Repository navigation
Configure Claude subagent models and caching per managed agent - #209
Merged
Merged
Conversation
Keep model routing, cache retention and delegation limits in each agent's launch settings, with role overrides and durable recovery. Refs #193.
added 3 commits
October 26, 2026 08:39
Register the memory hold before agents become idle and publish complete fake forge models with atomic replacement. Cover a reader during a paused write and allow the source CLI contract test time to start both forges.
Use queues for the single-reader settings fixtures and wait for fake model reads before advancing their clock. This prevents lost startup notifications and assertions racing a scheduled read.
madskronborg
pushed a commit
that referenced
this pull request
Oct 26, 2026
…and let the peer rig's listener bind its own port The schedule test took a Run fired before the stop, whose execution went with the engine; the oversight test ran a real verification under a clock moving a minute per millisecond, which timed it out; the board test compared two clients that could straddle the host seeing the finish. Each now waits for the state it asserts. The Flock chat race was fixed on master (#209) and is only reproduced here. The peer rig picked a port by binding port 0 and closing it, so another process could take it before the host listened. COLLIE_PEER_PORT=0 now makes the listener bind what is free and keep that port for every later opening, so the rig names no port at all.
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Managed Claude agents currently inherit native subagent settings, so choosing cheaper workers or longer cache retention also affects ordinary Claude sessions. Add an optional Collie policy scoped to each managed Run agent, with workflow, role, scope, and call overrides.
The policy controls model selection and enforcement, cache retention, delegation availability, nesting, and concurrency. It preserves existing behavior when unset, records choices for recovery, rejects incompatible live-agent reuse, and checks Claude compatibility before opening a pane. Other Harnesses and chat sessions keep their existing behavior.
The branch includes the latest master fix for the criteria hand-back fixtures. It also fixes test races: the memory-budget test registers its hold before agents become idle, the fake forge atomically replaces its shared model, and chat fixtures retain refresh notifications until their reader starts and wait for model reads before advancing the clock. The source CLI contract test has an explicit timeout for starting both forge fakes.
Validation:
63a44f4: the full suite finished with 4,396 passes, 11 skips, and zero failures; build, smoke, contract, desktop, macOS, and security checks passed.Live provider caching, billing, and model-quality effects were not tested. No installed binary or runtime settings were changed. One-hour cache retention remains opt-in because longer cache writes can cost more for short workers.
Closes #193.