pr/5c1db6641 fix discover claude models via cli initi#252
Merged
plusplusoneplusplus merged 20 commits intoMay 31, 2026
Merged
Conversation
Owner
plusplusoneplusplus
commented
May 31, 2026
plusplusoneplusplus
commented
- fix: discover claude models via cli initialize
- feat(ralph): zero system-prompt injection (AC-01..AC-05)
- feat(effort-tiers): AC-01 — add effortLevels.enabled admin config flag
- feat(effort-tiers): AC-02 per-provider effort-tier storage and REST endpoints
- feat(effort-tiers): AC-03 — Effort Tiers editor on AI Provider page
- feat(effort-tiers): AC-04 — Effort Tier selector in new-chat composer
- feat(effort-tiers): AC-05 — Effort Tier selector in follow-up composer
- fix: add OpenAPI docs for effort-tier endpoints and fix missing mock in pending-task-info test
- feat(work-items): add goal leaf type under PBI
- feat(work-items): mobile add-child button and type-picker (AC-01 to AC-04)
- feat(effort-tiers): hardcoded per-provider defaults with source markers
- feat: add Reasoning Effort to conversation metadata popover
- Make diff classification skill-based
- Fix admin prompt test after classification prompt removal
- fix: propagate provider through cold resume in queue-follow-up
- Show effort tier model in tooltip
- Map Codex modes to sandbox settings
- Use Claude Opus 4.7 for Claude provider effort tier defaults
- feat(codex): always grant ~/.coc as a Codex accessible directory
Remove all Ralph-specific content from system messages across every Ralph phase. All Ralph framing now lives exclusively in user messages. AC-01: Execution iterations - system message is generic only - Remove RALPH_BASE_INSTRUCTIONS, buildRalphSystemMessage, and all Ralph framework strings from system message builder in ralph-executor.ts - buildRalphIterationPrompt now accepts progressPath, currentIteration, maxIterations and outputs: skill pointer -> dynamic context -> <goal> - Executor overrides effectivePrompt each iteration (bridge reuses old prompt, so executor rebuilds it fresh with correct iteration counter) AC-02: Final-check - system message is generic only - Remove buildRalphFinalCheckSystemMessage and RALPH_FINAL_CHECK_BASE_INSTRUCTIONS from final-check path in buildModeOptions - buildFinalCheckPrompt user message already carries full Ralph framing AC-03: Grilling - grill suffix moved from system to user message - Remove RALPH_GRILL_SUFFIX append to systemMessage.content in chat-base-executor - Prepend RALPH_GRILL_SUFFIX to effectivePrompt when phase === 'grilling' so every grilling turn carries the instruction AC-04: Remove iteration prompt override and retriever scaffolding - Delete RALPH_WORK_INTENT_PROMPT, PROMPT_PREFIX, RALPH_SPEC_CONTRACT_PROMPT, RALPH_ITERATION_PROMPT_DEFAULT_HEAD - Remove promptOverride from BuildRalphIterationPromptInput and BuildRalphIterationTaskInput - Remove synthesis promptOverride (no callers used it) AC-05: Tests lock in the new split - Negative system-message assertions for AC-01, AC-02, AC-03 - Positive user-message assertions (skill pointer, progress path, iteration counter, <goal> block, grill prefix) - Updated ralph-start-routes.test.ts to assert ultra-ralph in prompt Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
- CLIConfig / ResolvedCLIConfig: add effortLevels?: { enabled?: boolean }
- DEFAULT_CONFIG: effortLevels.enabled defaults to false
- CLIConfigSchema (Zod): add effortLevels object
- namespace-registry: add EFFORT_LEVELS_SOURCE_KEYS + effortLevels descriptor
- admin-config-fields: bool('effortLevels.enabled', ...) entry (live runtime)
- coc-client AdminResolvedConfig / AdminConfigUpdate: add effortLevels field
- RuntimeDashboardConfig.features: add effortLevelsEnabled boolean
- buildRuntimeDashboardConfig: populate effortLevelsEnabled from config
- SPA utils/config.ts: DashboardConfig field + isEffortLevelsEnabled() helper
- AdminPanel.tsx: state, init, dirty-check, save, cancel, UI toggle row
Tests: update snapshot in config.test.ts; add effortLevels cases to
admin-config-fields.test.ts and runtime-config-handler.test.ts (unit +
live end-to-end via RuntimeConfigService).
Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
…ndpoints
- Extend CLIConfig and ResolvedCLIConfig provider settings with effortTiers
(low/medium/high → { model, reasoningEffort? })
- Add getProviderModelSettings / writeProviderModelSettings support for effortTiers
- Add EffortTierEntry / EffortTiersMap types and getProviderCatalog helper
- Register GET /api/agent-providers/:provider/effort-tiers (read tier map)
- Register PUT /api/agent-providers/:provider/effort-tiers (single-tier upsert or
full-map replace with catalog + reasoningEffort validation)
- Add EffortTierEntry / ProviderEffortTiersResponse to coc-client contracts
- Add getEffortTiers / setEffortTier client methods to AgentProvidersClient
- Add 19-case test file covering: GET default/configured, PUT upsert/merge/replace,
invalid tier key, missing model, invalid model (400), invalid effort (400),
null reasoningEffort allowed, per-provider isolation, and bad-request cases
Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
Add ProviderEffortTiersSection component and useProviderEffortTiers hook for the admin UI. Three rows (Low/Medium/High) each with a Model dropdown and a dependent Reasoning Effort dropdown. Standard dirty/save/cancel card semantics using full-map PUT replace. - coc-client: add replaceEffortTiers() (full-map PUT) to AgentProvidersClient - hooks/useProviderEffortTiers: fetch/local-draft/save/cancel with dirty tracking - features/models/ProviderEffortTiersSection: admin editor component (aip-*/ar-* classes) - admin/AIProviderPage: lazy-loads and renders ProviderEffortTiersSection below models section - admin-redesign.css: aip-tier-table/aip-tier-label styles - 25 tests: 12 hook tests + 13 component tests (all passing) Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
- Add EffortTierSelector component (Low/Medium/High dropdown chip, mirrors EffortPillSelector style; unconfigured tiers greyed with tooltip) - Add resolveEffortTier / resolveEffectiveTier pure utilities - Wire useProviderEffortTiers into NewChatArea; when effortLevels.enabled is true and the active provider has ≥1 tier configured, the tier selector replaces the model picker + effort pill in the toolbar - Zero-tier fallback: legacy controls reappear when all tiers unconfigured - Last-picked tier persisted to localStorage (coc:effort-tier:<workspaceId>); first-time default = Medium; auto-fallback when stored tier goes unconfigured - handleSend resolves the selected tier to model+reasoningEffort at send time; legacy path unchanged when flag is off or zero tiers - Tests: 9 resolveEffortTier, 11 EffortTierSelector, 9 NewChatArea effort-tier (all 66 affected tests pass; tsc --noEmit clean) Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
- Wire EffortTierSelector into FollowUpInputArea via 4 new props (useEffortTierMode, effortTierMap, selectedEffortTier, onEffortTierChange) - ChatDetail owns tier state (selectedFollowUpEffortTier), fetches tier map via useProviderEffortTiers(sessionProvider), persists last-picked tier to localStorage, and passes tier-resolved model+effort to useSendMessage - When effortLevels.enabled is true and provider has >=1 tier configured, EffortTierSelector replaces the model picker + effort pill in follow-up toolbar - Zero-tier fallback: legacy controls reappear when flag off or no tiers - Tests: 12 new FollowUpInputArea effort-tier tests (all pass) - Fix missing isEffortLevelsEnabled / getEffortTiers / useProviderEffortTiers mock entries in 4 existing test files that render NewChatArea or ChatDetail Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
…in pending-task-info test
GAP-01: add GET /api/agent-providers/{provider}/effort-tiers and
PUT /api/agent-providers/{provider}/effort-tiers path entries to
openapi.yaml, including EffortTierEntry, EffortTiersMap,
ProviderEffortTiersResponse, EffortTierUpsertRequest, and
EffortTiersMapRequest schemas in the components/schemas section.
