Telemetry
Auto-Skill includes privacy-first anonymous telemetry following the principles from the anonymous-telemetry agent specification.
Architecture
Telemetry operates on two layers:
- Local — SQLite database at
~/.claude/auto-skill/telemetry.dbfor effectiveness reports and event history - Remote — Fire-and-forget HTTP to
t.insightx.profor anonymous aggregate reporting (forwarded to Firebase GA4)
What We Collect
Anonymous, aggregate metrics only:
| Data | Example | Purpose |
|---|---|---|
| Event type | skill_used, search | Feature usage |
| Result counts | 5 results | Effectiveness |
| Timing | 45ms | Performance |
| Outcome | success / failure | Reliability |
| Agent name | claude-code | Compatibility |
| System info | darwin, node 22 | Compatibility |
| Tool version | 4.0.1 | Adoption |
What We Do NOT Collect
- No search queries
- No file names or paths
- No session IDs
- No skill content
- No IP addresses
- No personal information
Disable Telemetry
# Tool-specific
export AUTO_SKILL_NO_TELEMETRY=1
# Universal standard
export DO_NOT_TRACK=1
Add to ~/.bashrc or ~/.zshrc to disable permanently.
Automatic Opt-Out
Telemetry is automatically disabled in CI environments: GitHub Actions, GitLab CI, CircleCI, Travis CI, Buildkite, Jenkins.
Implementation
- Async fire-and-forget (never blocks user workflow)
- Silent failures (telemetry errors never surface to users)
- 2-second HTTP timeout
- Source:
src/core/telemetry.ts
Local Queries
auto-skill telemetry report # Aggregated effectiveness per skill
auto-skill telemetry events # Raw event log