github can1357/oh-my-pi v18.14.0
on GitHub

3 hours ago

@oh-my-pi/pi-agent-core

Breaking Changes

  • Removed the Bun event-loop workaround utilities from @oh-my-pi/pi-agent-core/utils/yield; agent loops no longer insert cooperative sleeps while streaming or between tool calls.

Added

  • Added AgentTool.approvalRestrictionPaths so tools can report the paths they read, allowing restricted-folder approvals to cover those paths without remembering approval separately for each path.
  • Added Agent.logContext, which adds custom context fields to every log record written during an agent run.

@oh-my-pi/pi-ai

Breaking Changes

  • Moved Anthropic header and endpoint helpers, along with Codex session/transport inspection and OpenAI Responses replay/storage helpers, to dedicated provider modules. Update imports from the previous locations.
  • Updated auth-gateway model resolvers to receive ModelResolveOptions, including the route's native API wire and the scoped client's accepted grants. Scoped client IDs no longer fall back to scanning all models; resolvers now select the best permitted candidate.

Added

  • Added mTLS support for Anthropic connections through the CLAUDE_CODE_CLIENT_CERT and CLAUDE_CODE_CLIENT_KEY environment variables.

Changed

  • Provider implementations are now loaded on demand, reducing startup and memory overhead while preserving existing stream timeouts, cancellation, and host overrides.
  • Account policy priority now takes precedence when usage information is temporarily unavailable, keeping new sessions on the preferred account.
  • Improved provider-specific credential usage lookups so they no longer wait for unrelated providers, while preserving cache freshness, cancellation isolation, and complete aggregate reports. The corresponding server-side latency improvement requires an updated auth broker.
  • Credential usage is now prefetched during session setup without selecting, pinning, or advancing the OAuth account rotation.

Fixed

  • Fixed Mistral thinking-message replay requests that could be rejected by LiteLLM with a 422 extra_forbidden error.
  • Fixed native Judge API requests missing the omp/<version> User-Agent, so omp usage is correctly identified in upstream statistics.
  • Fixed Codex capacity-error retries over WebSocket and SSE ignoring provider-supplied Retry-After delays and retrying too early.
  • Fixed cancelled OAuth refresh requests aborting the entire session, and ensured requests joining an existing refresh can still use the refreshed credential when the original request is cancelled.
  • Fixed Cursor models losing omp system-prompt context, including context files, skills, and rules, on subsequent turns or when earlier messages precede the active turn.
  • Fixed short-lived commands that refresh OAuth credentials, such as omp token --force-refresh, remaining open for several seconds after completing.
  • Fixed TypeSafe authentication when TYPESAFE_BASE_URL is configured by validating keys against the correct /v1/models endpoint.
  • Fixed forced tool calls on non-reasoning models causing an unhandled reasoning-effort error.
  • Fixed interrupted turns reporting a transport error instead of the caller-provided cancellation reason.
  • Fixed long-running pi-native turns being terminated before their first event by fetch or proxy timeouts by supporting keepalives and the configured first-event timeout.
  • Fixed pi-native clients crashing when the auth gateway reports an upstream failure during streaming.
  • Fixed Cursor turns failing after a mid-turn stream interruption when an automatic retry receives a zero-token provider error; retries now continue on the rotated conversation.
  • Corrected auth-gateway debug logs so they no longer claim a retry occurred when a usage-limit error was returned without retrying.

@oh-my-pi/pi-catalog

Breaking Changes

  • Removed the raw @oh-my-pi/pi-catalog/models.json export; use the bundled model registry APIs instead.

Added

  • Added nextBuildPriceChange(cost, now) to identify when a model's dated pricing card may change, allowing callers to safely reuse earlier builds until that time.
  • Added the selector-revision-suffix catalog axis for Google Vertex, so google-vertex/claude-opus-4-8 resolves exactly to claude-opus-4-8@default.

Changed

  • Reduced the bundled model catalog size and loading overhead.
  • Updated role defaults so Factory Droid image models are preferred for the image role, allowing generate_image to work with only a Factory login.
  • Updated smol and slow role defaults to use the currently bundled Cerebras and OpenAI Codex model identifiers.

