Conversation
…splay - Add models.dev client with API caching for provider/model validation - Update OpenCode plugin to accept any provider (75+ supported) - Add model info display to TUI header (provider/model format) - Add model info display to output panel timing summary - Pass currentModel prop through RunApp and RunAppWrapper
- Enter now submits the message (instead of creating newline) - Shift+Enter or Ctrl+J inserts a newline - Updated hint text to reflect new key bindings
Add null checks for trackerState to prevent crashes when session data is incomplete or from older versions
- Add currentModel prop to ProgressDashboard for dashboard display - Add agentName and currentModel to RightPanel for output view display - Show agent|model in task output header when viewing output (o key) - Display model info in dashboard (d key) alongside agent and tracker
Fixes crash when switching to output detail view (o key)
Previously ProgressDashboard and RightPanel were using the configured agent name from storedConfig, which doesn't reflect when the engine switches to a fallback agent. Now they use displayAgentName which prefers activeAgentState.plugin from the engine.
Previously the textarea wasn't cleared until the parent component's setInputValue('')
took effect, which could cause a visible delay. Now we clear the textarea's
internal buffer immediately when submit is triggered.
|
The latest updates on your projects. Learn more about Vercel for GitHub.
|
|
Warning Rate limit exceeded
⌛ How to resolve this issue?After the wait time has elapsed, a review can be triggered using the We recommend that you space out your commits to avoid hitting the rate limit. 🚦 How do rate limits work?CodeRabbit enforces hourly rate limits for each developer per organization. Our paid plans have higher rate limits than the trial, open-source and free plans. In all cases, we re-allow further reviews after a brief timeout. Please see our FAQ for further information. 📒 Files selected for processing (2)
WalkthroughThreads a new optional currentModel prop through run/resume and TUI components, adds a disk-cached models.dev client, relaxes OpenCode provider validation, introduces configurable PRD skills with a task engine, silences certain progress-append errors, and enhances ChatView keyboard/submit behaviour. Changes
Sequence Diagram(s)sequenceDiagram
participant User as CLI User
participant CLI as create-prd
participant FS as Filesystem (skills_dir)
participant Engine as createPrdChatEngine
participant TaskEngine as createTaskChatEngine
participant UI as PrdChatApp
User->>CLI: run `create-prd --prd-skill <name>`
CLI->>FS: resolve skill path, read SKILL.md (prdSkillSource)
alt skill loaded
CLI->>Engine: invoke runChatMode with prdSkill/prdSkillSource
Engine->>Engine: build systemPrompt (from skillSource or skillName)
Engine->>TaskEngine: createTaskChatEngine for task prompts
Engine->>UI: start PRD UI with both engines and currentModel (if any)
else load fails
CLI->>User: print actionable error and exit
end
Estimated code review effort🎯 3 (Moderate) | ⏱️ ~35 minutes Poem
🚥 Pre-merge checks | ✅ 3✅ Passed checks (3 passed)
✏️ Tip: You can configure your own custom pre-merge checks in the settings. Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out. Comment |
Validate custom PRD skill usage when configured.
![coderabbitai[bot]](https://avatars.githubusercontent.com/in/347564?s=60&v=4){kind=small}
There was a problem hiding this comment.
Actionable comments posted: 0
Caution
Some comments are outside the diff and can’t be posted inline due to platform limitations.
⚠️ Outside diff range comments (1)
src/tui/components/ChatView.tsx (1)
142-142: Default hint text is inconsistent with new keyboard behaviour.The default hint mentions
[Ctrl+Enter] Sendbut the actual behaviour now usesEnterto send andShift+Enter/Ctrl+Jfor newlines. Update the default to match.📝 Suggested fix
- hint = '[Ctrl+Enter] Send [Esc] Cancel', + hint = '[Enter] Send [Shift+Enter/Ctrl+J] Newline [Esc] Cancel',
🧹 Nitpick comments (11)
src/session/persistence.ts (1)
579-610: Defensive null-checks look good, but consider aligning the type definition.The optional chaining on
trackerStateis a sensible safeguard against crashes from legacy or malformed session files. However, thePersistedSessionStateinterface (line 97) declarestrackerState: TrackerStateSnapshotas required, creating a type/runtime mismatch.Consider making
trackerStateoptional in the interface to reflect reality:- /** Tracker state snapshot */ - trackerState: TrackerStateSnapshot; + /** Tracker state snapshot (may be missing in legacy sessions) */ + trackerState?: TrackerStateSnapshot;This ensures TypeScript will flag any other code paths that assume
trackerStateis always present.src/engine/index.ts (2)
917-922: Consider retaining minimal error logging for observability.Silently swallowing errors here is reasonable since progress append is non-critical. However, completely removing error logging could mask persistent issues (e.g., disk full, permission errors) that would be valuable to diagnose. Consider a debug-level log or incrementing a counter for monitoring, rather than complete silence.
