hummin-cli 1.0.6 → 1.1.1

package/CHANGELOG.md

@@ -1,5 +1,61 @@
 # Changelog
 
+## [1.1.1] - 2026-09-21
+
+### Added
+
+- Todo plan widget above the editor: live checklist with header counts (`TODOs 2/6`), the in-progress step plus up to three pending ones, and a collapsed `+N pending, M completed` summary line; hidden once every step is completed.
+- Clear-session choice: `/clear` (and `app.session.new`) with work in flight (active response, bash, queued messages, background tasks/monitors) asks whether to continue running and migrate to the new session, or abort all; queued messages are re-delivered into the new session and background jobs are adopted by its extensions (rebindable output/completion hooks on background jobs).
+- `ask_user` transcript rendering: the call row shows the question (expanded shows numbered choices) and the result shows the picked answer or `cancelled`, instead of a raw JSON dump.
+- Categorized settings UI: /settings now groups its entries behind a tab bar (General, Agent, Models, Memory, Fleet, Editor, Terminal, Network, Shell, Privacy); `tab` or clicking a tab switches categories, and search still filters the active list.
+- Settings previously missing from /settings are now configurable: global default thinking level, per-provider list (providers.showAll), retry on failure, memory (enabled, mode, vault directory, provider, model), local fleet instances and fleet auto-start, editor mode (vim), external editor command, WebSocket connect timeout, HTTP proxy, shell path, shell command prefix, npm command, and analytics opt-in. Free-text fields open an input submenu (Enter saves, Esc cancels).
+- Upstream sync: bug reporting (`/bug` with model-generated summary), prompt cache warming with off/streaming/idle modes (setting under Agent), progressive session picker loading, faster recent session discovery, compiled Node CLI module caching, deferred extension loader dependencies, and an experimental micro agent.
+- Per-model image resize profiles: `inputLimits.images.resize` in `models.json` applies to file attachments, image reads, and tool-result images ([#9631](https://github.com/earendil-works/pi/issues/9631)).
+- Upstream sync: crash diagnostics now hint at loaded extensions that appear in the stack trace.
+
+### Changed
+
+- Tool rows drop the padded 12-char label column: each label is followed by a two-space gap, so rows like `Bash  <command>` no longer carry a wide empty stretch.
+- Footer right-aligned clusters (provider + model + thinking level, and the ctx meter) no longer start with a dim `·` separator with nothing to their left.
+- Message timestamps and skill commands moved to the General category; image settings moved to Terminal; transport and HTTP idle timeout moved to Network.
+
+### Fixed
+
+- Upstream sync: restore OSC 52 clipboard fallback for headless sessions; ignore stale tool image conversions ([#8743](https://github.com/earendil-works/pi/issues/8743)); close compaction cancellation races; compact oversized trailing tool results; suppress repeated Anthropic thinking drop notices; suppress bug hints for expected failures; preserve multiline bug descriptions; await the terminal remote prompt event; rebuild expired idle prompt-cache warming caches only when the timer or extension decision is not delayed.
+- TUI (upstream sync): preserve fullscreen images in WezTerm; rank skill autocomplete by bare name ([#9120](https://github.com/earendil-works/pi/issues/9120)); handle CJK punctuation in file autocomplete ([#9746](https://github.com/earendil-works/pi/issues/9746)); improve LaTeX compatibility and layouts.
+
+## [1.1.0] - 2026-09-19
+
+### Added
+
+- Agent teamwork: inter-agent messaging across terminals and projects - UDS broker with offline inbox and replay (`agent_send`, `agent_inbox`, `/agents`, `/agent-name`), and a shared cross-session task board (`task_board`, `/team`).
+- Cron scheduler: schedule detached `hummin -p` wake-ups per project (`cron_create`/`cron_list`/`cron_delete`, `/cron`); single-scheduler lock, one queued wake per entry.
+- MCP client: stdio JSON-RPC transport for `mcpServers` in settings (project entries trust-gated); tools register as `mcp_<server>_<tool>`; `/mcp` status panel with restart.
+- LSP client for TypeScript via `typescript-language-server`: `lsp_diagnostics`, `lsp_definition`, `lsp_references`, `lsp_hover` (1-based coordinates); auto-activates on tsconfig/jsconfig projects; `/lsp` status panel.
+- `ask_user` tool: structured multiple-choice questions through the TUI with optional free text; sequential asks no longer hang; fleet-style bordered question panel.
+- Bash sandboxing: macOS seatbelt / Linux bubblewrap workspace mode (writes limited to the working directory and temp, secrets unreadable, optional network deny) with real capability probing; `/sandbox` toggle; spawned task children inherit the mode.
+- Declarative hooks: `hooks.json` (global and trust-gated project) runs shell commands on `tool_call`, `tool_result`, `agent_start`, and `agent_end`; `tool_call` hooks can block with a reason; `/hooks` table and reload.
+- BashGuard advisory matrix: in-band warnings for destructive commands, out-of-tree `sed -i`, and outside-cwd writes; opt-in blocking and read-only deny mode (`/bashguard`).
+- Usage visibility: `/context` (system prompt, tools, conversation, cache-hit ratio, compaction trigger; est vs exact labeled) and `/cost` (per-model totals, served-locally vs cloud split).
+- Footer navigator: `alt+down` (or `down` on an empty editor) opens a selectable list of Background tasks, TODOs, Agents, and Fleet, each opening its panel; native styled TODOs footer segment (`TODOs x/y · current item`).
+- Background panel: `ctrl+b` or `/background` opens a live panel with view-tail/cancel actions, and `ctrl+x` stops the selected running task/monitor or clears a finished one; the footer status reports counts by kind ("2 tasks, 1 monitor").
+- Fleet: selecting an offline fleet model in `/model` offers to start its server first (`fleet.autoStart` skips the prompt), with readiness-gated selection via the shared fleet-actions library.
+
+### Changed
+
+- Background tasks and monitors survive agent interrupts: the turn's abort signal is no longer forwarded to background jobs (foreground tasks remain abortable).
+- Task rows show what is running (`Task  "<prompt>" · model`); background task completions and monitor notices render as collapsible one-line rows instead of always-expanded blocks.
+- The plan checklist is labeled "TODOs" everywhere and an always-current footer segment shows progress and the current item.
+- Custom `statusline.left`/`statusline.right` segments and the default footer right side use dim middot separators; vim modal editing is available via `editorMode: "vim"` (`HUMMIN_VIM=1`).
+
+### Fixed
+
+- Collapsed bash tool rows are a single line again: the command is collapsed to one line and truncated to the terminal width, with the duration/status suffix preserved; the full multi-line command and output remain available expanded.
+- Fixed the sandbox extension hard-blocking bash in the default (`off`) mode; plain execution is used unless workspace mode is active, with an actionable `[Sandbox]` message when a mechanism is unavailable.
+- Fixed the agents broker crashing extension startup when two sessions raced for the broker socket; the loser now connects as a client.
+- Fixed the seatbelt capability probe using `/bin/true`, which does not exist on macOS.
+- Fixed background task completion notices being delivered twice.
+
 ## [1.0.6] - 2026-09-18
 
