genaicode 2.43.0 → 2.44.0
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.
- package/dist/agents/cli-agent.js +2 -1
- package/dist/agents/cli-agent.js.map +1 -1
- package/dist/agents/drivers/claude.d.ts +15 -2
- package/dist/agents/drivers/claude.js +89 -8
- package/dist/agents/drivers/claude.js.map +1 -1
- package/dist/agents/live-agent.js +2 -1
- package/dist/agents/live-agent.js.map +1 -1
- package/dist/agents/runtime.d.ts +5 -1
- package/dist/agents/runtime.js +7 -1
- package/dist/agents/runtime.js.map +1 -1
- package/dist/agents/sessions/claude.js +27 -1
- package/dist/agents/sessions/claude.js.map +1 -1
- package/dist/agents/sessions/codex-api.js +2 -0
- package/dist/agents/sessions/codex-api.js.map +1 -1
- package/dist/agents/sessions/codex.js +7 -0
- package/dist/agents/sessions/codex.js.map +1 -1
- package/dist/agents/sessions/group.d.ts +9 -2
- package/dist/agents/sessions/group.js +37 -2
- package/dist/agents/sessions/group.js.map +1 -1
- package/dist/agents/sessions/hooks.d.ts +10 -0
- package/dist/agents/sessions/hooks.js +24 -3
- package/dist/agents/sessions/hooks.js.map +1 -1
- package/dist/agents/sessions/index.d.ts +4 -1
- package/dist/agents/sessions/index.js +28 -4
- package/dist/agents/sessions/index.js.map +1 -1
- package/dist/agents/sessions/limits.d.ts +39 -0
- package/dist/agents/sessions/limits.js +279 -0
- package/dist/agents/sessions/limits.js.map +1 -0
- package/dist/agents/sessions/tools.js +3 -2
- package/dist/agents/sessions/tools.js.map +1 -1
- package/dist/agents/sessions/topics.d.ts +86 -0
- package/dist/agents/sessions/topics.js +242 -0
- package/dist/agents/sessions/topics.js.map +1 -0
- package/dist/agents/sessions/types.d.ts +14 -0
- package/dist/agents.d.ts +2 -2
- package/dist/agents.js +1 -1
- package/dist/agents.js.map +1 -1
- package/dist/cli/args.d.ts +39 -0
- package/dist/cli/args.js +86 -0
- package/dist/cli/args.js.map +1 -1
- package/dist/cli/main.d.ts +1 -1
- package/dist/cli/main.js +276 -5
- package/dist/cli/main.js.map +1 -1
- package/dist/ui/index.js +108 -104
- package/dist/ui/web-client.js +21 -16
- package/docs/CHANGELOG.md +13 -0
- package/docs/agents.md +61 -15
- package/docs/cli.md +66 -4
- package/package.json +1 -1
package/docs/CHANGELOG.md
CHANGED
|
@@ -8,6 +8,19 @@ fails. Merging to master refreshes the `release` PR; merging that publishes to n
|
|
|
8
8
|
|
|
9
9
|
## Unreleased
|
|
10
10
|
|
|
11
|
+
## 2.44.0 — 2026-11-21
|
|
12
|
+
|
|
13
|
+
### Added
|
|
14
|
+
|
|
15
|
+
- `genaicode topics`: your sessions as pieces of work, whichever agents and directories they were in, with what asks for you first. `topics refresh` has an agent sort the sessions no topic has yet (one line per session; kept in `~/.genaicode/topics.json`), and `topics rename|merge|move` correct it. From `genaicode/agents`: `topicsPrompt()`, `sortedTopics()`, `summarizeTopics()`, `readTopics()`, `writeTopics()` (#241)
|
|
16
|
+
- `genaicode sessions` leaves out the sessions nobody typed into (Claude Code's scheduled tasks and notifications, `codex exec` runs; `--automated` or `{ automated: true }` lists them, and they carry `automated: true`), and says what a waiting session's last message asks of you: `SessionInfo.ending` (`question`, `command`, `offer`, `unreported`) from `replyEnding()`. `sessions --waiting`, the session tools and the desktop menu show those right after the approvals (#240)
|
|
17
|
+
- `genaicode limits`: how much of each agent's plan is used (share of each window, when it resets), as last seen and without asking any agent. Runs through genaicode keep what the agents report, Codex's limits are read from its sessions, and `genaicode hooks install` makes `genaicode limits record` Claude Code's status line (unless you have your own), which keeps Claude's. `sessions --waiting --json --watch` carries them as `limits`, and the desktop app's menu shows them (#235)