💡 Optional: Add debug logging
try { const progressEntry = createProgressEntry(result); await appendProgress(this.config.cwd, progressEntry); - } catch { - // Don't fail iteration if progress append fails + } catch (err) { + // Don't fail iteration if progress append fails + // Log at debug level for troubleshooting persistent issues + console.debug('[progress] Failed to append progress entry:', err); }
953-958: Same observability consideration as above.Same pattern as lines 917-922. If you choose to add debug logging there, apply it here as well for consistency.
src/tui/components/ChatView.tsx (1)
79-80: Stale JSDoc comment.The comment states "presses Ctrl+Enter" but the new behaviour uses Enter (without modifiers) to submit. Update the comment to reflect the current implementation.
📝 Suggested fix
- /** Callback when user submits (presses Ctrl+Enter) with the current input value */ + /** Callback when user submits (presses Enter) with the current input value */ onSubmit?: (value: string) => void;src/models-dev/index.ts (4)
117-121: Consider adding a timeout to the fetch request.The
fetchcall has no timeout configured, which could cause the application to hang indefinitely if the models.dev API is unresponsive. Consider adding anAbortControllerwith a reasonable timeout.⏱️ Suggested fix
async function fetchAndCache(): Promise<CacheData> { + const controller = new AbortController(); + const timeoutId = setTimeout(() => controller.abort(), 10000); // 10 second timeout + - const response = await fetch(MODELS_DEV_API_URL); + const response = await fetch(MODELS_DEV_API_URL, { signal: controller.signal }); + clearTimeout(timeoutId); + if (!response.ok) { throw new Error(`Failed to fetch models.dev API: ${response.status} ${response.statusText}`); }
78-92: Type safety note on cache loading.The
JSON.parseresult is typed asCacheDatabutvalidModelIdsis actually an array at that point (JSON doesn't support Sets). The code correctly handles this at lines 88-90, but the intermediate state has an incorrect type. Consider using an intermediate type for the parsed JSON.🔧 Optional type-safe approach
+interface CacheDataJson { + timestamp: number; + providers: ProviderInfo[]; + modelsByProvider: Record<string, ModelInfo[]>; + validModelIds: string[]; // JSON stores as array +} + function loadCache(): CacheData | null { try { if (!existsSync(CACHE_FILE_PATH)) { return null; } const content = readFileSync(CACHE_FILE_PATH, 'utf-8'); - const cache = JSON.parse(content) as CacheData; + const parsed = JSON.parse(content) as CacheDataJson; // Check if cache is expired - if (Date.now() - cache.timestamp > CACHE_TTL_MS) { + if (Date.now() - parsed.timestamp > CACHE_TTL_MS) { return null; } - // Restore Set from array - if (Array.isArray(cache.validModelIds)) { - cache.validModelIds = new Set(cache.validModelIds); - } - - return cache; + return { + ...parsed, + validModelIds: new Set(parsed.validModelIds), + }; } catch {
328-333:clearCacheonly clears in-memory cache.The
clearCachefunction setscacheData = nullbut doesn't remove the disk cache file. The next call togetModelsDevDatawill reload from disk if the file exists and isn't expired. This may be intentional, but consider documenting the behaviour or adding an option to also clear the disk cache.