Fixed

  • Fixed Mistral reasoning models so streamed thinking appears in the transcript and is preserved for subsequent turns.
  • Fixed later-turn HTTP 422 errors in Mistral-hosted DeepSeek conversations when assistant reasoning history is included.
  • Fixed DeepSeek V4.1 Flash image attachments being dropped for publisher-namespaced model IDs such as ds/deepseek-flash.

@oh-my-pi/pi-coding-agent

Breaking Changes

  • Custom tool loaders now resolve the injected host SDK themselves; constructors take cwd as their first argument instead of a pi namespace.
  • WebMCP now uses Chrome's native WebMCP support through WebMcpController, which accepts a Puppeteer Page. The former page-hook installation helpers, bridge constant, and page-hook types have been removed; Tern tabs no longer support WebMCP.
  • Provider-qualified model selectors in --model, settings, roles, and task agents must identify an exact model ID or alias. Near matches now fail with a helpful error and suggestions instead of silently selecting a sibling model. Bare selectors such as opus retain fuzzy matching, and the Google Vertex Claude alias remains supported. Older migrated web fallback chains replace the retired Codex entry with openai-codex/gpt-5.6-sol.

Added

  • Agents can send an interim response without ending the current task by writing Markdown to out:// (or using > out:// in bash), with streaming output up to 64 KiB. This is enabled by default in the interactive TUI and can be configured with out.enabled; it is disabled by default in print, RPC, and ACP sessions.
  • Added computer-tool approval modes: tools.approval["computer.control"] can allow unattended foreground control on dedicated machines or deny the request without prompting. The approval mode does not itself grant access.
  • Added optional task-file input (task.file=true), allowing subagent instructions to be loaded from an absolute UTF-8 file path or local:// URL with the submitted path and SHA-256 recorded (#17133 by @iliaal).
  • Sessions can be pinned or unpinned from the resume picker with Alt+K and Alt+Shift+K, or the remappable app.session.pin and app.session.unpin actions. Pinned sessions move to the top and take priority in search results (#9837 by @zhang17-24).
  • Added PI_LINK_SCHEME and optional PI_LINK_HOST, allowing file links from SSH or tmux sessions to open through a desktop protocol handler (#15855 by @KunoLu).
  • Extensions and RPC clients can start live voice sessions using their own WebRTC peer, including phone-browser clients, through ctx.startLive({ sdpOffer }) or live_connect (#17013 by @materemias).
  • Extensions can customize assistant-text rendering and rerender individual replies through the opt-in extensionRenderers.assistantText setting (#16238 by @materemias), and can register interactive link schemes with extension callbacks and renderer fallbacks (#16239 by @materemias).
  • /debug reports now include session-logs.txt containing all log records written by the reporting session and its subagents for the current day.
  • Added prewalk.from and prewalk.into settings to choose the planning model and handoff target (default @default → @smol); /prewalk [into] and /prewalk restart [from] [into] override them for one command (#17373 by @H4vC).

Changed

  • Improved startup responsiveness by deferring nonessential UI, accessibility, browser, provider, slash-command, and interactive-composer loading until needed, and by overlapping independent session preparation work.
  • OpenAI live steering is now disabled by default for GPT models. Typed messages wait for the next tool boundary unless providers.openaiLiveSteering is enabled, avoiding unsupported in-flight message errors.
  • HTML exports now match the /stats visual design with a higher-contrast light theme, timeline and full-trace views, and navigation for nested subagents and advisors. Packaged viewers use integrity-pinned shared assets and require network access; source checkouts continue to embed the viewer.
  • tui.autoGraph now defaults to auto: charts are refined by the judge for supported native TypeSafe judge models and otherwise use the built-in estimate.
  • Filesystem changes are detected automatically, so manual scan-cache invalidation after edits is no longer needed.
  • Model refreshes use substantially less CPU by reusing unchanged models and avoiding repeated model-cache database opens.
  • Overlapping sloppy-edit operations are now applied sequentially, with each operation matched against the result of the preceding operation.
  • PI_INTENT_PROMPT and PI_INTENT_DESC can now customize the system guidance and tool-schema description for the i intent field.
  • Printed JSON values in eval, including large Python dictionaries and lists, are displayed as collapsible JSON trees.
  • Codex and Claude rate-limit reset warnings now appear once per reset across conversations and windows, only during the final six hours; the status line continues to show the countdown.
  • The . and c continue shortcuts now simply answer an unanswered last message instead of adding a hidden continue directive.
  • With bash.gitGuard enabled, checkout, switch, restore, reset, and stash operations run safely in linked worktrees, while discarded uncommitted changes are preserved for recovery.
  • Sessions with computer use enabled receive the desktop guide before the first computer call.
  • The Tern interface now presents warnings, errors, update notices, file reads, and completed background work consistently with chat messages.
  • Muted live voice sessions remain visibly muted while the voice model starts work.

