | Last report timestamp |
- No IP addresses stored — Worker does not persist IPs to D1
- No session or user identifiers — every report is a standalone row
- Summaries are LLM-generated — never raw user input, PII stripped before transmission
- Tags are author-defined constants — static, safe by construction
- No correlation possible — no device fingerprint, no cookie, no auth token
The anonymise module strips 13 patterns before any data leaves the client:
| Pattern | Example |
|---|---|
| Email addresses | user@company.com |
| Credit card numbers | 4111-1111-1111-1111 |
| US phone numbers | 555-123-4567 |
| International phone | +44 20 7946 0958 |
| IPv4 addresses | 192.168.1.1 |
| Social Security numbers | 123-45-6789 |
| macOS user paths | /Users/john/... |
| Linux user paths | /home/john/... |
| Windows user paths | C:\Users\john\... |
| UUIDs | a1b2c3d4-e5f6-... |
| OpenAI API keys | sk-abc123... |
| GitHub PATs | ghp_abc123... |
| Slack tokens | xoxb-abc123... |
All matches replaced with [REDACTED]. Summary capped at 80 characters.
Add a tracker block to your SKILL.md frontmatter to enable community telemetry:
---
name: my-skill
description: What my skill does
tags: [category-1, category-2]
tracker:
enabled: true
---| Setting | Effect |
|---|---|
tracker.enabled: true |
Reports outcomes to the community hive |
tracker.enabled: false or omitted |
Tracks locally only, does not report |
| Setting | Default | Description |
|---|---|---|
| Hive URL | https://skill-tracker-hive.stas6236.workers.dev |
Community telemetry endpoint |
| Timeout | 5000ms | HTTP timeout for hive requests |
| Scan dirs | /mnt/skills/ |
Directories to scan for installed skills |
MIT