|
|
18
|
+
- `claude()` inherits `permissions.defaultMode` from Claude Code user settings (`~/.claude/settings.json` / `$CLAUDE_CONFIG_DIR`) or `options.settings` instead of overriding it with `--permission-mode acceptEdits`, keeping `acceptEdits` as the fallback when no default mode is configured (#238)
|
|
19
|
+
|
|
20
|
+
### Fixed
|
|
21
|
+
|
|
22
|
+
- `genaicode ui` (and `chat --web` / `tower`) strips `?token=` from the browser address bar into origin-isolated `sessionStorage`, sends the token in headers for `/api/events` and commands, and keeps reloads working via a reload-only `HttpOnly` shell cookie (#236)
|
|
23
|
+
|
|
11
24
|
## 2.43.0 — 2026-11-21
|
|
12
25
|
|
|
13
26
|
### Added
|
package/docs/agents.md
CHANGED
|
@@ -327,6 +327,47 @@ the command that continues it. It looks a week back unless `--since` says otherw
|
|
|
327
327
|
browser's Sessions view has the same as a "N wait for you" button; in code,
|
|
328
328
|
`waitingSessions(sessions)` picks and orders them.
|
|
329
329
|
|
|
330
|
+
Not every waiting session asks for something: most just ended. `SessionInfo.ending` says what the
|
|
331
|
+
end of the agent's last message asks, read from its shape by `replyEnding(lastReply)`: `question`
|
|
332
|
+
(it ends on a question), `command` (it ends on a shell code block for you to run), `offer` ("I can
|
|
333
|
+
also...", "let me know") or `unreported` (it says it could not report back to the thread that sent
|
|
334
|
+
it). Those come right after the approvals in `--waiting`, marked `●`, and `sessions_list` says
|
|
335
|
+
`[ends with: question]`. It is a guess from the last 200 characters: a request put as a plain
|
|
336
|
+
statement is not seen.
|
|
337
|
+
|
|
338
|
+
Sessions nobody typed into are left out of every listing: scheduled tasks and notifications in
|
|
339
|
+
Claude Code (every prompt has `origin.kind` other than `human`, or the session opens with a
|
|
340
|
+
scheduled task nobody answered) and `codex exec` runs. They carry `automated: true`;
|
|
341
|
+
`sessions --automated`, or `{ automated: true }` in the query, lists them too.
|
|
342
|
+
|
|
343
|
+
**Topics.** Sessions can be kept as pieces of work that cross agents and directories
|
|
344
|
+
([`genaicode topics`](cli.md#your-work-as-topics-topics)). An agent of your choice does the
|
|
345
|
+
sorting; genaicode writes the prompt, reads the answer and keeps the result:
|
|
346
|
+
|
|
347
|
+
```ts
|
|
348
|
+
import {
|
|
349
|
+
listSessions,
|
|
350
|
+
readTopics,
|
|
351
|
+
sortedTopics,
|
|
352
|
+
summarizeTopics,
|
|
353
|
+
topicsPrompt,
|
|
354
|
+
unsortedSessions,
|
|
355
|
+
writeTopics,
|
|
356
|
+
} from 'genaicode/agents';
|
|
357
|
+
|
|
358
|
+
const { sessions } = await listSessions(sources, { since: Date.now() - 30 * 86_400_000 });
|
|
359
|
+
const store = await readTopics();
|
|
360
|
+
const todo = unsortedSessions(sessions, store).slice(0, 120);
|
|
361
|
+
const { text } = await agent.run({ prompt: topicsPrompt(todo, store), cwd: emptyDir }).result;
|
|
362
|
+
const sorted = sortedTopics(text ?? '', todo, store); // { store, placed }
|
|
363
|
+
await writeTopics(sorted.store);
|
|
364
|
+
const { topics, unsorted } = summarizeTopics(sessions, sorted.store); // asking first, then the latest
|
|
365
|
+
```
|
|
366
|
+
|
|
367
|
+
The prompt holds one line per session (agent, last two directory names, title or first and last
|
|
368
|
+
prompt), quoted as data. `renameTopic()`, `mergeTopics()` and `moveSessions()` correct a store; a
|
|
369
|
+
session that has a topic is never moved by a later answer.