Fixed

  • Fixed live configuration reloads missing edits immediately after a symlink switches to another file in the same directory.
  • Fixed auth-gateway model routing after catalog refreshes so bare model IDs select the highest-priority credentialed provider, with wire-format compatibility used as a tie-breaker.
  • Fixed chart recommendations that incorrectly inferred whether higher or lower values were better from row or column names; smart charting now uses the judge and avoids charting unsuitable numeric text such as page or step numbers.
  • Fixed binary data flowing through background Bash pipelines being displayed as corrupted text or counted as huge line output.
  • Fixed browser-tab closing delays when reattaching after a worker exits.
  • Fixed startup warnings being obscured by the Tern interface; they now appear as normal in-app warnings.
  • Fixed todo init clearing existing flat items when list was empty.
  • Fixed legacy extensions using detached event handlers failing to load.
  • Fixed MCP tools connected during vibe mode being activated prematurely and exceeding provider tool limits.
  • Fixed Bash services rejecting port 0 when a log readiness condition was also provided.
  • Fixed context usage and automatic compaction overstating token usage for inline tool descriptors.
  • Fixed terminal titles not being restored after exit on terminals without title-stack support, including Windows consoles.
  • Fixed cached approval directory grants that did not cover every target in a request.
  • Fixed text typed during startup being delayed until startup completed, including resumed drafts.
  • Fixed pull-request metadata and diffs remaining stale after a successful git push.
  • Fixed live voice playback clicks when audio stalls mid-word.
  • Fixed read rejecting artifact URLs from truncation hints under cached approvals or restricted folders.
  • Fixed peer task acknowledgements being mistaken for final answers, so wait continues until the peer replies.
  • Failed Hindsight memory recalls and retention operations now produce visible warnings instead of silently continuing without memory.
  • Fixed long conflict-path handling, conflict-content selectors, and stale conflict IDs after aborted merges or external file changes.
  • Fixed cancelled MCP OAuth flows and authentication lookups from crashing the session.
  • Fixed progress HUD entries for idle local-model workers that were no longer used.
  • Improved LSP diagnostics reliability across cold starts, push-diagnostics servers, save-only servers, retries, multi-server and multi-file edits, and reopened files; incomplete results are no longer reported as clean, and repeated clangd checks are faster.
  • Fixed live job widgets showing incidental text from files, web pages, or search results instead of the actual job status.
  • Fixed repeatedly failing advisor batches being retried indefinitely; they are now dropped after three failures.
  • Fixed thinking overrides on prefix-bound Claude models from losing their provider prefix behavior during compaction and retries.
  • Fixed failed large-value JSON queries from displaying the rejected value instead of a useful error.
  • Fixed F5 retry reporting no retry when a background result arrived after cancellation or interruption.
  • Fixed WebMCP invocation errors being empty and timed-out page invocations continuing to run.
  • Fixed HTTP 413 errors for Factory Droid GPT and Grok after computer-use sessions by removing obsolete screenshots from requests.
  • Fixed unavailable configured default or dictation models being silently replaced; omp now warns and identifies the fallback model.
  • Fixed LSP references for files in sibling Git worktrees by adding those checkouts to the active workspace.
  • Improved macOS computer-use behavior for applications without windows by opening the app and inspecting its returned window in one operation.

@oh-my-pi/pi-natives

Breaking Changes

  • Removed the invalidateFsScanCache API and the associated filesystem scan cache environment variables (FS_SCAN_CACHE_TTL_MS, FS_SCAN_EMPTY_RECHECK_MS, FS_SCAN_CACHE_MAX_ENTRIES, and FS_SCAN_CACHE_MAX_BYTES); filesystem listing caches are now managed automatically.

Added

  • Added a built-in rsync implementation to the bash tool, compatible with rsync 3.5.1 and protocol versions 29–33 across Linux, macOS, and Windows. It supports local copies, SSH remote shells via -e, and rsync:// daemons without requiring an installed binary; daemon serving is not supported, and PI_DISABLE_UUTILS_DESTRUCTIVE continues to use the system rsync.
  • Added built-in ast-grep (also available as sg) support to the bash tool for structural code search and rewriting, including pattern matching, rewrites, context, headings, and JSON output.