198-210: Silent fallback to empty cache on fetch failure.When the API fetch fails and no valid cache exists, the code silently returns an empty cache. This is resilient but could lead to confusing behaviour where all provider/model validations pass (since there's nothing to validate against). Consider logging a warning or exposing the failure state.
src/tui/components/Header.tsx (1)
160-166: Consider extracting model parsing to a shared utility.This model parsing logic is duplicated across
Header.tsx,ProgressDashboard.tsx, andRightPanel.tsx. Consider extracting it to a shared utility function intheme.tsor a dedicatedutils.tsfile to reduce duplication and ensure consistent behaviour.♻️ Suggested utility function
// In src/tui/utils.ts or src/tui/theme.ts export function parseModelInfo(currentModel: string | undefined): { provider: string; model: string; full: string } | null { if (!currentModel) return null; const [provider, model] = currentModel.includes('/') ? currentModel.split('/') : ['', currentModel]; return { provider, model, full: currentModel }; }Then use it in components:
- const modelDisplay = currentModel - ? (() => { - const [provider, model] = currentModel.includes('/') ? currentModel.split('/') : ['', currentModel]; - return { provider, model, full: currentModel }; - })() - : null; + const modelDisplay = parseModelInfo(currentModel);src/tui/components/RightPanel.tsx (2)
473-479: Minor: Non-null assertions can be simplified.Since the IIFE is only executed when
timing.modelis truthy (checked in the ternary condition), the non-null assertions (!) on lines 476-477 are safe but could be avoided by capturing the value in a closure variable.♻️ Cleaner approach
const modelDisplay = timing.model ? (() => { - const [provider, model] = timing.model!.includes('/') ? timing.model!.split('/') : ['', timing.model!]; - return { provider, model, full: timing.model! }; + const m = timing.model; + const [provider, model] = m.includes('/') ? m.split('/') : ['', m]; + return { provider, model, full: m }; })() : null;
573-579: Same model parsing pattern duplicated within the same file.This is the third instance of the model parsing IIFE within
RightPanel.tsxalone (also at lines 473-479). Extracting this to a local helper or the suggested shared utility would reduce repetition.
📜 Review details
Configuration used: Organization UI
Review profile: CHILL
Plan: Pro
📒 Files selected for processing (13)
src/commands/resume.tsxsrc/commands/run.tsxsrc/engine/index.tssrc/models-dev/index.tssrc/plugins/agents/builtin/opencode.tssrc/session/persistence.tssrc/tui/components/ChatView.tsxsrc/tui/components/Header.tsxsrc/tui/components/PrdChatApp.tsxsrc/tui/components/ProgressDashboard.tsxsrc/tui/components/RightPanel.tsxsrc/tui/components/RunApp.tsxsrc/tui/types.ts
🧰 Additional context used
📓 Path-based instructions (2)
**/*.{js,ts,jsx,tsx}
📄 CodeRabbit inference engine (CLAUDE.md)
Start all code files with a file-level JSDoc comment section explaining the file's purpose, prefixed with 'ABOUTME: '
Files:
src/tui/components/ChatView.tsxsrc/tui/components/PrdChatApp.tsxsrc/session/persistence.tssrc/engine/index.tssrc/tui/types.tssrc/plugins/agents/builtin/opencode.tssrc/commands/run.tsxsrc/tui/components/RunApp.tsxsrc/commands/resume.tsxsrc/tui/components/ProgressDashboard.tsxsrc/tui/components/RightPanel.tsxsrc/tui/components/Header.tsxsrc/models-dev/index.ts
**/*.{js,ts,tsx,jsx}
📄 CodeRabbit inference engine (AGENTS.md)
Start all code files with a file-level JSDoc comment section explaining the file's purpose, prefixed with "ABOUTME: "
Files:
src/tui/components/ChatView.tsxsrc/tui/components/PrdChatApp.tsxsrc/session/persistence.tssrc/engine/index.tssrc/tui/types.tssrc/plugins/agents/builtin/opencode.tssrc/commands/run.tsxsrc/tui/components/RunApp.tsxsrc/commands/resume.tsxsrc/tui/components/ProgressDashboard.tsxsrc/tui/components/RightPanel.tsxsrc/tui/components/Header.tsxsrc/models-dev/index.ts
🧬 Code graph analysis (4)
src/tui/components/RunApp.tsx (1)
src/tui/components/ProgressDashboard.tsx (1)
ProgressDashboard(76-151)
src/tui/components/ProgressDashboard.tsx (1)
src/tui/theme.ts (1)
colors(9-58)
src/tui/components/RightPanel.tsx (1)
src/tui/theme.ts (1)
colors(9-58)
src/tui/components/Header.tsx (2)
src/tui/types.ts (1)
HeaderProps(76-103)src/tui/theme.ts (2)
formatElapsedTime(203-215)colors(9-58)
🔇 Additional comments (22)
src/tui/components/ChatView.tsx (1)
161-195: LGTM!The keyboard handling logic correctly differentiates between plain Enter (submit) and modified Enter keys (newline). The implementation properly captures the textarea value before clearing, and the optional chaining on
key.preventDefault?.()handles cases where the method may not exist.src/plugins/agents/builtin/opencode.ts (3)
403-438: LGTM!The relaxed provider validation appropriately delegates to the OpenCode CLI, which has comprehensive knowledge of supported providers. The defensive type check ensures invalid types are caught early whilst allowing any valid provider string through.