|
|
370
|
+
|
|
330
371
|
For Copilot and Gemini shell approvals, `waitingCommands` contains the tool name and up to
|
|
331
372
|
200 characters of each matching command. Genaicode keeps a digest in its private hook record
|
|
332
373
|
and reads the text on demand from the agent's own local session journal. A session with a
|
|
@@ -345,6 +386,11 @@ The command in the settings is the `genaicode` that installed it (`node` and the
|
|
|
345
386
|
`cli.js`): install it from a `genaicode` that stays where it is (a project's dependency or a
|
|
346
387
|
global install), and run `hooks install` again after moving it.
|
|
347
388
|
|
|
389
|
+
For Claude Code, `hooks install` also sets `statusLine` in the same settings to `genaicode limits
|
|
390
|
+
record`, which keeps the plan limits Claude Code reports for [`genaicode limits`](cli.md#how-much-of-your-plans-is-used-limits)
|
|
391
|
+
and prints them as the line. A status line you set yourself is never replaced; `hooksStatus()`
|
|
392
|
+
says which it is (`statusLine: 'genaicode' | 'other' | 'none'`).
|
|
393
|
+
|
|
348
394
|
### Sending a turn to a session
|
|
349
395
|
|
|
350
396
|
`sendSession()` accepts a user's message for the local session's own driver. It checks the
|
|
@@ -703,19 +749,19 @@ Artifacts are agent-written content, so they are handled as untrusted:
|
|
|
703
749
|
|
|
704
750
|
## Drivers
|
|
705
751
|
|
|
706
|
-
| Driver | Command | Mode | Notes
|
|
707
|
-
| --------------------- | -------------- | ---------------------------------------- |
|
|
708
|
-
| `claude(options?)` | `claude` | `-p --output-format stream-json` | `
|
|
709
|
-
| `codex(options?)` | `codex` | `exec --json` | `sandbox` defaults to `workspace-write`; effort via `model_reasoning_effort`
|
|
710
|
-
| `muse(options?)` | `muse` | `exec --json --trust-workspace` | `maxTurns` maps to `--max-model-steps`
|
|
711
|
-
| `codexLive(options?)` | `codex` | `app-server` (JSON-RPC) | `steer()`, approvals, MCP. See [Live sessions](#live-sessions-steering-and-approvals)
|
|
712
|
-
| `museLive(options?)` | `muse` | `serve` (JSON-RPC) | `steer()`, approvals. See [Live sessions](#live-sessions-steering-and-approvals)
|
|
713
|
-
| `gemini(options?)` | `gemini` | `--output-format stream-json --prompt=…` | `approvalMode` defaults to `auto_edit`; `--skip-trust` unless `trustWorkspace: false`
|
|
714
|
-
| `cursor(options?)` | `cursor-agent` | `-p --output-format stream-json` | `--force --approve-mcps` by default (`--trust` if `force: false`); `partialOutput`
|
|
715
|
-
| `opencode(options?)` | `opencode` | `run --format json` | `model` is `provider/model`; `effort` maps to `--variant`; `autoApprove` adds `--auto`
|
|
716
|
-
| `copilot(options?)` | `copilot` | `--output-format json --prompt=…` | `--allow-all-tools --no-ask-user` by default; `allowTools` / `denyTools` patterns
|
|
717
|
-
| `vibe(options?)` | `vibe` | `--output streaming --prompt=…` | `agent` defaults to `accept-edits`; `--trust`; `model` via `VIBE_ACTIVE_MODEL`
|
|
718
|
-
