Releases: manishiitg/agentworks
Releases · manishiitg/agentworks
Release list
1.25.133
Desktop release v1.25.133.
Install (macOS, Apple Silicon)
Install from Terminal with curl. Do not download the DMG in a browser: AgentWorks is not notarized by Apple, so macOS blocks a browser-downloaded copy ("is damaged and can't be opened"). A curl download is not quarantined.
curl -fL -o /tmp/AgentWorks.dmg https://github.com/manishiitg/coding-agent-loop/releases/download/v1.25.133/AgentWorks-1.25.133-arm64.dmg
hdiutil attach -nobrowse -quiet -mountpoint /tmp/agentworks-dmg /tmp/AgentWorks.dmg
rm -rf /Applications/AgentWorks.app && cp -R /tmp/agentworks-dmg/AgentWorks.app /Applications/
hdiutil detach -quiet /tmp/agentworks-dmg && rm /tmp/AgentWorks.dmg
open /Applications/AgentWorks.appAlready installed? The app updates itself from this release.
Changes since v1.25.132:
- agy: explain alpha terminal and final reply behavior (749e13d)
- agy: expose alpha provider in UI (9ae369a)
- agy: finish local app P0 and native tools UI (05b038d)
- agy: wire durable receipt watcher and document record contract (ef08c9b)
- docs: describe AGY structured turn proof and limits (2f7dcd5)
- agy-exp: builder agy provider wiring + setup guide (83a4c3a)
- desktop-release: cap the changelog at 100 commits with a compare link (e1ee644)
- desktop-release: previous release is the highest plain vX.Y.Z tag, not the repo's Latest (a27d5b9)
- Desktop DMG: green again; release notes say to install with curl (c971f70)
- Crew calls from non-owners run as their guest, not with the owner's access (e4f616f)
- PLAT-363: stop losing scheduled runs silently; Pulse sees and repairs lost runs (35d7623)
- Suggestions over MCP and the CLI: suggest_crew_change, suggest_workflow_change (2b4836c)
- Scheduler: Pulse fix runs no longer block the workflow's own schedules (9fccab4)
- Crew suggestions: other users ask the owner for changes; decisions API checks access (368e740)
- PLAT-280 diagnostic fires only for a granted-but-missing DB_PATH (663219d)
- code-authoring skill: native Begin Patch and multi-file patches; no splice edits (b002bb0)
- diff_patch_workspace_file: tool description names the Begin Patch and multi-file support (919c70c)
- diff_patch_workspace_file: multi-file "*** Begin Patch", all or nothing (a6ed4db)
- Begin Patch: prefix anchors, anchor-as-first-line, unique loose matches; more tests (7c83e1a)
- diff_patch_workspace_file accepts the Codex/Cursor "*** Begin Patch" format (9ea2899)
- PLAT-362 review follow-ups: token-file sweep, CDP owner visibility (99d68e9)
- Security (PLAT-362 D5): session routes act only on the caller's sessions (44c3ff1)
- PLAT-362: agents could act as another session or workflow — done and left (df4aa27)
- Terminal hover: plan usage for launch-only chats (Crew) from the CLI's files (eb70094)
- MCP/CLI function calls return at once too; skills say so (004d85a)
- Crew functions: call_function returns at once; wait_seconds is opt-in (3b129ce)
- Crew functions: an identical call while one is running joins it (48eb757)
- Pin working Codex CLI for SparkQuill and verify startup on deploy (7eaaf4e)
- Security (H1 review): shells, virtual scopes and execute routes bind the caller's session (b1a08db)
- Crew authoring: export_crew is owner-only (db62fc5)
- Crew authoring over the external MCP and agentworks CLI (2f69e80)
- Security (H1 follow-up): sign session tokens with an in-memory secret (4be8622)
- Active work: every row says what started it; Ctrl+K @Active lists every run (d46c8e1)
- Crew run tabs: show the Crew's name and the full title on hover (49d2f34)
- Security (H1): agent tool calls act only as the caller's own session (f48d158)
- Goal setup: one card with segmented progress; slim line after the goal (7385bab)
- Goal setup: shorter card, one line once the goal is set (d0c1c90)
- Goal setup: 'No goal needed' as a clear secondary choice (48b7f5b)
- Goal setup bar: a guided setup card instead of a row of pills (da96869)
- Coding watchdog: a main pane replaced by a running turn does not fail the session (5283a59)
- Builder prompt: a CLI's read-only native sandbox is not the session's access (f89e149)
- docs: coding CLI usage and limits (hover, watchdog, P0 plan-usage contract) (92126aa)