Changed

  • Improved native filesystem search performance, including faster globbing and more efficient grep, fuzzy finding, workspace listing, and structural search/edit operations, while preserving deterministic result ordering and streaming behavior.
  • Native search and the bash tool's built-in rg, grep, fd, find, and ast-grep now stop filesystem work promptly when cancelled or timed out, including searches over JavaScript-provided filesystems and searches interrupted while reading large files.
  • The bash tool's built-in rg and recursive grep now search files in parallel while retaining path-ordered output; grep can also stop promptly after reaching maxCount.
  • Enabled safe reuse of unchanged directory listings when cache: true, with timestamp validation to prevent stale results.

Fixed

  • Fixed cancelled native directory searches leaving filesystem-provider requests running during startup.
  • Fixed built-in bash commands such as cat, grep, head, wc, read, and mapfile hanging at the end of FIFO input on macOS; read -t now returns at EOF.
  • Fixed ANSI text wrapping for styled underlines, background colours, and colon-form RGB colours.
  • Fixed clicking in native audio playback when the queue runs dry mid-waveform by smoothly fading output to silence and back.

@oh-my-pi/omp-stats

Changed

  • Improved omp stats session synchronization performance by approximately 2–5× on large histories. The first sync after upgrading may take up to a minute while the stats database is optimized.

Fixed

  • Fixed /stats and omp stats failing when another dashboard was syncing sessions or the configured port was already in use; busy dashboards are now reused, and a free port is selected when necessary.

@oh-my-pi/pi-tui

Breaking Changes

  • Moved PINNED_HUD_TOGGLE_ID to the coding agent's task output manager so reserving subagent IDs no longer loads the interactive editor.
  • Updated StatusLineHost.resetExpiryNotice to receive the session and manage warning deduplication; custom status-line components must use the new signature.
  • Removed ToolExecutionHandle.setToolActivityVisible; tool-activity visibility is now controlled by transcript components, while reply blocks remain visible by default.
  • Removed the @oh-my-pi/pi-tui/theme/defaults export. Bundled palettes are now available through getBuiltinThemes() from theme/defaults.json.

Fixed

  • Fixed table-chart direction detection so measures are not incorrectly marked as better based on header words; single-measure tables are now sent for judging when local charting is appropriate.
  • Fixed LSP diagnostics that incorrectly showed “No issues” before a language server had reported; such files are now shown as unverified. (#17293 by @JustasMasiulis)
  • Fixed Esc-Esc rewind in Tern selecting interrupted turns with no visible output instead of the last displayed turn.
  • Fixed streamed code blocks losing lines that scrolled off screen before the closing fence arrived; completed lines now remain available in scrollback. (#17333 by @gzkGit777)

@oh-my-pi/pi-utils

Added

  • Added logger.withContext(fields, fn) to attach contextual fields to all records produced during a function and its asynchronous work, plus logger.withoutContext(fn) for work that should not inherit caller context.

Fixed

  • Fixed untilAborted so promises passed to an already-aborted signal are still observed, preventing later rejections from being reported as unhandled.

@oh-my-pi/pi-wire

Added

  • Added the TSP json kind for displaying raw JSON values as interactive, expandable trees in the terminal.
  • Added TspEditorProps.typeahead support so programs that cannot yet read input can preserve typed text, display it at the caret, and submit it when input becomes available.