87-108: LGTM!The provider field and initialisation logic correctly accept any non-empty string, aligning with the PR's goal of supporting 75+ providers through delegation to the OpenCode CLI.
455-461: Model string with multiple slashes is silently truncated before validation.The
split('/')destructuring only captures the first two segments. If the model string contains multiple slashes (e.g.,provider/model/extra), everything after the second segment is discarded without error. This is problematic because:
- The error message explicitly documents the expected format as
provider/model- The comment states validation is delegated to OpenCode CLI, but the truncated input is passed instead of the original
- A user providing
provider/model/extrawould receive no feedback that part of their input was removedTo address this, either reject strings with more than one slash, or preserve everything after the first slash:
model.split('/', 2)and rejoin asmodelName = segments.slice(1).join('/').src/tui/components/PrdChatApp.tsx (1)
500-504: LGTM!The hint text correctly reflects the new keyboard behaviour where Enter submits and Shift+Enter/Ctrl+J inserts newlines.
src/commands/resume.tsx (3)
89-95: LGTM!The
currentModelparameter is cleanly threaded through therunWithTuifunction signature, maintaining consistency with the run command implementation.
151-167: LGTM!The
currentModelprop is correctly passed to theRunAppcomponent, enabling the UI to display the current provider/model information during resumed sessions.
398-404: LGTM!The call site correctly passes
config.modelas thecurrentModelargument, ensuring the resumed session displays the same model information as a fresh run.src/models-dev/index.ts (1)
265-266: Edge case: model strings with multiple slashes.The
split('/')approach has a latent vulnerability. According to the models.dev API documentation, model IDs can contain slashes (e.g.,openai/gpt-5). If such a model ID is used in the input format, the destructuring at line 266 will discard everything after the second slash. For instance, an input likeprovider/openai/gpt-5would be incorrectly split into["provider", "openai", "gpt-5"]with only the first two parts captured, causing the validation to fail.Currently, the codebase examples use hyphens rather than slashes in model identifiers (e.g.,
anthropic/claude-3-5-sonnet), but the code should be defensive. Usesplit('/', 2)or similar to properly handle multi-part model identifiers.src/tui/types.ts (1)
101-102: LGTM!The new optional properties are well-documented with JSDoc comments and follow the existing naming conventions. The additions maintain backward compatibility as all new props are optional.
Also applies to: 138-139, 158-161
src/tui/components/Header.tsx (1)
219-232: LGTM!The conditional rendering logic correctly handles the optional model display and properly manages the pipe separators between agent, model, and tracker information.
src/commands/run.tsx (1)
544-546: LGTM!The
currentModelprop is correctly threaded through the component hierarchy fromconfig.modeltoRunAppWrapperand down toRunApp. The changes are minimal and focused on data flow wiring.Also applies to: 565-566, 691-691, 910-910
src/tui/components/ProgressDashboard.tsx (2)
18-19: LGTM!The
currentModelprop addition and parsing logic are correctly implemented. Note: The model parsing logic duplication was flagged in theHeader.tsxreview—the same refactor suggestion applies here.Also applies to: 79-79, 92-98
127-132: LGTM!The conditional model display rendering integrates well with the existing agent/tracker information display.