- Terminal icon hover: context fill, session tokens, cost and CLI details (0b2007d)
- Show coding CLI plan usage with reset times on terminal icon hover (2e94232)
- Coding watchdog: the CLI's own usage numbers outrank screen text (3f10b26)
- Expand reviewed Crew playbooks across four categories (5a22afe)
- Coding watchdog: the user's own words about a limit are not a limit wall (7f74cc3)
- Add Human actions panel and shared schedules overview (e2f0c85)
- Pulse review fixes: direct reviewer caller, no Crew work in scheduled turns, orphaned steps, dropped completions (6f697f2)
- Security: only a person in their own Builder chat can answer a decision (e1e179b)
- Security: Pulse platform tools act only for the calling session's workflow and owner (1738265)
- Add Product GTM and Support Crew playbook routes (fe21dfe)
- Goal setup starts with an optional playbook (899eb6e)
- Goal setup: add the Dashboard step (1858664)
- Automations start with an optional goal setup: goal, plan, metrics (ea73f1e)
- Crews: a deleted attachment no longer breaks the crews that attached it (af5693d)
- Pulse can read how a Crew handled this workflow's calls, and judges Crew vs message sequence (e6682ff)
- Crew: typing in chat no longer re-renders the workspace pane (3b8f212)
- Reports: typing in chat no longer re-runs the dashboard (b4a690b)
- Workshop: only an owner's session may write workflow.json (66c32ea)
- Pulse: the runtime starts due reviewers itself; reviewers record their own result (407e312)
- Enforce per-workflow access in the /api/wp proxy and workflow routes (0edf755)
- Restarted CLI sessions get the archive path, not a pasted transcript (eca917f)
- PLAT-361: scheduled runs and background agents stopped before their steps finished (b51acc6)
- Scheduled runs and background agents own the steps they start (0afb9f2)
- Fix agent startup panic: keep http.DefaultTransport an *http.Transport (f0db758)
- Require the workspace token on every route; route /api/wp through the agent (0b3b084)
- Balance Crew categories with Product GTM and Support roles (e5fead4)
- Group QA and Security under Engineering browse (91ea6b9)
- Add trial account to reviewed sales assist playbook (297220d)
- Add dispute Playbook and enforce Crew handoff schemas (b38fe4f)
- Scope Product Crew setup to verified decision route (d7e592d)
- Require source-record joins in order update playbook (ed1afad)
- Add generic order exception to reviewed update playbook (2a8c118)
- Correct launch library counts and Crew versions (a580cbd)
- Deepen Operations Crew setup evidence (ad8936a)
- Deepen Marketing and GTM Crew setup evidence (0768376)
- Deepen support and security Crew setup checks (6a039e2)
- Guide Builder through validated Crew handoffs (6ab9555)
- Add access exception to verified fix playbook (ca4d6e2)
- Deepen QA Crew setup evidence and acceptance checks (149011a)
- Promote engineering operations intelligence to Crew playbook (35deba7)
- Add vendor evaluation purchase decision playbook (0cd2397)
- Add released feature adoption decision playbook (635ca54)
- Add contract-aware renewal Crew playbook (46c33ab)
- Choose workspace landing view from saved preference and content (c62d4c4)
- Add signed deal to onboarding playbook and library quality gate (220802f)
- feat(panes): Ask AI lives in the walkthrough popup; refresh Crew Models and Files (7f43ce1)
- Add post-incident review and verified actions playbook (cce7a68)
- Add experiment execution and outcome Crew playbook (878ec25)
- Crews shared root, step 1: one crew-path resolver + Crew/ access gates (67ceec7)
- Add maturity-aware retention Crew and playbook handoff (5ed6638)
- Design: move Crews to a shared Crew/ root (aaaf618)
- Add signup funnel Crew and typed experiment playbook (db602db)
- Reports: crew owner's Dashboard can run scripts (relative project path) (c83a4d0)
- Crew MCPs: name apps that still need connecting (#205 E2) (50b3896)
- Improve provider cost explorer and attribution (5e5ff17)
- Add typed AI visibility Crew playbook (411fbc1)
- Add typed FinOps Crew savings playbook (0349b30)
- Verify SEO proposal and Workflow guide installation semantics (b12e242)
…and 2628 more. Full list: v1.25.132...v1.25.133
1.25.132
Desktop release v1.25.132.
