BitSquare/Codeman
2
0
Fork 0
mirror of https://github.com/Ark0N/Codeman.git synced 2026-10-06 07:29:42 +02:00
Code Issues Packages Projects Releases Wiki Activity

fix(sessions): scope the launch model to claude and pin it with the advisor (#514, #515, #530 landing)

Browse Source 
Maintainer merge-time fixes for the three PRs that landed together on the
session create / launch / persistence path.

#514 findings (bot verdict merge-with-fixes):
- minor, fixed: SessionState.model was published and persisted for every
  mode, so a codex/opencode cron session reported the app-wide Claude
  default it never ran on. toState() now emits it only where the new
  cliTakesSessionModel() holds (registry capability model.source ===
  'claude-settings-file', no CLI id branch). POST /api/sessions uses the
  same helper for its non-claude refusal, so refusal and publication cannot
  drift. Recovery then hands back undefined for other modes on its own.
- nit, fixed: the `model` schema admitted a leading dash (and '.', '[').
  The first character must now be a letter or digit; still a subset of the
  registry's model-claude pattern, so nothing accepted is refused at launch.
- nit, fixed (reject, the consistent choice): `model` with
  attachRemoteSession was silently dropped. Now a 400 INVALID_INPUT, as
  #514 does for non-claude CLIs and quick-start does for remote cases.
  advisorModel (#530) gets the same refusal there. effort and envOverrides
  keep their older silent ignore on that branch so no existing caller breaks.

#515 finding (bot verdict merge, one nit):
- nit, fixed: the types/session.ts @fileoverview described CodexConfig as
  (model, resumeSessionId); it now lists reasoningEffort, bypass,
  animations and renderMode too.

Audit of the merged combination (not reviewed before):
- The conflict resolutions in session.ts (toState), types/session.ts,
  reboot-restore-routes.ts, server.ts (restoreMuxSessions), CLAUDE.md and
  skills/codeman/reference/endpoints.md (+ plugin mirror) keep both sides
  correctly; nothing was lost or doubled.
- A claude session with both `model` and `advisorModel` launches with
  `--model <id>` and ONE merged `--settings` JSON (ultracode + advisorModel,
  or advisorModel beside `--effort <level>`), on the tmux template
  (including the resume || new variant and with the statusLine exporter)
  and on the direct-PTY fallback. Both values (and effort) survive
  restoreMuxSessions onto a dead pane, a reboot restore into a fresh pane,
  and restartCli/dead-pane respawn via _buildRespawnPaneOptions.
- quick-start and ralph-loop take no per-session `model` (matching #514's
  scope, POST /api/sessions only) and launch on the app-wide default, which
  toState now persists for claude, so recovery stays consistent.
- No defect found in the combination beyond the findings above. Noted, not
  changed: advisorModel is still published for any mode a caller sends it
  with (launch-inert there; the UI and skill send it for claude only).

Tests: test/session-model-recovery.test.ts pins the pair through both
recovery shapes for effort ultracode/high/none, the recovery constructors'
fields, the tmux-manager builder hop, and the codex/opencode/shell
non-publication; test/advisor-model.test.ts pins the launch lines and a
real direct-PTY Session's pty.spawn argv; the route test covers flag-shaped
models, attach refusals and the published fields. Docs: SessionState.model
docstring, the reboot-restore-registry header, the golden test comment and
the CLAUDE.md model/advisor bullets.

Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
This commit is contained in:
Codeman maintainer
2026-10-04 23:52:41 +02:00
parent 7917273188
commit 06c4c7da16
10 changed files with 365 additions and 18 deletions
Download Patch File Download Diff File Expand all files Collapse all files
src/session.ts
+18 -1
View File
@@ -213,6 +213,21 @@ export function isExternalCliMode(mode: SessionMode): boolean {
return getCli(mode)?.capabilities.external ?? true;
}
/**
* Does this CLI take the top-level session `model` (claude's per-session `--model`)?
*
* Read off the registry's model-source capability: only a `claude-settings-file` CLI
* (claude) launches on that field. Every other CLI takes its model in its own config object
* (`codexConfig.model` and so on), so for them the field is inert, and cron hands the
* app-wide default (always a Claude id) to any CLI that has a model at all. `toState()`
* publishes and persists the field only where this holds, so a codex cron session never
* reports a Claude model it did not run on, and `POST /api/sessions` refuses a `model` for
* any CLI where it does not.
*/
export function cliTakesSessionModel(mode: SessionMode): boolean {
return getCli(mode)?.capabilities.model.source === 'claude-settings-file';
}
/** Display name for a run mode. Falls back to the raw id for an unregistered one. */
function getModeLabel(mode: SessionMode): string {
return getCli(mode)?.label ?? mode;
@@ -1838,7 +1853,9 @@ export class Session extends EventEmitter {
ompConfig: this._ompConfig,
resumeSessionId: this._resumeSessionId,
effort: this._effort,
model: this._model,
// Claude only: for any other CLI `_model` is inert (its model lives in its own config
// object) and may be the app-wide Claude default cron handed it.
model: cliTakesSessionModel(this.mode) ? this._model : undefined,
advisorModel: this._advisorModel,
customModel: this.customModel,
// COD-118: runtime-only — surfaced so the frontend can require explicit user
src/types/session.ts
+4 -2
View File
@@ -12,7 +12,7 @@
* - ClaudeMode — CLI permission mode ('dangerously-skip-permissions' | 'auto' | 'normal' | 'allowedTools')
* - SessionColor — visual differentiation color
* - OpenCodeConfig — OpenCode-specific settings (model, autoAllowTools, continueSession)
* - CodexConfig — Codex (OpenAI CLI)-specific settings (model, resumeSessionId)
* - CodexConfig — Codex (OpenAI CLI)-specific settings (model, reasoningEffort, resumeSessionId, bypass, animations, renderMode)
* - GeminiConfig — Gemini CLI-specific settings (model, approvalMode, resumeSession)
* - AntigravityConfig — Antigravity CLI (agy) settings (model, dangerouslySkipPermissions, resumeConversationId)
* - PiConfig — Pi CLI (pi.dev) settings (model, provider, thinking, resume/continue, project trust)
@@ -833,7 +833,9 @@ export interface SessionState {
* The model the session was LAUNCHED with (`--model`): the caller's per-session `model`, or
* the app-wide default when there was none. Persisted so a recovered session relaunches on
* the same model rather than whatever the default is by then. Not `cliModel`, which is what
* the CLI's banner reports.
* the CLI's banner reports. Claude sessions only (`cliTakesSessionModel()`): every other CLI
* keeps its model in its own config object (`codexConfig.model` and so on), and this is
* absent for them.
*/
model?: string;
/**
src/web/reboot-restore-registry.ts
+2 -1
View File
@@ -18,7 +18,8 @@
* session record involved. A dropped plan therefore returns the user to
* resuming by hand, one at a time, which is where they are without this
* feature. What the plan held that a transcript does not is the owner, the
* name, the env overrides, the effort, the advisor model and the lineage.
* name, the env overrides, the effort, the model, the advisor model and the
* lineage.
* - Module-level singleton in the style of `web/approval-inbox.ts`: no `Session`
* import and no IO, which keeps it unit-testable and cycle-free.
* - Spending is take-then-build: `take()` removes entries synchronously, before
src/web/routes/session-routes.ts
+18 -2
View File
@@ -30,7 +30,13 @@ import {
type OmpConfig,
type RemoteHost,
} from '../../types.js';
import { Session, isAltScreenStripMode, isExternalCliMode, isMuxAltScreenOnlyStripMode } from '../../session.js';
import {
Session,
cliTakesSessionModel,
isAltScreenStripMode,
isExternalCliMode,
isMuxAltScreenOnlyStripMode,
} from '../../session.js';
import type { PaneCaptureOptions } from '../../mux-interface.js';
import { SseEvent } from '../sse-events.js';
import { webviewCapabilities } from '../../webview-capabilities.js';
@@ -892,12 +898,22 @@ export function registerSessionRoutes(
// The top-level `model` is Claude's per-session `--model`. Every other CLI takes its model
// in its own config object (`codexConfig.model` and so on), so a `model` here would be
// dropped without a word; refuse it before anything is written for the session.
if (body.model && getCli(body.mode ?? 'claude')?.capabilities.model.source !== 'claude-settings-file') {
if (body.model && !cliTakesSessionModel(body.mode ?? 'claude')) {
return createErrorResponse(
ApiErrorCode.INVALID_INPUT,
'model applies to claude sessions only; other CLIs take their model in their own config object, such as codexConfig.model'
);
}
// An attach launches nothing (the remote agent is already running), so a launch model
// or advisor would be dropped the same way, so both are refused, as they have been since
// they were added. The older launch fields (effort, envOverrides) predate this and keep
// their silent ignore here, since refusing them now would break existing callers.
if (body.attachRemoteSession && (body.model || body.advisorModel)) {
return createErrorResponse(
ApiErrorCode.INVALID_INPUT,
'model and advisorModel shape a new launch, and attachRemoteSession launches nothing; leave them out when attaching'
);
}
let workingDir = body.workingDir || process.cwd();
let remote = undefined;
src/web/schemas.ts
+7 -5
View File
@@ -548,15 +548,17 @@ export const CreateSessionSchema = z.object({
modelOverride: z.string().max(50).optional(),
/**
* Claude model for THIS session only, passed as `claude --model <id>`; nothing is written to
* disk. Wins over the app-wide default model. Same character set as the registry's
* `model-claude` pattern, so a value accepted here is never rejected at launch. An empty
* string means no per-session model, as it does for `modelOverride`. Claude only: the route
* refuses it for any other CLI.
* disk. Wins over the app-wide default model. A subset of the registry's `model-claude`
* pattern, so a value accepted here is never rejected at launch. The first character must be
* a letter or digit: the value lands in argv, and no model id opens with `-`, so a
* flag-shaped value is refused here rather than left to the launch quoting. An empty string
* means no per-session model, as it does for `modelOverride`. Claude only: the route refuses
* it for any other CLI and on a remote attach.
*/
model: z
.string()
.max(100)
.regex(/^[a-zA-Z0-9._\-[\]]+$/)
.regex(/^[a-zA-Z0-9][a-zA-Z0-9._\-[\]]*$/)
.or(z.literal(''))
.optional(),
openCodeConfig: OpenCodeConfigSchema,