genaicode 2.48.0 → 2.49.0

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
Files changed (59) hide show
  1. package/dist/agents/drivers/anthropic-managed.js +1 -0
  2. package/dist/agents/drivers/anthropic-managed.js.map +1 -1
  3. package/dist/agents/drivers/codex-live.d.ts +14 -0
  4. package/dist/agents/drivers/codex-live.js +64 -10
  5. package/dist/agents/drivers/codex-live.js.map +1 -1
  6. package/dist/agents/drivers/codex.d.ts +11 -1
  7. package/dist/agents/drivers/codex.js +27 -10
  8. package/dist/agents/drivers/codex.js.map +1 -1
  9. package/dist/agents/drivers/mcp-isolation.d.ts +1 -1
  10. package/dist/agents/drivers/mcp-isolation.js +2 -2
  11. package/dist/agents/drivers/mcp-isolation.js.map +1 -1
  12. package/dist/agents/live-agent.d.ts +2 -2
  13. package/dist/agents/live-agent.js +95 -79
  14. package/dist/agents/live-agent.js.map +1 -1
  15. package/dist/agents/session.d.ts +6 -1
  16. package/dist/agents/session.js +33 -8
  17. package/dist/agents/session.js.map +1 -1
  18. package/dist/agents/sessions/files.d.ts +7 -0
  19. package/dist/agents/sessions/files.js +19 -0
  20. package/dist/agents/sessions/files.js.map +1 -1
  21. package/dist/agents/sessions/index.d.ts +1 -0
  22. package/dist/agents/sessions/index.js +20 -2
  23. package/dist/agents/sessions/index.js.map +1 -1
  24. package/dist/agents/sessions/shares.d.ts +31 -0
  25. package/dist/agents/sessions/shares.js +68 -0
  26. package/dist/agents/sessions/shares.js.map +1 -0
  27. package/dist/agents/sessions/stores.js +8 -2
  28. package/dist/agents/sessions/stores.js.map +1 -1
  29. package/dist/agents/sessions/topics.js +2 -6
  30. package/dist/agents/sessions/topics.js.map +1 -1
  31. package/dist/agents/sessions/window.d.ts +18 -0
  32. package/dist/agents/sessions/window.js +300 -0
  33. package/dist/agents/sessions/window.js.map +1 -0
  34. package/dist/agents/types.d.ts +6 -1
  35. package/dist/cli/args.d.ts +2 -0
  36. package/dist/cli/args.js +2 -0
  37. package/dist/cli/args.js.map +1 -1
  38. package/dist/cli/main.d.ts +10 -1
  39. package/dist/cli/main.js +24 -8
  40. package/dist/cli/main.js.map +1 -1
  41. package/dist/react/agent-chat.css +162 -6
  42. package/dist/react/agent-chat.d.ts +27 -6
  43. package/dist/react/agent-chat.js +125 -24
  44. package/dist/react/agent-chat.js.map +1 -1
  45. package/dist/react/format.d.ts +20 -0
  46. package/dist/react/format.js +163 -0
  47. package/dist/react/format.js.map +1 -1
  48. package/dist/react/markdown.js +25 -2
  49. package/dist/react/markdown.js.map +1 -1
  50. package/dist/react.d.ts +2 -2
  51. package/dist/react.js +1 -1
  52. package/dist/react.js.map +1 -1
  53. package/dist/ui/index.js +154 -119
  54. package/dist/ui/web-client.js +190 -24
  55. package/docs/CHANGELOG.md +22 -0
  56. package/docs/agents.md +9 -2
  57. package/docs/cli.md +11 -4
  58. package/docs/react.md +19 -16
  59. package/package.json +1 -1
package/docs/CHANGELOG.md CHANGED
@@ -8,6 +8,28 @@ fails. Merging to master refreshes the `release` PR; merging that publishes to n
8
8
 