src/tui/components/RightPanel.tsx (1)
583-600: LGTM!The agent and model information is correctly propagated through the component hierarchy from
RightPanel→TaskDetails→TaskOutputView. The header rendering inTaskOutputView(lines 583-600) provides clear visual context for the active agent and model.Also applies to: 721-722, 754-755
src/tui/components/RunApp.tsx (7)
94-96: LGTM!The new
currentModelprop is well-documented and follows the existing prop pattern. The optional typing is appropriate since model info may not always be available.
298-298: LGTM!Prop destructuring follows the established pattern in this component.
392-394: LGTM!The
displayAgentNamecomputation correctly prioritises the active agent from engine state (which reflects runtime fallback scenarios) over the static config value. The nullish coalescing operator appropriately handles bothnullandundefinedcases fromactiveAgentState?.plugin.
1151-1151: LGTM!Propagating
currentModeltoHeaderenables consistent model display in the toolbar.
1158-1159: LGTM!
ProgressDashboardnow receives the runtime-deriveddisplayAgentNameandcurrentModel, which aligns with the component's updated interface shown in the relevant code snippet.
1197-1198: LGTM!Consistent prop propagation to
RightPanelin the tasks view mode.
1230-1231: LGTM!Consistent prop propagation to
RightPanelin the iterations view mode, maintaining parity with the tasks view branch.
✏️ Tip: You can disable this entire section by setting review_details to false in your review settings.
- Option + Left/Right: move cursor by word - Option + Up/Down: move cursor to buffer start/end - Option + Delete: delete word to the left - Option + Fn + Delete: delete word to the right - Shift + Option + Left/Right: select by word - Shift + Option + Up/Down: select to buffer start/end - Shift + Left/Right/Up/Down: select by character/line - Shift + Cmd + Left/Right: select to line start/end - Shift + Cmd + Up/Down: select to buffer start/end
The main branch had a cleaner implementation that defines trackerState once and uses it throughout. Using that approach.
There was a problem hiding this comment.
Actionable comments posted: 2
Caution
Some comments are outside the diff and can’t be posted inline due to platform limitations.
⚠️ Outside diff range comments (2)
src/tui/components/ChatView.tsx (2)
79-79: JSDoc comment is inconsistent with actual behaviour.The comment states "presses Ctrl+Enter" but the implementation at line 183 uses plain Enter (without modifiers) to submit. Update the documentation to reflect the new behaviour.
📝 Suggested fix
- /** Callback when user submits (presses Ctrl+Enter) with the current input value */ + /** Callback when user submits (presses Enter) with the current input value */
142-142: Default hint text does not match the new keyboard shortcuts.The hint says
[Ctrl+Enter] Sendbut Enter (without modifiers) now submits, and Shift+Enter/Ctrl+J insert newlines. Consider updating the default hint to accurately guide users.📝 Suggested fix
- hint = '[Ctrl+Enter] Send [Esc] Cancel', + hint = '[Enter] Send [Shift+Enter] New line [Esc] Cancel',
🤖 Fix all issues with AI agents
In `@src/commands/create-prd.tsx`:
- Line 9: Remove the unused readFile import to fix the TS6133 error: edit the
import statement that currently reads "import { access, readFile, stat } from
'node:fs/promises';" and drop readFile so only used symbols (access, stat) are
imported; ensure no other references to readFile exist in the file (e.g., in
functions like create-prd related code) before committing.
- Around line 45-46: Remove the dead field prdSkillSource from the CreatePrdArgs
interface: locate the CreatePrdArgs declaration and delete the prdSkillSource?:
string; member, and ensure no other code references it (notably
parseCreatePrdArgs and any callers) so the type reflects only used args; run
type checks to confirm nothing else relies on that property.