GAP-02: add isEffortLevelsEnabled: () => false to the vi.mock for
utils/config in pending-task-info.test.tsx so ChatDetail no longer
errors when calling isEffortLevelsEnabled() during render. Fixes 16
previously failing tests (exit code 1 → 0).
Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
Introduce a third leaf work-item type, 'goal', alongside work-item and bug under PBI. A goal is outcome-oriented and carries optional success criteria. Add a 'drafting' status for the pre-planning spec phase and a grillSessionId field to link a future spec-drafting session. - types + contracts: goal type (leaf, parent=pbi), drafting status, successCriteria/grillSessionId fields, transitions, rollup typings - routes: thread successCriteria + grillSessionId through create/update - SPA: orange GOAL pill/label/prefix, create dialog goal option + success criteria field, drafting status badge, goal success-criteria panel in detail - tests: type rules, route create/update, drafting transition, SPA layout Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
…C-04)
- WorkItemHierarchyNode: add isMobile + onAddChild props; render always-visible
'+' button on container nodes (epic, feature, pbi) when isMobile is true
- WorkItemHierarchyTree: add isMobile prop; add handleAddChild callback that
calls onCreateItem directly for single-child-type parents (epic→feature,
feature→pbi) and opens a bottom-sheet TypePickerModal for multi-child-type
parents (pbi→work-item/bug/goal); TYPE_LABELS used for picker labels
- WorkItemDetail: add isMobile + onCreateChild props; render 'Add Child' button
for container items on mobile, not gated by hierarchyEnabled; inline
child-type-picker modal mirrors tree flow
- WorkItemsTab: pass isMobile={isMobile} to WorkItemHierarchyTree and
WorkItemDetail; pass onCreateChild={openCreateDialog} to WorkItemDetail
- Tests: 27 new assertions in WorkItemHierarchy.layout.test.ts covering
AC-01 (node button), AC-02 (type picker in tree), AC-03 (detail pane button),
AC-04 (mobile-only gate); all 61 tests pass
Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
Defaults for copilot/codex/claude effort tiers (low/medium/high) live in coc-agent-sdk via getDefaultEffortTiers and mergeEffortTiersWithDefaults. The agent-providers GET endpoint merges stored config with defaults and returns a per-tier 'source' marker plus a separate 'defaults' map; PUT continues to persist only user-configured tiers (defaults never baked in). Admin UI shows a 'Default' badge for default-sourced rows and the Clear action reverts to the default rather than deleting. Hook tracks remote/local/defaults separately and dirty-checks only config-sourced entries. Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
Show the reasoning effort level (low/medium/high/xhigh) in the ConversationMetadataPopover, sourced from config.reasoningEffort (set at process creation time) with fallback to metadata.reasoningEffort. The row appears after Agent Provider. Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
Co-authored-by: Codex <codex@openai.com>
Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
When a chat session could not be warm-resumed (cold path), the new task was built without forwarding payload.provider from the parent task. This caused getTaskChatProvider() on the WS broadcast to return undefined, falling back to Copilot green even when the original chat used Claude/Codex. Fix: carry payload.provider from the parent task into the cold-resume task spec, matching how workingDirectory and workspaceId are forwarded. Add focused unit tests (queue-follow-up-provider.test.ts) that directly exercise the cold-resume path via a mock bridge: - provider: 'claude' is inherited by the new task payload - absent provider leaves the new task payload provider-free Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
Update medium and high tier defaults for the claude provider from claude-opus-4.8 to claude-opus-4-7. The copilot-sdk provider defaults are unchanged. Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
Mirror the Claude provider behavior by always appending ~/.coc to the Codex thread additionalDirectories, alongside caller-supplied skill and additional directories. The entry is de-duplicated case-insensitively on Windows and only added when not already present. Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
…vior The 'Map Codex modes to sandbox settings' change made interactive Codex threads use read-only sandbox (only plan/autopilot use danger-full-access), and the effort-tier tooltip now prefixes the tier label. Update the stale forge codex-sdk and coc EffortTierSelector tests to match. Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
plusplusoneplusplus
deleted the
pr/5c1db6641-fix-discover-claude-models-via-cli-initi
branch
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
1 participant
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.