What's Changed

  • fix(coding-agent): allow artifact reads through approval checks by @iliaal in #17282
  • fix(catalog): keep DeepSeek V4.1 Flash images for namespaced ids by @ZanDev32 in #16470
  • fix(coding-agent): warn when a Hindsight memory recall fails by @enieuwy in #17278
  • fix(ai): kept cancelled OAuth refresh rejections observed by @H4vC in #17290
  • feat(session-picker): add pin and unpin actions by @zhang17-24 in #9837
  • feat(tui): support host-qualified file links in remote sessions by @KunoLu in #15855
  • feat(coding-agent): add opt-in assistant text renderers for extensions by @materemias in #16238
  • feat(coding-agent): run live voice sessions on a caller-owned WebRTC peer by @materemias in #17013
  • fix(coding-agent): keep idle tiny-worker unloads out of the progress HUD by @H4vC in #17305
  • fix(lsp): stopped diagnostics stalling, timing out, or reporting unanswered files as clean by @JustasMasiulis in #17293
  • fix(approval): only offer directory grants that cover all targets by @iliaal in #17309
  • feat(task): dispatch subagent instructions from UTF-8 files by @iliaal in #17133
  • fix(conflict): follow live conflict blocks across lists, ranges and the registry by @alphastorm in #17177
  • fix(github): refresh pull request diffs after git push by @alphastorm in #17183
  • fix(wait): keep a peer request owed until its reply arrives by @alphastorm in #17176
  • feat(coding-agent): route clicks on extension-owned links to their extension by @materemias in #16239
  • fix(coding-agent): clear terminal title on exit for terminals without title stack by @asuffield in #17321
  • fix(ai): carried omp system prompt on cursor history-bearing roots by @robfeldmann in #17325
  • fix(ai): rank account policy priority ahead of a missing usage report by @will-bogusz in #15124
  • fix(lsp): add another checkout to the language server workspace before opening its files by @will-bogusz in #16271
  • fix(coding-agent): fail provider-qualified model selectors that name no exact id instead of running a sibling by @will-bogusz in #16297
  • fix(coding-agent): drop computer screenshots the Factory image budget evicts by @will-bogusz in #16564
  • fix(ai): stop logging a usage-limit retry the gateway does not make by @will-bogusz in #16579
  • feat(coding-agent): tag each log record with the session that wrote it by @will-bogusz in #16587
  • fix(browser): use Chrome's native WebMCP on demand instead of injecting a page hook by @will-bogusz in #16756
  • feat(coding-agent): let tools.approval grant computer foreground control on unattended machines by @will-bogusz in #16917
  • feat(coding-agent): start desktop tasks without a documentation read by @will-bogusz in #16918
  • fix(coding-agent): avoid failed window lookups when starting macOS desktop tasks by @will-bogusz in #16925
  • feat(pi-shell): keep the git guard's refusals to checkouts other sessions share by @will-bogusz in #17349
  • feat(catalog): fall back to Factory Droid image models in the image role by @will-bogusz in #17351
  • fix(catalog): prevented mistral deepseek reasoning replay by @roboomp in #17366
  • fix(ai): report the caller's abort reason instead of the transport's abort text by @MaisonnatM in #17216
  • fix(coding-agent): preserved large json query error reasons by @roboomp in #17270
  • fix(coding-agent): retried past trailing async-result after an interrupt by @roboomp in #17272
  • fix(coding-agent): kept catalog prefix binding under thinking overrides by @roboomp in #17279
  • fix(ai): handle disabled reasoning without supported efforts by @wyyufeng in #17283
  • fix(advisor): keep the three strikes across a context-maintenance re-prime by @joshuaswarren in #17291
  • fix(voice): faded playback edges around underruns by @roboomp in #17299
  • fix(ai): Apply inference user agent to native judge requests by @BizerNotNull in #17311
  • test(ai): remove racing timeout fallback by @DROOdotFOO in #17315
  • fix(ai): retry a rotated Cursor conversation after a 400 by @morali in #17319
  • fix(context): counted pruned tool specs when descriptions are inlined by @roboomp in #17332
  • fix(tui): retire a streamed open fence's finished lines by @gzkGit777 in #17333
  • fix(coding-agent): bind EventBus prototype methods for detached pi.events calls by @Zynoo71 in #17338
  • fix(coding-agent): kept mcp refreshes off the vibe director toolset by @roboomp in #17343
  • fix(coding-agent): kept flat todo init items when list is empty by @roboomp in #17348
  • fix(ai): preserve Retry-After headers on Codex stream errors by @iFwu in #17362
  • fix(ai): kept silent pi-native streams alive past bun's fetch timeout by @roboomp in #17364
  • fix(ai): captured and replayed mistral thinking content parts by @roboomp in #17377
  • feat(prewalk): add prewalk.from/prewalk.into and /prewalk selector arguments by @H4vC in #17373

New Contributors

Full Changelog: v18.13.3...v18.14.0

Check out latest releases or
releases around can1357/oh-my-pi v18.14.0

Don't miss a new oh-my-pi release

NewReleases is sending notifications on new releases.

Get notifications