| `antigravity(opts?)` | `agy` | `acp` (JSON-RPC) | `steer()`, approvals, MCP, resume; `mode` defaults to `accept-edits`; `--sandbox`
|
|
752
|
+
| Driver | Command | Mode | Notes |
|
|
753
|
+
| --------------------- | -------------- | ---------------------------------------- | --------------------------------------------------------------------------------------------------------------- |
|
|
754
|
+
| `claude(options?)` | `claude` | `-p --output-format stream-json` | inherits user `permissions.defaultMode` (fallback `acceptEdits`); `allowedTools` / `disallowedTools`; approvals |
|
|
755
|
+
| `codex(options?)` | `codex` | `exec --json` | `sandbox` defaults to `workspace-write`; effort via `model_reasoning_effort` |
|
|
756
|
+
| `muse(options?)` | `muse` | `exec --json --trust-workspace` | `maxTurns` maps to `--max-model-steps` |
|
|
757
|
+
| `codexLive(options?)` | `codex` | `app-server` (JSON-RPC) | `steer()`, approvals, MCP. See [Live sessions](#live-sessions-steering-and-approvals) |
|
|
758
|
+
| `museLive(options?)` | `muse` | `serve` (JSON-RPC) | `steer()`, approvals. See [Live sessions](#live-sessions-steering-and-approvals) |
|
|
759
|
+
| `gemini(options?)` | `gemini` | `--output-format stream-json --prompt=…` | `approvalMode` defaults to `auto_edit`; `--skip-trust` unless `trustWorkspace: false` |
|
|
760
|
+
| `cursor(options?)` | `cursor-agent` | `-p --output-format stream-json` | `--force --approve-mcps` by default (`--trust` if `force: false`); `partialOutput` |
|
|
761
|
+
| `opencode(options?)` | `opencode` | `run --format json` | `model` is `provider/model`; `effort` maps to `--variant`; `autoApprove` adds `--auto` |
|
|
762
|
+
| `copilot(options?)` | `copilot` | `--output-format json --prompt=…` | `--allow-all-tools --no-ask-user` by default; `allowTools` / `denyTools` patterns |
|
|
763
|
+
| `vibe(options?)` | `vibe` | `--output streaming --prompt=…` | `agent` defaults to `accept-edits`; `--trust`; `model` via `VIBE_ACTIVE_MODEL` |
|
|
764
|
+
| `antigravity(opts?)` | `agy` | `acp` (JSON-RPC) | `steer()`, approvals, MCP, resume; `mode` defaults to `accept-edits`; `--sandbox` |
|
|
719
765
|
|
|
720
766
|
Every driver accepts `command` to point at a specific executable. `task.extraArgs` is
|
|
721
767
|
inserted before the prompt for flags that have no portable field.
|
|
@@ -784,8 +830,8 @@ Per agent:
|
|
|
784
830
|
- `claude()`: with `onApproval`, the driver serves Claude Code's `--permission-prompt-tool`
|
|
785
831
|
from a private MCP endpoint on 127.0.0.1 (random port, random bearer token), next to the
|
|
786
832
|
task's own `mcpServers`, and raises `MCP_TOOL_TIMEOUT` to a day unless `env` sets it. The
|
|
787
|
-
permission mode still applies first (`
|
|
788
|
-
is asked. Approving allows exactly the input Claude asked about (`updatedInput` unchanged).
|
|
833
|
+
permission mode still applies first (the user's `permissions.defaultMode` when set, else
|
|
834
|
+
`acceptEdits`): only what it does not allow is asked. Approving allows exactly the input Claude asked about (`updatedInput` unchanged).
