Oh My Pi v18.13.3

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

Changed

• Updated Bun engine requirement to ≥1.4.4

Fixed

• Fixed chat-completions context estimates counting past thinking that the request drops (no reasoning replay), which triggered compaction far below the threshold (#15797, #17231 by @ghisguth)

@oh-my-pi/pi-ai

Changed

• Updated Bun engine requirement to ≥1.4.4

@oh-my-pi/browser-relay

Changed

• Updated Bun engine requirement to ≥1.4.4

@oh-my-pi/pi-catalog

Added

• Added DeepSeek V4.1 Flash 1017 to the Factory Droid provider for accounts with Factory's 1017 flag; turning thinking off on its Baseten route disables reasoning instead of falling back to low effort (#17234 by @skipper019)

Changed

• Made Kimi Code model discovery authoritative and cached per credential, added K3.1 Pro and K3.1 Pro Highspeed to the bundled Kimi Code catalog, and preserved live thinking metadata there (#17224 by @trukhinyuri)
• Updated Bun engine requirement to ≥1.4.4

Fixed

• Codex model refresh now lists models available to an entitled account even when other saved accounts return HTTP 402 (#17198).
• LiteLLM discovery now reads every /v2/model/info page before selecting the proxy's model roster, rather than caching only the first 50 entries; malformed or incomplete pagination falls back without caching a partial page (#17237 by @maykonmichel).
• Fixed hosted web search on OpenAI Responses and Codex Responses selecting GPT-5.6 Luna instead of GPT-6 Luna (#17202 by @mateo-bolanos).

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

Added

• Manual compactions (RPC compact, /compact, extension ctx.compact()) emit manual_compaction_start and manual_compaction_end session events, so RPC hosts can show compaction progress; docs/rpc.md documents these with the other progress events (retry_fallback_applied, ttsr_triggered, goal_updated) (#17247 by @Quelaan1)
• Added retry.rateLimitWalksChain (default false): with it on, a 429 rate limit after a turn with signed Claude thinking moves to the next model in retry.fallbackChains at once instead of retrying the throttled model until retry.maxRetries is spent (#17232 by @sjawhar)
• Sessions can be renamed from the resume picker: the new title persists as a user title without loading the session, and renames of live sessions are refused instead of forking the transcript (#17262 by @Twilight0)
• Added display.advisorCollapsedLines (Settings → Appearance → Display) to cap how many note lines a collapsed advisor card shows; the default keeps the full preview (#17228 by @bluecapone)
• omp config list --scope global|project lists only the keys that layer sets, with the values as written: path-scoped enabledModels, enabledProviders and disabledProviders entries ({ path, … }) stay unresolved, so a host can edit the global value without dropping them (#17246 by @Quelaan1)
• tools.xdevNotices: false hides the xd://: mounted … notices shown as MCP servers connect, without turning on startup.quiet (#17263 by @MaisonnatM)

Changed

• Updated Bun engine requirement to ≥1.4.4

Fixed

• A running job's glance keeps the whole a meter counts instead of the bare numerator: a widget whose headline restated 140 beside a 140/190 bar now draws 140/190 in the job card, the HUD pill and the status line (#17226 by @iacore)
• Fixed the managed-skills root guard rejecting a symlinked store outright; a root that resolves to a directory now accepts writes inside its target, while a broken link or a link to a non-directory is still refused (#17201 by @iacore)
• Fixed experimental context rollover repeatedly running during tool calls when the fixed prompt and tool schemas already exceed the recovery band (#17205)
• Fixed Ctrl+D during deferred extension startup falsely reporting an extension exit error and exiting 1 instead of closing cleanly (#17259).
• Resuming a session, or reviving a parked or restored subagent, after a prewalk or plan-yolo handoff continues on the model and effort the handoff switched to instead of falling back to the planning model (#17229 by @shgew)
• Fixed one interrupt silently stopping a session from ever waking on its own background job results again: an interrupt (Esc, the RPC abort command, an ACP cancel, a collab guest stop, an extension's ctx.abort()) that stops no run and no queued work no longer suppresses those wakes, and after an interrupt that did stop a run they come back once a later turn has run and ended. Esc on the "waiting on background jobs" row still ends that wait: its results land in context for your next prompt, as before (#17243 by @sjawhar)
• Fixed ! commands on zsh/fish waiting forever at interactive prompts (e.g. the Oh My Zsh update question on !code .); keystrokes now go to the running command while it runs, and Esc still cancels (#17210)
• Fixed auto-compaction firing early on OpenAI-compatible chat-completions models that don't replay past reasoning (#15797, #17231 by @ghisguth)
• Eval's tool.<name>(…) now accepts xd://<name> and the Claude Code mcp__<server>__<tool> spelling like direct tool calls, and reports registered but unavailable tools instead of "Unknown tool" (#17250)
• Updated the ask prompt to encourage assistants to show missing decision context before the dialog opens (#17242 by @jasonharrison)
• Custom LiteLLM providers now refresh cached model rosters after paginated /v2/model/info discovery was fixed; cached first-page-only lists no longer hide later-page models (#17237 by @maykonmichel).
• Fixed hosted search repeating the same request when the active model is also its search target but uses a cloned or refitted model object (#17202 by @mateo-bolanos).

…