Changes since v1.25.130:
- Remove Gmail from the interactive Bots modal (restores a broken test contract) (14eb0a5)
- Batch WhatsApp uploads, pin codex model, zoom-scroll fix, harder/animation prompts (fecdb44)
- Bump desktop version to 1.25.131 (7c5cc26)
- Answer marking goes on-demand, page anchors, WhatsApp send retry, Pulse trace, quick actions (069694c)
- suggest_actions reliability, drag/zoom controls for child mode, find_image, richer scenes (32132d1)
- Reap orphaned tmux sessions, own our tmux namespace, and show workspace size (c887e7e)
- Fix a real dangerous bug (shared .cursor/hooks race), plus tool-call visibility and chat cleanup (c0da6e1)
- Weave her genuine interests into tutoring, add refine-it quick actions, actually say her name (7ed12bf)
- Add a JSON-transport/no-steer test toggle, and fix the start-burst animation (c60ae90)
- child mode: stop silently deleting past scenes/stars, and use more of the space (0b32202)
- child mode: unblock steering, keep her place, clearer feedback, start animation (fca771a)
- Pulse rework, turn latency tracing, activity archiving, and the missing goal field (44fffae)
- workflow: coding-agent CLI steps always use structured JSON transport (59705b5)
- agent_go: migrate to mcpagent's single WithCodingAgentTransport option (19c59c7)
- guidance: drop doc-read gate, fix path convention, tighten Pulse/plan prompts (9eddc08)
- Show current Pi models in LLM UI (d21fa73)
- wire exa-search (free hosted MCP) into SparkQuill coding-agent sessions (aedc41c)
- SparkQuill desktop app: Electron shell, curl|bash installer, release workflow (e2e7b70)
- Workspace tab: the activity becomes the card, and the parent can act without digging (fcee09d)
- All files as its own icon, not a sub-toggle; progress report answers "how is she doing / what next" (293c6c2)
- prompts+skills: cut duplication, tighten, trust the model more (ecb3be4)
- fix: read_image degrades clearly on Pi CLI (no vision support); remove generate_image (aecca85)
- feat: child steer (replaces silent drop), steer-first everywhere, Gemini/Vertex image-gen fallback (557ff53)
- Add CLI workspace security configuration (2f64040)
- Update workflow profile tests for Opus 5 (127640e)
- remove parent_tools_test.go — tests not wanted in SparkQuill yet (7550eed)
- Expose older Claude Opus models for manual selection (e67e525)
- fix two session-continuity P1s: foreign-engine handles, and history loss on cold start (38cdf88)
- Use Opus 5 medium for Claude managed roles (dd89506)
- family-server: one canonical parent tool manifest across all parent surfaces (05feba8)
- fix: live tool-call streaming, cursor-cli streaming, robust auto-scroll, activity goal/give-to-child polish (6dc9d9b)
- remove reply_sanitize.go stopgap - the real fix landed upstream (283f0c4)
- deps: re-pin mcpagent + multi-llm-provider-go to latest shared-branch fixes (aa7aaf0)
- feat: chat-driven show_scene interactivity, activity goal field, model tier tweaks (6390a05)
- feat: activity-folder workspace redesign, encrypted secrets store, codex reply-leak stopgap (b9b46e3)
- Support scripted step route convergence (6faa540)
- docs: dev note on pinned private upstream deps + GOPRIVATE + re-pin flow (680b4e0)
- agent_go: remove Agy provider + pin mcpagent/multi-llm to shared branches (15636f4)
