“See where skills actually activate across agent sessions, not just where they’re installed. Mine agent logs to find recurring mistakes.” One of the six platform components, surfaced through the web app rather than as a single CLI command.
The distinction in that sentence is the contribution. Installed is not loaded. A skill only fires when its description matches the task, so an organization can have a skill deployed everywhere and used nowhere, and the two states are indistinguishable from any install-side view. Activation telemetry is the only way to tell them apart — and the fix for a skill that never fires is a description problem, not a content one, which is why tessl-eval ships a mode that measures activation alone.
This is the wiki’s first capability that measures agent behavior in production rather than in a test. Every other measurement here runs against a scenario or a diff; this reads what actually happened across real sessions, and it is the input the tessl-agent mines for recurring mistakes.
It also completes the layer’s loop, which is worth stating as a whole: tessl-inventory finds what exists, observability finds what is used, tessl-eval proves whether use helps, tessl-review-security checks whether it is safe, and tessl-workspace policy decides what is allowed. Discover → observe → measure → verify → govern, applied to a corpus of instructions.
See Also
- tessl-agent — mines this data and proposes fixes.
- tessl-eval — its activation-only mode answers the same question in a test.
- tessl-inventory — what exists, as against what runs.
- pattern-measure-first — baseline before changing.