 ### Changed
@@ -38,9 +94,11 @@
 ### Changed
 
 - Exact session-ID session lookups skip transcript scans ([#9601](https://github.com/earendil-works/pi/issues/9601)).
+- Formatted Bash and PowerShell tool durations of at least one minute as minutes and seconds, with hours when needed ([#9628](https://github.com/earendil-works/pi/issues/9628)).
 
 ### Fixed
 
+- Fixed mid-run threshold compaction silently skipping oversized trailing tool results ([#9740](https://github.com/earendil-works/pi/issues/9740)).
 - Fixed signal-terminated local shell commands being reported as successful with partial output ([#9577](https://github.com/earendil-works/pi/issues/9577) by [@BrendanJMurphy](https://github.com/BrendanJMurphy)).
 - Fixed local clipboard failures reporting success when the terminal ignored the fallback OSC 52 write, and added platform-specific setup guidance when no clipboard backend works ([#9618](https://github.com/earendil-works/pi/issues/9618)).
 - Fixed `before_agent_start` handlers returning `systemPrompt` (and `forceSystemPrompt`) on models with mid-conversation system messages: the forced prompt is now sent as the provider's leading system prompt instead of being appended as a section patch after the original prompt.
@@ -104,6 +162,7 @@ First hummin release. Hummin is a thin-overlay fork of pi: zero deletions from u
 - Fixed premature missing-model errors after login by waiting for catalog discovery. Radius now defaults to `balanced`, falling back to the first available Radius model when needed.
 - Fixed fullscreen mode reserving a blank row for custom footers that render zero rows ([#8919](https://github.com/earendil-works/pi/issues/8919)).
 - Fixed extension tools without parameter schemas to be rejected during registration instead of breaking provider requests ([#9300](https://github.com/earendil-works/pi/issues/9300)).
+- Fixed loaded llama.cpp models with `enable_thinking` chat templates ignoring Pi's thinking level ([#9528](https://github.com/earendil-works/pi/issues/9528)).
 
 ## [0.85.1] - 2026-09-05
 

package/README.md

@@ -45,6 +45,16 @@ hummin -p "summarize this repo"   # oneshot mode
 
 GLM-5.3, GLM-5.3-Flash and GLM-5.3-highspeed ship in the `zai` provider catalog (1M-token context, reasoning variants mapped to `reasoning_effort`). Pick models with `/model`; set a persistent default with the picker's "set as default" action.
 
+## Features
+
+- **Agent teamwork**: sessions message each other across terminals and projects (`agent_send`, `/agents`), share a task board (`/team`), spawn bounded subagents (`task`, `/background` on `ctrl+b`), and run scheduled wake-ups (`/cron`).
+- **Local fleet**: one provider across all your OpenAI-compatible servers (colibri, llama.cpp, Ollama, ...), health-probed and startable from the model picker, with per-origin serialization and real context windows.
+- **Integrations**: MCP client (`mcpServers`), TypeScript LSP tools (diagnostics/definition/references/hover), declarative `hooks.json`, `ask_user` questions, DuckDuckGo search and SSRF-guarded fetch.
+- **Safety**: bash sandboxing (macOS seatbelt / Linux bubblewrap) with `[Sandbox]` in-band blocks, destructive-command advisories (`/bashguard`), loop + budget guardrails, project trust, file checkpoints with `/rewind`.
+- **Memory**: session distillation into lessons plus a self-curating Obsidian-compatible vault with BM25-ranked recall injected into every session.
+- **Visibility**: `/context` token breakdown with cache-hit ratio, `/cost` with local-served vs cloud split, `/status` + `/doctor` dashboards, two-row statusline (user-definable segments).
+- **UX**: vim editing mode, grouped command palette, custom statusline segments, fullscreen transcript search, themeable, remote browser control over loopback.
+
 ## Local inference (colibri, llama.cpp, ...)
 
 The bundled `hummin-colibri` extension registers ONE provider (`colibri`) that exposes every model found on your local OpenAI-compatible servers. The engine behind each server does not matter - colibri (MoE streaming), llama.cpp, Ollama all work; the server's own `/v1/models` is the source of truth for model ids, and the first server in the list that serves a model wins (duplicates dedupe into one entry with fallback ordering).