- feat: activity-scoped child sandbox (no copy), open_activity, coding/discover skills, teaching-style→guide_note (0ba7cb0)
- feat: WhatsApp voice notes (local STT + toggle), PDF-in-chat links, multi-account WhatsApp, streaming, steer (278b660)
- feat: refine Pulse workflows and notification UX (40e48b7)
- feat: WhatsApp media stays in inbox (text-only turns), current date/time in prompts (bc200c2)
- feat: dark/light theme, Uploaded Material tab, package handoff confirm + instructions on first turn (c4ce9ef)
- feat: faster child (haiku) + medium reasoning, HTML notify email, package-only handoffs (abb1b9d)
- fix(child): mirror whole package, scope tutor to current assignment, print-in-new-tab, package handoff button (f3257f9)
- feat: inline PDF preview, file-type icons, metadata + download for non-previewable files, read-file skill (397ff25)
- feat: durable cross-restart context (--resume handle) + Pulse, WhatsApp bot, Gmail, browser reuse (b394963)
- Make Goal Advisor strategy-first (61a6484)
- Zustand store migration, agent_browser integration, and a round of UX fixes (16efa71)
- Add per-workflow notification preferences + surface global Gmail UI (5262fcd)
- Add generic agent lifecycle controls (2649341)
- Compact Pulse review orchestration (00ca7a4)
- feat: real parent->child handoff (approve_for_child) + child file viewer + Give-to-child UI (74948b4)
- Honor live CDP download grants (3bd8f7c)
- feat: parent teaching-preference memory (parent/preferences.md) + Reference tab (900d465)
- Fix workflow Slack delivery across bridge sessions (3c799e8)
- frontend: finish arrow-key image navigation (wire viewerImageList to thumbnail clicks) (06a071d)
- frontend: real academic map, subject/topic/type file browser, thumbnails, arrow-key nav (8302b78)
- security: sandbox the parent's shell (Isolator), fix skills embed, add child-handoff + image-gen tools (b798e89)
- fix: create-test skill adherence — "short" test was collapsing to plain markdown (37bf93e)
- fix: remove injection-shaped inbox nudge; persist failed turns; live "Quill is:" status (430ff95)
- docs: record the reply-truncation fix found via browser testing (cd677df)
- docs: defer streaming — not a clean reuse in warm-resume/tmux mode (ea0acf2)
- docs: enshrine AgentWorks chief-of-staff + automations as the gold standard (3d33dac)
- refactor: match AgentWorks — one process-global bridge, per-turn agent, no session cache/LRU (3c6208d)
- fix: warm-resume sessions keep a live bridge across turns (fixes tool calls on resumed turns) (3506dea)
- process-file: add local OCR path (agent installs via shell), keep read_image for vision (9d0347a)
- fix: read_image works through multi-llm-provider (no claude -p) — deep fix (dcac14d)
- fix: read_image reads images reliably (exec claude CLI directly) (68b8cdb)
- feat: warm session resume — turns continue the same agent, not cold-start (4f6d581)
- fix: pin react + react-dom to 19.2.0 (were "*", causing a version mismatch) (f230740)
- UI: render chat with react-markdown + remark-gfm (same as main app); reorder tabs (9227056)
- fix: clean-markdown output + robust list rendering (terminal-wrapped replies) (62444b1)
- feat: use the gws CLI for notify (Gmail) + backup/publish (Drive); drop Materials tab (ae6af9f)
- feat: WhatsApp connector (agentic core + simulator) + plain-language + image-aware backup (8ff8781)