🧹 Nitpick comments (1)
src/plugins/agents/builtin/opencode.ts (1)
420-425: Consider using a limited split to handle potential slashes in model names.If a model string contains multiple slashes (e.g.,
"provider/model/variant"),split('/')will produce more than two elements, and the destructuring will only capture the first two, potentially losing part of the model name.💡 Suggested fix using limit parameter
if (model.includes('/')) { - const [provider, modelName] = model.split('/'); + const slashIndex = model.indexOf('/'); + const provider = model.slice(0, slashIndex); + const modelName = model.slice(slashIndex + 1); if (!provider || !modelName) { return `Invalid model format "${model}". Expected format: provider/model (e.g., anthropic/claude-3-5-sonnet)`; } // Provider and model name are passed through - OpenCode CLI validates them }This ensures that only the first slash is used as the delimiter, preserving any subsequent slashes as part of the model name.
📜 Review details
Configuration used: Organization UI
Review profile: CHILL
Plan: Pro
📒 Files selected for processing (8)
src/chat/engine.tssrc/commands/create-prd.tsxsrc/config/index.tssrc/config/schema.tssrc/config/types.tssrc/plugins/agents/builtin/opencode.tssrc/tui/components/ChatView.tsxsrc/tui/components/PrdChatApp.tsx
🚧 Files skipped from review as they are similar to previous changes (1)
- src/tui/components/PrdChatApp.tsx
🧰 Additional context used
📓 Path-based instructions (2)
**/*.{js,ts,jsx,tsx}
📄 CodeRabbit inference engine (CLAUDE.md)
Start all code files with a file-level JSDoc comment section explaining the file's purpose, prefixed with 'ABOUTME: '
Files:
src/config/schema.tssrc/tui/components/ChatView.tsxsrc/config/types.tssrc/chat/engine.tssrc/commands/create-prd.tsxsrc/config/index.tssrc/plugins/agents/builtin/opencode.ts
**/*.{js,ts,tsx,jsx}
📄 CodeRabbit inference engine (AGENTS.md)
Start all code files with a file-level JSDoc comment section explaining the file's purpose, prefixed with "ABOUTME: "
Files:
src/config/schema.tssrc/tui/components/ChatView.tsxsrc/config/types.tssrc/chat/engine.tssrc/commands/create-prd.tsxsrc/config/index.tssrc/plugins/agents/builtin/opencode.ts
🧬 Code graph analysis (1)
src/commands/create-prd.tsx (1)
src/config/index.ts (1)
loadStoredConfig(194-216)
🪛 GitHub Actions: CI
src/commands/create-prd.tsx
[error] 9-9: TypeScript error TS6133: 'readFile' is declared but its value is never read. (Command: tsc --noEmit during bun run typecheck)
🔇 Additional comments (14)
src/plugins/agents/builtin/opencode.ts (3)
87-88: LGTM!The type widening from a specific provider union to
stringaligns well with the objective of supporting 75+ providers and delegating validation to the OpenCode CLI. The comment clearly documents this design decision.
105-108: LGTM!The simplified provider validation correctly accepts any non-empty string, consistent with the design to delegate provider validity checks to the OpenCode CLI.
371-380: LGTM!The validation correctly ensures provider is a string type when provided, while appropriately deferring value validation to the OpenCode CLI.
src/tui/components/ChatView.tsx (4)
9-10: LGTM!The new imports for
useKeyboardandKeyEventare correctly added and used in the keyboard handling implementation.
161-170: LGTM!The submit handler correctly captures the current value before clearing the textarea, ensuring no data loss while providing immediate visual feedback to the user.
172-302: Well-implemented macOS-style keyboard shortcuts.The keyboard handler is well-structured with proper modifier key checks and early returns. The dependency array correctly includes all referenced values from the enclosing scope.
304-304: LGTM!The
useKeyboardhook is correctly wired up with the callback, and the conditional logic is appropriately handled inside the handler rather than around the hook call.src/config/types.ts (1)
179-180: LGTM!The new
skills_diroptional field is consistent with the existing pattern for optional string configuration properties and aligns with the corresponding schema and merge logic additions.src/config/index.ts (1)
166-166: LGTM!The
skills_diroverride follows the established pattern for scalar field merging, consistent with adjacent fields likeprogressFileandautoCommit.src/config/schema.ts (1)
129-130: LGTM!The schema addition for
skills_dircorrectly uses Zod 4's optional string validation and is appropriately placed alongside the relatedprompt_templatefield.src/commands/create-prd.tsx (2)
125-152: LGTM! Well-structured validation with clear error messages.The
validatePrdSkillfunction properly validates that the skill directory exists and contains the requiredSKILL.mdfile, with actionable error messages for each failure case.
