lsm / neokai / 25383057283

83%

Build Type

push

github

Committed by web-flow

Commit Message

Add configurable thinking levels for space agents (#1786)

* Add configurable space agent thinking levels

Allow space agents and workflow node slots to override thinking levels so runtime sessions can resolve slot-specific thinking before falling back to agent/app defaults.

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>

* Fix thinking level review issues

Persist agent thinking overrides from the editor, preserve node-slot overrides when switching modes, and clear stale session thinking config when init-derived overrides are removed.

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>

* Keep task thinking config out of space chat

Avoid applying Task Agent thinking-level overrides to user-facing space chat sessions so chat remains governed by normal app defaults.

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>

* Persist inherited thinking behavior correctly

Preserve single-agent shorthand overrides during workflow serialization and let sessions without explicit thinking overrides inherit the global app default at query time.

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>

* Address Codex P2 feedback and fix test schemas for thinkingLevel

- Preserve thinkingLevel in import replace path (space-export-import-handlers)
- Add migration 117 to backfill agent template hashes with thinkingLevel in fingerprint
- Carry disabledSkillIds when collapsing multi-agent to single-agent (NodeConfigPanel)
- Add thinking_level column to test helper schemas (space-agent-schema, space-test-db)
- Update agent-template-hash test to expect thinkingLevel in fingerprint

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>

* Remove thinkingLevel from template drift hash — user preference, not preset identity

thinkingLevel is a per-agent user preference, not part of the preset
identity. Including it in the template hash caused false drift when users
changed their thinkingLevel and sync-from-preset would silently overwrite
their prefer... (continued)

Coverage Stats

7975 of 11057 branches covered (72.13%)

Branch coverage included in aggregate %.

56 of 66 new or added lines in 13 files covered. (84.85%)

1 existing line in 1 file now uncovered.

64132 of 89891 relevant lines covered (71.34%)

251.26 hits per line