|
|
789
835
|
- `codexLive()`: runs with approval policy `on-request` when `onApproval` is set (`never`
|
|
790
836
|
otherwise). Commands and file changes are answered `accept` / `decline`. A permission
|
|
791
837
|
profile request (`item/permissions/requestApproval`) is asked with `scope: 'turn'`; approving
|
package/docs/cli.md
CHANGED
|
@@ -101,6 +101,12 @@ must finish first. The check uses the latest store and hook information availabl
|
|
|
101
101
|
`npx genaicode notify on` then shows a desktop notification when a session asks for approval
|
|
102
102
|
(`--done`: also when one finishes its turn; see [notify](agents.md#being-told-notify)).
|
|
103
103
|
|
|
104
|
+
`sessions --waiting` puts the sessions whose last message asks you something right after the
|
|
105
|
+
approvals (`●`: a question, a command to run, an offer to go on, or a report that could not be
|
|
106
|
+
sent back), before the ones that just ended. Scheduled tasks and headless runs nobody typed into
|
|
107
|
+
are left out of `sessions`; `--automated` lists them too (see
|
|
108
|
+
[agents](agents.md#which-sessions-need-you-hooks)).
|
|
109
|
+
|
|
104
110
|
With Copilot or Gemini hooks installed, `sessions --waiting`, the tower's `/waiting` command
|
|
105
111
|
and the browser's Sessions view show up to 200 characters of a shell command awaiting approval.
|
|
106
112
|
Refresh the queue to see which commands still need a decision. Answer in the agent that asked.
|
|
@@ -117,7 +123,60 @@ asking again and again: `npx genaicode sessions --waiting --json --watch` prints
|
|
|
117
123
|
`{"sessions":[...]}` at once and another whenever the waiting sessions or their open questions
|
|
118
124
|
change. With `--exit-with-stdin` it ends when its stdin closes, so it cannot outlive its parent.
|
|
119
125
|
It looks every 3 seconds: a question or anything else a hook reports shows at once, a session
|
|
120
|
-
that only finished its turn within 15 seconds.
|
|
126
|
+
that only finished its turn within 15 seconds. Once limits are known, the line also carries them
|
|
127
|
+
as `"limits":[...]`.
|
|
128
|
+
|
|
129
|
+
### How much of your plans is used: `limits`
|
|
130
|
+
|
|
131
|
+
`npx genaicode limits` shows how much of each agent's plan is used, in the plan's own measure: the
|
|
132
|
+
share of each window and when it resets, never dollars.
|
|
133
|
+
|
|
134
|
+
```
|
|
135
|
+
claude 5h 62% used resets in 2 h 10 min (read 4 min ago)
|
|
136
|
+
7d 41% used resets in 3 d (read 4 min ago)
|
|
137
|
+
codex 5h 12% used resets in 48 min (read 1 h 5 min ago)
|
|
138
|
+
```
|
|
139
|
+
|
|
140
|
+
No agent is asked for it, so looking costs nothing. The readings come from three places:
|
|
141
|
+
|
|
142
|
+
- every run through genaicode (`run`, `chat`, `ui`, the library) keeps what the agent reports in
|
|
143
|
+
`~/.genaicode/limits` (`GENAICODE_LIMITS=off` keeps nothing);
|
|
144
|
+
- Codex notes its limits in its own session files, and the newest ones are read;
|
|
145
|
+
- Claude Code gives them to its status line only. `npx genaicode hooks install` makes
|
|
146
|
+
`genaicode limits record` that status line (it prints `5h 62% · 7d 41%`), and `hooks uninstall`
|
|
147
|
+
takes it out again. A status line of your own is left as it is: have it pass its input on to
|
|
148
|
+
`genaicode limits record` as well.
|
|
149
|
+
|
|
150
|
+
A window that has reset since it was read is left out: what is used of it now is not known until
|
|
151
|
+
the agent says so again. `--json` prints `{"limits":[{"agent","window","status","usedPercent","resetsAt","at"}]}`.
