boluopai.Evidence of use
Investigate the smallest relevant task and period through the current host's available history, event records, tool outputs, or direct consumer observation. This method requires no particular telemetry service or new collector.
Bind the controlling request, instruction surface, task/turn identity, actual Skill body, resource read or script execution, and returned result where these records exist. Distinguish an explicit request to load a Skill from the model's selection, and distinguish reading a Skill as an audit target from applying it to the work.
Keep claims separate:
| Signal | Supported claim | Still unresolved |
|---|---|---|
| Catalog entry | Visible discovery metadata | Selection or successful retrieval |
| Read or injection receipt | Text reached the consumer | Application or effect |
| Task actions attributable to the method | Application in that task | Whether it improved the outcome |
| Observable task result | Outcome in the examined conditions | General or enduring effectiveness |
| Aggregate counters | Events counted within stated coverage | Task identity, causality, or unrecorded use |
Inspect collection coverage, retention, event meaning, and known gaps before interpreting counts. No received records is not proof of no use. Stop receipts are not automatically completion counts. A detected or injected Skill is not automatically an effective Skill. Report confidence and missing evidence rather than filling gaps with assumptions.
Compare relevant behavior with the user's purpose and actual instruction contract. Use observed failures to locate whether the problem belongs to wording, relational design, discovery, access, installation, or another owner. Avoid attributing the whole outcome to one Prompt when shared context or user participation also shaped it. Retain only the useful evidence permitted by the task's data boundaries; availability of a history tool grants no broader collection or publication authority.