Skip to content
Merged
Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
50 commits
Select commit Hold shift + click to select a range
65fdf26
Share CLI effects with Tob broker without sharing caller authority
DoodleBears Nov 15, 2026
ba8ec96
Resolve CLI family effects from dispatched arguments and preserve opa…
DoodleBears Nov 15, 2026
9b47608
Include external card settlements and guard completed CLI owner journeys
DoodleBears Nov 15, 2026
72805e0
Declare the landed IM reconnect owner after main integration
DoodleBears Nov 15, 2026
1b77aee
Keep strict CLI option parsing independent of runtime handlers
DoodleBears Nov 15, 2026
92b8175
Show Bot-owned card rules in composer permissions
DoodleBears Nov 15, 2026
adea9ad
chore: align deletion development with published E3 history
DoodleBears Nov 15, 2026
02f62ad
Keep irreversible external tool calls under common approval
Nov 15, 2026
2937f9a
feat(tob): execute typed CLI requests in admitted model-free turns
Nov 15, 2026
662508c
fix(perf): audit every native session compression frame
Nov 15, 2026
d09fdc1
Show the canonical directory for wrapped approval rules
Nov 15, 2026
70ade0a
Confirm template deletion and move affected Bots to global inheritance
Nov 15, 2026
d721d1a
Preserve remote and destructive effects in CLI declarations
Nov 15, 2026
a8e1f1d
Cancel model-free Tob turns during native Session preparation
Nov 15, 2026
4abc715
Reserve native preparation until cancelled publication is cleaned up
Nov 15, 2026
0206917
Integrate effective approval templates and current main into CLI effe…
Nov 15, 2026
5ec6e68
Integrate protected external effects with retained cards and admitted…
Nov 15, 2026
072e362
fix: release setup-owned bridge callbacks with Agent Context
Nov 15, 2026
ea8789b
Integrate Bot card rules with retained approval cards and current main
Nov 15, 2026
6a488e5
Integrate confirmed template deletion with three-part authoring
Nov 15, 2026
07e9e80
Record verified template accessibility fix ancestry
Nov 15, 2026
2e74bcb
Integrate Bot card rules with grouped template authoring
Nov 15, 2026
6d030c2
Integrate grouped guidance cards with confirmed template deletion
Nov 15, 2026
8a2d1e1
Guard real persistence costs for referenced template deletion (#4246)
Nov 15, 2026
08f1fcb
Integrate Bot card-rule source and revocation with confirmed template…
Nov 15, 2026
2aaba24
Integrate approval rules, template deletion and native CLI effect adm…
Nov 15, 2026
031ff90
Keep typed Tob effects in common approval under Full Access (#4210)
Nov 15, 2026
db8c391
Bound native Session audit reads and stop failed performance captures…
Nov 15, 2026
d0efc6e
Share the effects wire schema between Web and mobile (#4210)
Nov 15, 2026
be9d681
Stop native performance capture on coded cancellation refusal (#4212)
Nov 15, 2026
192237d
fix: expose explicit template deletion confirmation through Gateway
Nov 15, 2026
9d0adcf
fix: keep owned Tob CLI approval with the broker
Nov 16, 2026
033d8d9
Merge template retry and recovery fixes into approval integration
Nov 16, 2026
b92c21e
fix: refuse invalid owned CLI calls before transport approval
Nov 16, 2026
a8489d5
fix: close approval stop and clean CI verification gaps
Nov 16, 2026
a541133
fix: retain native approval result and chosen action
Nov 16, 2026
07a4188
fix: keep CI permission fixtures independent and precisely approved
Nov 16, 2026
5f25624
Merge commit '2e67c1f455236282b294ef0bb357c03f35ae577a' into t3/bot-a…
Nov 16, 2026
e799d09
Merge reviewed E3 policy recovery fixes
Nov 16, 2026
607bf88
fix: allow confirmed recovery deletion of unreadable templates
Nov 16, 2026
629bf76
Merge latest main while preserving Bot approval and Skill command con…
Nov 16, 2026
3cb4592
Merge canonical template recovery after partial save failures
Nov 16, 2026
eade34c
Merge main principal governance with Bot approval implementation
Nov 16, 2026
999711b
fix: keep shared Bot approvals in the requesting participant DM
Nov 16, 2026
c2c11e6
Merge Workspace Grant Project Skills from main
Nov 16, 2026
6202f23
fix: preserve approval audiences and native refusal boundaries
Nov 16, 2026
5ceed5a
Merge main Skill profile and Assignment Memory providers
Nov 16, 2026
19c110a
fix(approval): preserve outcomes across concurrent lifecycle changes
Nov 16, 2026
35c1b98
fix(client): preserve the latest approval template read
Nov 16, 2026
f1172fd
test(approval): assert trusted caller context and refresh delivery docs
Nov 16, 2026
File filter

