Skip to Content

Indexing, Refresh, and Context Packs

Context Engine indexing should start with a narrow, approved repository scope.

Do not index every repository at once for a first rollout. Begin with one service where ticket-to-code automation is already working.

Confirm:

  • repositories are synced and enabled in the controller
  • repository aliases and default branches are correct
  • Context Engine is enabled for the workspace
  • the selected runtime is ready
  • provider authorization is ready when AI-assisted indexing is required
  • include and exclude patterns match the repository
  • generated, vendor, build, and dependency folders are excluded

Use the Context Engine dashboard to estimate indexing scope before starting.

The estimate helps you understand:

  • selected repositories
  • approximate source size
  • include/exclude effect
  • expected index work
  • expected provider usage where available

If the estimate is too large, narrow the source set before indexing.

After reviewing the estimate, start indexing from the Context Engine page.

Expected outcomes:

  • index status changes to running
  • progress events appear in the UI
  • terminal status becomes completed, failed, cancelled, paused, or another explicit final state
  • audit events are written where the deployment supports them

If indexing is already in a terminal state, a repeated start action may be ignored with a message that indexing already finished.

Vault refresh creates or updates Context Vault documents from indexed evidence.

Use refresh when:

  • repository code changed
  • documentation sources changed
  • previous vault documents are stale
  • reviewers found missing or incorrect context

If the refresh estimate is high, reduce scope or confirm provider credit availability before continuing.

During ticket execution, MergeLoom can request a bounded context pack for the job. The pack should be limited to repositories and documentation sources authorized for the workspace and relevant to the selected work repository.

Context packs should support the Investigation and Implementation stages. They do not replace repository guidance files such as MERGELOOM.md or AGENTS.md.

SymptomCheck
No repositories are availableRepository catalog sync, repository enabled state, and Context Engine feature access.
Cloud Context Engine is not readyEmail verification, payment method, cloud provisioning, model authorization, and Cloud AI credits.
Self Hosted Context Engine is not readyContext Engine URL, tokens, provider readiness, worker network path, and durable storage.
Indexing fails on generated filesAdd exclude patterns for dependency, build, generated, and runtime directories.
Context pack misses expected evidenceRepository scope, context source applicability, and whether the relevant source has been indexed/refreshed.