- feat: agentic publish + notify (AgentWorks-style) (4a71ea7)
- feat: agentic backup skill (AgentWorks-style) + WhatsApp icon (6d27d56)
- UI: clearer drawer tabs + Workspace Basic/All-files view (559a7e3)
- feat: read_image tool — the agent can actually see uploaded images (no OCR) (2a9159a)
- UI + skills: brand header, child-identity subline, type-specific designs (7ac7d9c)
- feat: interactive HTML persists to workspace + viewer/heading fixes (e88d4a7)
- UI: quick-actions menu in the composer (skill shortcuts) (152084c)
- feat: web-search-backed coaching + drawer/header cleanup (e74fc05)
- feat: default to interactive HTML + agent-suggested next-step pills (4150738)
- feat: agent open_file tool — created files auto-open in the right viewer (bf8f479)
- UI: right drawer always open + auto-scroll chat to latest (3e6bc8a)
- UI: drop redundant Subject&Topic card; New conversation in rail only (c737959)
- UI: fix composer input + de-clutter chat (d143170)
- UI: tighten chat/header side padding and widen assistant responses (3bace68)
- UI: Files tab — workspace tree + HTML/markdown viewer (AgentWorks-style) (61e208a)
- UI: readable notebook chat + rail-closed layout + resume-by-default (3b4c88e)
- skills: shared SparkQuill HTML design system + interactive process-file (ask when in doubt) (8671843)
- family-server: skills system + agentic file intake (process/generate) (4f37e54)
- UI: make the whole right drawer dynamic (Map + Progress), drop all mock (9aa37cd)
- family-server + UI: live Child Mode tutor screen (#5) (5ae6d24)
- family-server + UI: past conversations, file-system based (#12) (ac8542a)
- family-server: Child Mode — tutor persona + reused sandbox (#13) (3c2e41e)
- family-server: agentic file access via existing execute_shell_command (4e18d61)
- Drawer assets: dynamic from the workspace file system (not mock) (791b92a)
- Materials: file-system based, not a CRUD route (e4731e8)
- Upload: add /api/upload (multipart) + compose attach → workspace by subject/topic (c2685d0)
- SparkQuill (Family Learning) MVP: onboarding, Quill parent chat, Subject & Topic bridge tool (a477b46)
- Use equal split for tablet preview (27837f6)
- Default agent Slack notifications to rich Block Kit (3db327d)
- Teach Pulse Bug Review to flag self-asserted validation gates (7d2236b)
- Validate message_sequence steps on real output, not a forced JSON marker (4f754b9)
- Harden workflow runtime and browser recording lifecycle (6531194)
- Fix message sequence permission and freshness checks (f026291)
- Track workflow store freshness and harden message sequences (9678fcf)
- Harden terminal live attach rendering (2018c62)
- Fix synthetic terminal visibility (6ea947e)
- Test workflow knowledge store guidance (73e7115)
- Harden workflow runtime and browser contracts (d1d8ee3)
- Inherit account notification channels (90de11e)
- Harden shared CDP browser workflows (fce6b9c)
- fix: run manual Pulse without workflow execution (253cca0)
- feat: harden dynamic Pulse review pipeline (3169ee7)
- Repaint terminals without resizing (0b9ef63)
- Keep human feedback inside AgentWorks (de2f1f5)
- fix global human feedback for workflow agents (798bc58)
- Fix coding CLI bridge guid...
1.25.130
1.25.129
1.25.128
1.25.127
1.25.126
Desktop release v1.25.126.
Changes since v1.25.125: