clearingsBack to the site

Manage Clearings

Use ordinary language in the coding client. The bundled management skill resolves names, reads current state, applies the requested change, and reports the result.

Common requests

RequestBehavior
Show Clearings statusReport learning, client access, recent results, and service health
What did you learn?Show the working project's routine library
Open the workbenchOpen the local visual library and test interface
Pause learningStop automatic learning; keep saved routines usable
Learn twice daily, every two days, or weeklyChange the interval
Stop suggestionsDisable prompt hints independently of learning
Exclude this project or workflowRemove it from learning scope
Pause this routinePreserve its records but stop reuse
Undo the last automatic changeRestore the prior accepted version or deactivate the new routine

Workbench updates are user-directed changes. Use the named routine's undo control for them; the global automatic-change marker tracks automatic learning and improvement.

Library and history

clearings_library and library expose a bounded, project-scoped view shared with the workbench. It includes names, examples, current versions, controls, recent calls, and whether undo is available. Follow next_after for more entries.

Run history is paginated and scoped to the receiving project. Large records and missing evidence remain explicit. A history record identifies the version used, not just the current routine name.

Usage

Usage windows cover 7, 30, and 90 UTC calendar days, including today. reuse_calls, test_calls, and unclassified_calls are separate. These are caller-labelled observations, not a proof of savings. Acceptance evaluations are separate from real use.

Explicit trials use run-routine --purpose test or purpose: test through MCP. Test failures do not trigger automatic regression rollback. Only labelled reuse influences usage ranking and automatic-improvement eligibility.

Do not retire a routine merely because test or unclassified counts are low. Consider its purpose, age, recent real use, and whether another routine replaces it.

Retention and retirement

prune previews expired activity and run records; prune --apply removes those records under the configured retention period. Compact usage totals, immutable requirements, versions, evaluations, budget records, and change history remain available. Pruning does not modify source conversations.

Retiring a routine deactivates it and preserves a paused, excluded record. It is not permanent data erasure. Pause, exclusion, and retirement never grant additional access.

Advanced commands

The CLI and MCP also expose routine, manage, digest, background-jobs, performance, model-usage, and runs. Update operations use current revisions or versions where required. After a stale-state error, reread the target before retrying.

See daily use, workbench, and background learning.


View Markdown source

On this page