9
9
  ## Unreleased
10
10
 
11
+ ## 2.49.0 — 2026-11-22
12
+
13
+ ### Added
14
+
15
+ - `/effort [level|default]` in `genaicode chat`, `ui` and `tower` changes the reasoning effort from the next prompt, and the browser has an effort picker in the row under the prompt field for an agent that takes one. `AgentSession.setEffort()` does it for a host; switching to an agent that does not know the level goes back to that agent's default (#269)
16
+ - `genaicode chat`, `ui` and `tower` run Codex without the MCP servers of its own config, which it would start with every turn; `--agent-mcp` keeps them. `codexLive({ onlyTaskMcpServers })` does it for the live driver (`'when-listed'`: a Codex that cannot list its servers keeps them, with a note), and `liveAgent`'s `prepare` may be async (#267)
17
+ - `genaicode chat`, `ui` and `tower` run Codex over its live driver (`codexLive`), as a pack already did: "Ask first" asks before a step outside the sandbox, a message sent mid-turn steers the run, Codex's plan limits show, and an edit by Codex shows its diff. `genaicode run` keeps the headless driver (#266)
18
+ - `<AgentChat>`: an edit an agent reports as changed files (Codex) shows its diff like an edit step, and a link to a file in a reply (`[a.txt](/path/a.txt)`) shows as its label and opens the file with `onOpenFile`. The `file-change` event and the session `files` entry take an optional `diff` (#265)
19
+ - `genaicode ui`: the plan limits, agent, model, approval mode and the attach button sit in one row under the prompt field, which holds the text and Send only. `<AgentChat composerExtras>` renders in that row, with the attach buttons (#264)
20
+ - `genaicode ui`: the path of a file the agent edited or read opens the file beside the chat (text files inside the project only). `<AgentChat>` takes `onOpenFile` to make those paths buttons, and `toolPath` is exported from `genaicode/react` (#262)
21
+ - `<AgentChat>`: a step that edits a file shows what it changed, added and removed lines, when opened, with the counts on its line. `toolDiff` is exported from `genaicode/react` (#261)
22
+ - `genaicode ui`: the agent, the model, the approval mode and the plan's limits sit in the prompt box instead of the top bar, and files can be attached, pasted or dropped on the chat (kept in `.genaicode/uploads/`, their paths go to the agent). `<AgentChat>` takes `composerExtras` for host controls in the composer (#260)
23
+ - `<AgentChat completions>` offers the host's commands while the user types one (arrows, Tab or Enter to complete, Escape to close), and `genaicode ui` lists its `/` commands that way (#259)
24
+ - The start screen of `ui --sessions` (the desktop app) says, under each agent's plan limits, which topics that agent's work of the last 7 days went to, in percent; `topicShares()` in `genaicode/agents` computes it (#258)
25
+
26
+ ### Fixed
27
+
28
+ - The start screen's topic shares no longer read a long Codex rollout from its start every time: what was read is kept under `~/.genaicode/cache/rollouts`, and a later look reads only what the rollout gained (a 2 GB thread took about 9 s per look). `partsSince(sessions, since, { env })` says where home is (#270)
29
+ - `genaicode ui`: "Its work in the last 7 days went to" counts only what each session did in those days. A long session touched once in the period no longer counts with its whole life (read from Codex rollouts and Claude Code transcripts, estimated from the lifetime elsewhere). `topicShares` takes these parts from the new `partsSince()` (#268)
30
+ - `genaicode ui`: the agent, model and approval controls in the prompt box have an outline, so they read as controls. A note Codex sends about a hook timeout it clamps no longer shows as an error in the conversation (it is a `stderr` event) (#263)
31
+ - Sessions a program started in a temporary directory or in a `.genaicode/worktrees` worktree, and Antigravity command-line runs outside any project, no longer show in session listings or topics (`sessions --automated` lists them) (#256)
32
+
11
33
  ## 2.48.0 — 2026-11-21
12
34
 
13
35
  ### Added
package/docs/agents.md CHANGED
@@ -354,7 +354,9 @@ Sessions nobody typed into are left out of every listing: scheduled tasks and no
354
354
  Claude Code (every prompt has `origin.kind` or `turnOrigin` other than `human`, or the session
355
355
  opens with a scheduled task nobody answered), headless runs of it with a single prompt (a program
356
356
  driving it through more prompts is somebody's chat), `codex exec` runs, threads the Codex desktop
357
- app's server ran for another program, and Muse Code's probes. They carry `automated: true`;
357
+ app's server ran for another program, Muse Code's probes, Antigravity runs a program started
358
+ outside any project, and any agent's session in a temporary directory or in a worktree under
359
+ `.genaicode/worktrees` (unless it asks for approval right now). They carry `automated: true`;
358
360
  `sessions --automated`, or `{ automated: true }` in the query, lists them too.
359
361
 
360
362
  **Topics.** Sessions can be kept as pieces of work that cross agents and directories
@@ -386,7 +388,12 @@ prompt), quoted as data. `renameTopic()`, `mergeTopics()` and `moveSessions()` c
386
388
  session that has a topic is never moved by a later answer. `projectTopics(sessions, store)` places
387
389
  the sessions relayed from a claude.ai project (`SessionInfo.project`: its id, and its name where
388
390
  the session gives it; such a session is titled as its thread there) in a topic named as the project, without an agent. What ran in a pack member's worktree
389
- (`.genaicode/worktrees`) or a temporary directory (`inRunDirectory()`) is listed as ever but takes no topic.
391
+ (`.genaicode/worktrees`) or a temporary directory (`inRunDirectory()`) takes no topic.
392
+
393
+ `topicShares(sessions, store)` says where each agent's work went: per agent, the part every topic
394
+ took (0 to 1), with `Automatic runs` and `Not sorted yet` as parts of their own. Sessions are
395
+ weighed by `costUsd` where the agent keeps one, else by tokens; only proportions are returned.
396
+ List the sessions with `automated: true`, or scheduled runs are missing from the picture.
390
397
 
