version: "1.0.0" name: diagnosing-analytics description: 'Use when the user asks how TraceDecay is used: tool or skill usage, hook capture, attribution, or adoption analytics. Load this skill before the first tool call.' metadata: opencode/slash: "true"

Diagnosing analytics

Durable usage analytics, project hook JSONL, and session archives are different sources. Read their supported diagnostics and attribution rather than querying private databases or treating one source's count as the complete history.

The analytics diagnostics sync imports new hook rows; its no-sync mode is a read-only view. Analytics MCP reads do not themselves refresh ingestion. Inspect included versus total rows, malformed input, provider attribution, and temporal coverage. A null project can mean global scope, not a broken project join.

For missing native hooks, inspect host registration and doctor evidence; Codex trust gating can prevent capture. Import is incremental, so replayed source rows must not be mistaken for new activity.

Invocation counts measure use, not task success. Compare outcomes, unnecessary calls, false triggers, and missed useful routing before recommending more skill or hint exposure. Trace a discrepancy to its capture, import, attribution, or read scope before changing configuration.

Report: each count with the source it came from (analytics, hook JSONL, or session archive) and its included and total rows. Do not present one source's count as the complete history.