- Improved config display formatting to properly render arrays and objects as JSON instead of `[object Object]`.
- Enhanced type display in config list output to show correct type indicators for number, array, and record settings.
- Added test coverage for record settings rendering in config list output.
- Increased default async max jobs limit from 15 to 100 for improved concurrent task handling.
- Updated fallback default in SDK initialization to match schema configuration.
- Added support for four new exec message types: listMcpResources, readMcpResource, recordScreen, and computerUse.
- Imported four new result schemas to handle responses for MCP resource operations and computer use capabilities.
- Updated bash execution handler to use standardized 'cwd' parameter name instead of 'workdir'.
- Fixed skill discovery to continue loading project skills when user skills directory is missing.
- Added error handling in scanSkillsFromDir to gracefully handle ENOENT and log other read failures.
- Added test case verifying project skills load successfully despite missing user skills directory.
- Added topP, topK, minP, presencePenalty, and repetitionPenalty sampling control options to StreamOptions and AgentOptions interfaces.
- Implemented getter and setter properties on Agent class for runtime configuration of model sampling parameters.
- Added UI configuration and preset value providers for five new sampling parameters in coding-agent settings schema and components.
- Integrated sampling control parameters through proxy layer and agent session configuration for end-to-end provider support.
Validate report_finding details before extraction and normalize findings before task rendering to avoid crashes when tool errors emit empty details.
Fixes#173
After plan approval, handleClearCommand() creates a new session with a
fresh artifacts directory. The local:// URL is session-scoped, so the
renamed plan file in the old session local dir was unreachable from
the execution session.
Write the plan content to the new session local dir after the session
switch so the titled plan artifact resolves correctly.
Clarify inclusive end semantics in the hashline prompt with bad/good examples.
Add range-replace duplicate-boundary auto-correction with a safety guard so correction only triggers for off-by-one ends.
Expand hashline heuristics tests to cover brace/); correction, blank-line guard, and boundary-already-in-range false-positive prevention.
Support Perplexity web search via session cookies extracted from
the desktop app (Electron/AppImage). Accepts either a raw cookie
string or a bare session token value.
Auth resolution order: PERPLEXITY_COOKIES > agent.db OAuth > PERPLEXITY_API_KEY
The #getGitStatus() method fires an async `git status --porcelain` but
only updates the TTL timestamp after the async completes. In large repos
where git status takes seconds, every TUI render cycle during that window
spawns a new process — cascading into hundreds of concurrent git processes,
exhausting file descriptors and creating .git/index.lock contention.
Add a #gitStatusInFlight boolean guard that ensures exactly one git status
process runs at a time. Use a finally block to unconditionally reset the
guard and refresh the TTL to completion time.
Request contents.summary from Exa /search API and combine up to
3 non-empty summaries into SearchResponse.answer, replacing the
previous 'No answer text returned' output.
Changes:
- Add contents.summary to every Exa search request body
- Add summary field to ExaSearchResult interface
- Export synthesizeAnswer() to build combined answer from summaries
- Export buildExaRequestBody() for testability
- Export normalizeSearchType() (was private)
- Prefer summary over text/highlights for snippet field (falsy check)
- Only synthesize answer from results that have a URL (mirrors sources filter)
Tests: 37 new tests covering normalizeSearchType, buildExaRequestBody,
synthesizeAnswer (pure unit tests) and searchExa (mocked fetch)
The disabledServers mechanism introduced in 4bfa3b0c (Merge branch
'pr-82', 2026-02-16) was defeated by a level guard added after merge:
servers with _source.level === "user" were exempt from the disabled
check. This meant servers discovered from ~/.claude.json (which the
Claude provider tags as level "user") were never actually filtered out,
even when present in disabledServers.
Remove the level guard so disabledServers applies unconditionally. This
is safe because the /mcp disable command already uses updateMCPServer
(setting enabled: false inline) for servers defined in omp own configs;
the disabledServers path only fires for third-party discovered servers.
Also fix the /mcp list display to cross-filter discovered servers
against the disabled list, preventing a server from appearing both as
"connected" and "disabled" when the MCP manager has stale state.
The submit_result tool marked 'data' as optional (Type.Optional) in its
JSON schema, but runtime validation requires it when status is 'success'.
Models that follow the schema literally omit the data parameter, causing
every call to fail with 'data is required when status is success (got
null/undefined)'. The agent then retries indefinitely.
Change data from Type.Optional(dataSchema) to dataSchema so the schema
matches runtime requirements. Models are now forced to include the data
object when calling submit_result with success, resolving the loop.
For aborted status, models can pass data: {}; validation is skipped.
Co-authored-by: Cursor <cursoragent@cursor.com>
- Added a shared CHANGELOG_CATEGORIES constant and reused it across changelog tooling.
- Validated changelog entry and deletion payloads to require arrays and flag invalid categories.
- Exported the changelog tool schema for generator validation and referenced it during parsing.
Co-authored-by: Can Bölük <can1357@users.noreply.github.com>
- Guard Container.render() to ensure width >= 1 for ALL components
- Guard TwoColumnBody.rightWidth with Math.max(0, ...)
- Validate description type before passing to native function
- Fall back to truncation when width <= 2
Fixes crash when opening extensions in narrow terminals.
The Exa MCP endpoint at https://mcp.exa.ai/mcp is publicly accessible
without an API key. Previously, all Exa tools bailed early with
'EXA_API_KEY not found' when no key was set, even though the basic
tools (web_search_exa, company_research_exa, get_code_context_exa)
work fine unauthenticated.
Changes:
- fetchExaTools/callExaTool: accept null apiKey, omit exaApiKey param
- MCPWrappedTool.execute: only require key for Websets tools
- createExaTool (factory): remove early-return guard on missing key
Fixes#151
The non-interactive environment variables (pager, editor, and prompt
suppression) were only applied in the PTY (interactive) bash path.
The non-PTY executeBash path had no such defaults, so commands could
block on pagers or credential prompts.
- Extract NO_PAGER_ENV from bash-interactive.ts into shared
non-interactive-env.ts module (renamed to NON_INTERACTIVE_ENV)
- Apply it in executeBash (non-PTY) with user overrides taking
precedence
- Fix ordering in interactive PTY path so user env overrides
the defaults (was reversed)
- Add test verifying defaults are applied alongside user env
- Standardized prompt format, removing XML-like tags.
- Emphasized RFC 2119 keywords using bolding across all prompts.
- Rewrote the core system prompt with enhanced structure and directives.
can1357