274-284: LGTM! Good defensive validation.The validation logic correctly ensures
skills_diris configured before allowing--prd-skillusage, with clear error messages guiding users to the correct configuration location.src/chat/engine.ts (2)
22-37: LGTM! Good refactoring for configurable PRD skills.The extraction of
buildPrdSystemPromptandDEFAULT_PRD_SKILLenables runtime flexibility whilst maintaining backward compatibility via the staticPRD_SYSTEM_PROMPTexport.
347-359: LGTM! Clean conditional system prompt selection.The factory function correctly uses the provided
prdSkillto build a custom system prompt, falling back to the default when not specified.
✏️ Tip: You can disable this entire section by setting review_details to false in your review settings.
Clear node_modules/.cache, node_modules/.pnpm, and .tsbuildinfo before installing dependencies to prevent stale TypeScript/bun cache from causing false positives in unused variable checks.
Only remove .tsbuildinfo and node_modules/.cache to fix stale TypeScript cache issues without reinstalling all packages.
subsy
changed the title
Document new create-prd options and custom skills_dir configuration
- Add --prd-skill option to create-prd CLI docs - Add skills_dir configuration option docs - Add examples for custom skills usage - Update README with new commands
Add PRD format specification that custom skills must follow for downstream task creation compatibility
feat: models.dev integration, create-prd improvements, custom skills_dir, and various fixes
1 participant
Summary
This PR adds models.dev integration for provider/model validation, improves the create-prd chat mode, adds custom skills_dir configuration, and fixes several bugs.
Features
models.dev Integration
src/models-dev/index.tsclient that fetches and caches the models.dev API{agent} | {provider/model} | {tracker}create-prd Improvements
--skillflag)Custom Skills Directory
skills_dirconfiguration option to specify custom skills locationBug Fixes
getSessionSummarywhen session data is incomplete<span>to<text>for pipe separatorFiles Changed
New Files
src/models-dev/index.ts- models.dev API client (+333 lines)Modified Files
src/plugins/agents/builtin/opencode.ts- Less restrictive provider validationsrc/tui/components/ChatView.tsx- Enter to submit, textarea clearing, text editing shortcutssrc/tui/components/Header.tsx- Model display with pipe separatorssrc/tui/components/ProgressDashboard.tsx- Agent/model displaysrc/tui/components/RightPanel.tsx- Agent/model in output viewsrc/tui/types.ts- Added agentName and currentModel to RightPanelPropssrc/session/persistence.ts- Null checks for trackerStatesrc/commands/create-prd.tsx- Custom skill selection, skills_dir support.github/workflows/ci.yml- Clear TypeScript build cacheCommits
f969d94- fix: isolate task creation from PRD chatb071fd3- ci: clear only TypeScript build cache (not node_modules)a8c4a16- ci: clear build cache before install to avoid stale cache issuese808f62- feat: use PRD skill sources in prompts9ab242c- merge: resolve conflict in getSessionSummary with main1304809- feat: allow custom PRD skill selection0d34f7c- feat: add macOS-style text editing shortcuts to create-prd chat input73fac93- fix: honor skills_dir in merged configd9a40b8- feat: allow skills_dir configuration0590da9- debug: revert progress append logging3d27fcb- fix: clear textarea immediately on submit in create-prd mode8da528c- fix: use active agent from engine state in dashboard and output viewa62c18e- fix: use text instead of span for pipe separator in output viewec61696- feat: show agent and model info in dashboard and output vieweeb42c2- style: use pipe separator in toolbar for agent/model/tracker display629710e- fix: handle undefined trackerState in getSessionSummarycb00c7a- feat: change chat input to submit on Enter in create-prd mode9676709- feat: add models.dev integration for provider/model validation and displaySummary by CodeRabbit
New Features
Bug Fixes
Refactor
UX
Docs
Chores
✏️ Tip: You can customize this high-level summary in your review settings.