|
|
152
|
+
|
|
153
|
+
### Your work as topics: `topics`
|
|
154
|
+
|
|
155
|
+
A project directory is not a piece of work: one holds many, and one piece of work is often done
|
|
156
|
+
with several agents and from several directories. `npx genaicode topics` shows the sessions of all
|
|
157
|
+
agents as topics, what asks for you first:
|
|
158
|
+
|
|
159
|
+
```
|
|
160
|
+
● Pack mode t1 claude 9, codex 3 12m ago 1 ask you src/genaicode
|
|
161
|
+
▶ Desktop app t2 claude 6 1h ago 1 working src/genaicode, src/site
|
|
162
|
+
YouTube pins t3 codex 2 2d ago src/site
|
|
163
|
+
14 sessions not sorted yet: "genaicode topics refresh" sorts them.
|
|
164
|
+
```
|
|
165
|
+
|
|
166
|
+
What belongs together is a matter of what was said, so an agent decides it, and only when you ask:
|
|
167
|
+
`npx genaicode topics refresh` gives an agent one line per session no topic has yet (agent,
|
|
168
|
+
directory, title or first and last prompt; nothing else of the conversation) and keeps its answer
|
|
169
|
+
in `~/.genaicode/topics.json`. The next refresh sorts only the new sessions into the topics there
|
|
170
|
+
are. `-a` picks the agent and `-m` the model (Claude Code sorts with `haiku` unless told
|
|
171
|
+
otherwise); `--since` looks further back than the 30 days it and `topics` cover by default.
|
|
172
|
+
|
|
173
|
+
The result is yours to correct, and corrections stay: `topics rename <topic> <name>`,
|
|
174
|
+
`topics merge <topic> <into>`, `topics move <session id...> --to <topic or new name>` (`--out`
|
|
175
|
+
takes sessions out, to be sorted again). A topic is named by its id, its name, or a part of the
|
|
176
|
+
name only one has. `topics show <topic>` lists its sessions with their ids, and `--json` prints
|
|
177
|
+
`{"topics":[{"id","name","agents","projects","updatedAt","asking","active","sessions":[...]}],"unsorted":n}`.
|
|
178
|
+
A sorting run's own session is left out of every listing, like other
|
|
179
|
+
[sessions nobody typed into](agents.md#sessions-across-agents).
|
|
121
180
|
|
|
122
181
|
`npx genaicode sessions archive --stale` puts away every session left waiting or half way for 2
|
|
123
182
|
days or more (`sessions archive <id>` for one, `unarchive` to bring it back, `--archived` to list
|
|
@@ -261,9 +320,12 @@ An app that starts `genaicode ui` itself can add `--exit-with-stdin` and stop it
|
|
|
261
320
|
its stdin, which also reaches the server when a launcher such as npx sits in between.
|
|
262
321
|
|
|
263
322
|
`genaicode ui` listens on the loopback address and prints a link with a random access
|
|
264
|
-
token.
|
|
265
|
-
|
|
266
|
-
|
|
323
|
+
token. Opening it in a browser moves the token into origin-isolated `sessionStorage`, strips
|
|
324
|
+
`?token=` from the address bar, and sets a reload-only `HttpOnly` cookie so refreshing the
|
|
325
|
+
tab still serves the empty shell while every API request and the event stream keep requiring
|
|
326
|
+
the token in the `x-genaicode-token` header; only the bundled script and the four mascot
|
|
327
|
+
files (`/app.js`, `/assets/…`) are served without it, since they hold nothing about your
|
|
328
|
+
session. Both front ends are bundled into the package and load
|
|
267
329
|
only for these commands; importing `genaicode` or `genaicode/agents` pulls in no UI code.
|
|
268
330
|
In code, the same conversation is `createAgentSession` from `genaicode/agents` (see
|
|
269
331
|
[agents.md](agents.md#chat-sessions)); with a live driver (`codexLive`,
|
package/package.json
CHANGED
|
@@ -1,6 +1,6 @@
|
|
|
1
1
|
{
|
|
2
2
|
"name": "genaicode",
|
|
3
|
-
"version": "2.
|
|
3
|
+
"version": "2.44.0",
|
|
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": {
|