391
398
  For Copilot and Gemini shell approvals, `waitingCommands` contains the tool name and up to
392
399
  200 characters of each matching command. Genaicode keeps a digest in its private hook record
package/docs/cli.md CHANGED
@@ -206,7 +206,10 @@ updates: counts per store, never content, exit code 1 when one looks broken (see
206
206
 
207
207
  With `ui --sessions`, and once `genaicode topics refresh` has sorted your sessions, the empty
208
208
  chat page starts with your topics: those that wait for you, those agents work on now, and how
209
- much of each agent's plan is used. A topic opens the Sessions view on its sessions.
209
+ much of each agent's plan is used. Under each plan it says which topics that agent's work of the
210
+ last 7 days went to, in percent, with automatic runs as a part of their own. The parts are weighed
211
+ by what the agent says its sessions used (its price estimate, else tokens) in their whole life, so
212
+ they are a guide, not a bill. A topic opens the Sessions view on its sessions.
210
213
 
211
214
  In `npx genaicode chat --sessions`, ask "Which sessions need my attention?" The
212
215
  `sessions_waiting` tool shows the same sessions as `genaicode sessions --waiting`, including
@@ -320,11 +323,11 @@ Each prompt after the first continues the agent's own session (`AgentTask.resume
320
323
  can say "now add a test for that". Prompts sent while the agent works wait in a queue and
321
324
  run next; Esc stops the agent. When the agent asks for permission, `y` approves, `n`
322
325
  denies and `a` approves everything from then on (`/auto`, **Auto-approve all** on the request, or the
323
- Auto-approve chip in the browser; `-y` starts that way). `!` on an empty prompt switches to shell mode, as in Claude
326
+ the mode button in the browser's prompt box; `-y` starts that way). `!` on an empty prompt switches to shell mode, as in Claude
324
327
  Code: `!npm run dev` runs in the project with its output in the transcript, and `/stop` (or
325
328
  Esc) ends it. `/agent`, `/model` and `/new` switch agent,
326
- model and session; `/fallback codex,gemini` picks the agents to switch to when the current one
327
- runs out (`/fallback off` stops it); `/help` lists the rest. `/agent codex` hands the
329
+ model and session; `/effort high` sets how hard the agent thinks from the next prompt, in the agent's own words (`/effort` lists them, `/effort default` clears it); `/fallback codex,gemini` picks the agents to switch to when the current one
330
+ runs out (`/fallback off` stops it); `/help` lists the rest, and the browser offers them as you type `/`. In the browser the agent, the model, the effort (for an agent that takes one), the approval mode and the plan's limits sit in a row under the prompt field, and files can be attached, pasted (a copied screenshot) or dropped on the chat: they are kept in `.genaicode/uploads/` in the project (git-ignored, 20 MB each, 8 per prompt) and the agent gets their paths. An edit step shows its diff, and the path of a file the agent read or edited opens the file beside the chat (text files inside the project only). `/agent codex` hands the
328
331
  conversation over: codex gets a recap and a transcript file in `.genaicode/handoff/`, and
329
332
  switching back resumes the earlier agent's own session (`/agent codex --fresh` starts with no
330
333
  context).
@@ -372,6 +375,10 @@ The terminal UI rings the terminal bell instead. `/bark off` (or the Bark button
372
375
  prompt, e.g. `genaicode ui --mcp game=http://127.0.0.1:7777/mcp` from a game editor whose
373
376
  running game serves tools there. Only the agents that take MCP servers in genaicode (claude,
374
377
  codex, copilot) are offered then.
378
+ Codex runs in `chat`, `ui` and `tower` without the MCP servers of its own config (`~/.codex/config.toml`,
379
+ plugins): it starts every one of them with each turn, which is slow and can open apps you did not ask
380
+ for. `--agent-mcp` keeps them. A Codex too old to list its servers (`codex mcp list --json`) keeps
381
+ them too, with a note.
375
382
  The agent can publish artifacts: a report, chart or mockup as a self-contained HTML page, an
376
383
  image, a Markdown document. The browser UI shows each as a card that opens a preview; the
377
384
  terminal shows a line, and `/web` opens the browser on the same session. Pages run in a
package/docs/react.md CHANGED
@@ -71,22 +71,25 @@ HTML; links open in a new tab; GFM tables, strikethrough and task lists), collap
71
71
  own approval requests, queued prompts and a working indicator, and an input where Enter
72
72
  sends, Shift+Enter adds a line and Esc stops. Props:
73
73
 
74
- | Prop | |
75
- | ------------------------------------------------------------------------------ | ------------------------------------------------------------------------------------------------------------------ |
76
- | `state` | `SessionState`, or null before there is a session |
77
- | `onSend(text)`, `onStop()` | Enter or Send; Stop or Esc. No `onStop`, no Stop button |
78
- | `onApprove(id, decision)` | answers `state.approvals`; without it they show as waiting |
79
- | `onAutoApprove()` | optional third button ("Auto-approve all") beside Approve and Deny to approve pending and future requests |
80
- | `onPackAction(turnId, id, action)` | pass the session's `actOnPackTask` to enable worker controls and return a diff or result text |
81
- | `renderPrompt(text, turn?)` | how a sent prompt shows (return null to hide it) |
82
- | `renderAfterTurn(turn, i)`, `renderFooter()` | your content after a turn, or at the end |
83
- | `notices` | `{ id, afterTurn, text, tone? }[]`: your own lines between turns |
84
- | `artifactUrl(artifact)`, `renderArtifact(artifact, url)` | where your server serves a published artifact; your own card ([artifacts](agents.md#artifacts)) |
85
- | `onArtifactSubmit(artifact, data)` | a page the agent published sent data and the user confirmed it; default sends it as a message; `null` turns it off |
86
- | `foldSteps` | fold each turn's tools and edits into one "N steps" line |
87
- | `header`, `headerExtras` | replace the header (null for none), or add to its right end |
88
- | `emptyState`, `placeholder`, `hint` | text before the first turn, in and under the input |
89
- | `formatToolName(name)`, `disabled`, `autoFocus`, `theme`, `className`, `style` | |
74
+ | Prop | |
75
+ | ------------------------------------------------------------------------------ | --------------------------------------------------------------------------------------------------------------------------- |
76
+ | `state` | `SessionState`, or null before there is a session |
77
+ | `onSend(text)`, `onStop()` | Enter or Send; Stop or Esc. No `onStop`, no Stop button |
78
+ | `onApprove(id, decision)` | answers `state.approvals`; without it they show as waiting |
79
+ | `onAutoApprove()` | optional third button ("Auto-approve all") beside Approve and Deny to approve pending and future requests |
80
+ | `onPackAction(turnId, id, action)` | pass the session's `actOnPackTask` to enable worker controls and return a diff or result text |
81
+ | `renderPrompt(text, turn?)` | how a sent prompt shows (return null to hide it) |
82
+ | `renderAfterTurn(turn, i)`, `renderFooter()` | your content after a turn, or at the end |
83
+ | `notices` | `{ id, afterTurn, text, tone? }[]`: your own lines between turns |
84
+ | `artifactUrl(artifact)`, `renderArtifact(artifact, url)` | where your server serves a published artifact; your own card ([artifacts](agents.md#artifacts)) |
85
+ | `onArtifactSubmit(artifact, data)` | a page the agent published sent data and the user confirmed it; default sends it as a message; `null` turns it off |
86
+ | `foldSteps` | fold each turn's tools and edits into one "N steps" line. An edit step shows its diff (added and removed lines) when opened |
87
+ | `header`, `headerExtras` | replace the header (null for none), or add to its right end |
88
+ | `completions` | commands the host understands (`{ name, args?, help? }`), offered while the message starts one |
89
+ | `composerExtras` | host controls in a row under the message field, beside the attach button (agent, model, mode) |
90
+ | `onOpenFile` | makes file paths in the transcript (edited files, the file a tool step read) buttons; the host shows the file |
91
+ | `emptyState`, `placeholder`, `hint` | text before the first turn, in and under the input |
92
+ | `formatToolName(name)`, `disabled`, `autoFocus`, `theme`, `className`, `style` | |
90
93
 
91
94
  Theme it with CSS custom properties on the component or any ancestor: `--gc-bg`,
92
95
  `--gc-panel`, `--gc-border`, `--gc-text`, `--gc-muted`, `--gc-accent`, `--gc-accent-text`,
package/package.json CHANGED
@@ -1,6 +1,6 @@
1
1
  {
2
2
  "name": "genaicode",
3
- "version": "2.48.0",
3
+ "version": "2.49.0",
4
4
  "description": "Drive the coding-agent CLIs you already have (Claude Code, Codex, Copilot, Cursor, Gemini, opencode...) behind one API and one CLI, plus a thin provider-neutral LLM layer.",
5
5
  "author": "Grzegorz Tańczyk",
6
6
  "repository": {