Filter by extension

Filter by extension

Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
6 changes: 6 additions & 0 deletions .agents/skills/dsh-dev/SKILL.md
Original file line number Diff line number Diff line change
Expand Up @@ -125,6 +125,12 @@ The regression uses the real native Session fold and LLM projection in

BotHarness is a plugin layer inside DSH: unit tests do not cover the real boot. Every trap in this skill was invisible to `pnpm test` and only appeared in a running `dsh --profile web-dev`. **Validate against the running shell, not the test suite.**

### Native approval requires a real open turn (#4212)

On pinned DSH 0.2.2-rc.1, scoped Tools.execute supports an actual Agent outside a model step, but ApprovalService.request rejects outside an open native turn before appending approval/asked. A successful model-free read therefore does not prove protected-write approval parity. The public Agent.runMaintenance API is explicitly non-turn; do not append manual turn/start events to bypass the approval audit check or desynchronize the loop counter.

For an admitted application-owned request, keep the canonical BotRuntime owner and exact live Agent. Queue one identified native input, use the supported agent/pre-step waterfall after session admission to execute the scoped Tool, then reject model-step entry. The native loop owns turn/start and turn/end; approval/asked and approval/decided remain genuinely enclosed, and no model request is made. Validate the exact claimed native input, cancellation, no queued foreign work, and all original source/permission fences. Keep a model-free request marker on its genuine Human DM so normal retry or continuation cannot replay an expired execution intent.

## Proactive Client verification

After every isolated launch, Client reload and relevant UI action, read the real browser console and DOM yourself before accepting the result; do not wait for Human-copied errors. Read [the Client diagnostic loop](../../../docs/dev/guides/client-startup-diagnostics.md) for the agent-runnable reader, bounded attempt history, background-tab/freshness limits and native RC1 reproductions. Treat API health, Client shell evidence and browser-control failures as separate signals. Preserve first failure and each retry; recovery by reload does not establish a fix.
Expand Down
1 change: 1 addition & 0 deletions .agents/skills/dsh-plugin-dev/CHANGELOG.md
Original file line number Diff line number Diff line change
Expand Up @@ -10,6 +10,7 @@ Preparing the next DSH Skill release independently from downstream product relea

### Documentation

- Documented the downstream model-free Tob CLI journey and pinned native approval audit boundary; stable DSH/Cordis Context, Decision Tree and upstream provenance are unchanged ([#4212](https://github.com/BotHarness/DeepSeekBot/issues/4212)).
- Added the downstream anonymous Bot token telemetry verification journey and snapshot query contract; stable DSH/Cordis Context, Decision Tree and upstream qualification are unchanged ([#4255](https://github.com/BotHarness/DeepSeekBot/issues/4255)).

- Added the downstream first-Tob settings responsiveness journey; stable DSH/Cordis Context, Decision Tree and upstream qualification are unchanged ([#4226](https://github.com/BotHarness/DeepSeekBot/issues/4226)).
Expand Down
1 change: 1 addition & 0 deletions .agents/skills/dsh-plugin-dev/CHANGELOG.zh.md
Original file line number Diff line number Diff line change
Expand Up @@ -9,6 +9,7 @@ artifact;DSH 版本与上游 revision 记录这些内容是基于什么版本

### Documentation

- 补充下游无模型 Tob CLI 验证路径及原生审批的真实回合审计边界;稳定 DSH/Cordis Context、Decision Tree 及上游验证来源不变([#4212](https://github.com/BotHarness/DeepSeekBot/issues/4212))。
- 增加下游匿名 Bot Token 统计的验证路径及快照查询约定;稳定 DSH/Cordis Context、Decision Tree 及上游验证来源不变([#4255](https://github.com/BotHarness/DeepSeekBot/issues/4255))。

- 增加下游首次 Tob 准备期间的设置响应性验证路径;稳定 DSH/Cordis Context、Decision Tree 及上游验证来源不变([#4226](https://github.com/BotHarness/DeepSeekBot/issues/4226))。
Expand Down
2 changes: 1 addition & 1 deletion .agents/skills/verify-deepseekbot/features/README.md
Original file line number Diff line number Diff line change
Expand Up @@ -27,7 +27,7 @@
| DeepSeekBot device-local appearance | no | `scripts/e2e-local-appearance.mjs` | [local-appearance.md](local-appearance.md) |
| Functional Panel appearance | no | `scripts/e2e-panel-appearance.mjs`, `scripts/e2e-nested-appearance.mjs` | [panel-appearance.md](panel-appearance.md) |
| Global panel materials | no | `scripts/e2e-global-materials.mjs` | [global-materials.md](global-materials.md) |
| Protected Tob DM and mobile pull-to-Tob (management, models, runtime, privacy and updates) | yes for reply and approval | `scripts/e2e-tob-home.mjs`, `scripts/e2e-tob-model-runtime.mjs`, `scripts/e2e-tob-bot-management.mjs`, `scripts/e2e-tob-channel-management.mjs`, `scripts/e2e-tob-group-settings.mjs`, `scripts/e2e-tob-group-decisions.mjs`, `scripts/e2e-tob-delegation.mjs` | [tob-home.md](tob-home.md) |
| Protected Tob DM and mobile pull-to-Tob (management, models, runtime, privacy and updates), plus typed CLI admission | yes for model reply; no for typed CLI approval | `scripts/e2e-tob-home.mjs`, `scripts/e2e-tob-model-runtime.mjs`, `scripts/e2e-tob-bot-management.mjs`, `scripts/e2e-tob-channel-management.mjs`, `scripts/e2e-tob-group-settings.mjs`, `scripts/e2e-tob-group-decisions.mjs`, `scripts/e2e-tob-delegation.mjs` | [tob-home.md](tob-home.md) |
| DSH Session list Bot folders | for fresh execution | `scripts/e2e-bot-session-folders.mjs` | [bot-session-folders.md](bot-session-folders.md) |

One file per user-facing feature. Each file has four sections: `Sub-features`, `How to get to it (user POV)`, `Driving it with verify-app`, and `Gotchas`.
Expand Down
Original file line number Diff line number Diff line change
Expand Up @@ -10,6 +10,7 @@ A Bot inherits the optional global default, selects one complete shared template
- `select`: Channel settings → Auto approval, or composer Workspace Grants panel → Auto approval, uses the same DSH Select. Inheritance shows its effective global source. Specific selection replaces global; Off retains deterministic rules.
- `mobile`: Bot profile → Settings → Permissions → Workspace grants → Auto approval template, inside the existing iOS sheet / Android `ModalBottomSheet`. Inherit global shows the resolved global template (or Off); selecting Off then Inherit global writes null then @global through the same Host owner. Legacy Hosts without a global-selection field retain their supported choices.
- `reviewer`: a covered call leaves an auto-approved record with template identity/version; an uncovered or failed review posts a Human card with the reason and captured version when available. No new model rejection or rejection pause occurs.
- `delete`: custom template → Delete opens a read-only impact preview. It lists visible direct/global users plus protected private-reference count and the resulting global template/Off/unavailable state. Confirming its revision switches direct users to inherit and clears a deleted global default. Stale references/policy/content require a fresh preview and another confirmation. Failed reads do not delete or change choices.
- `guard`: effective template/global/Bot policy changes bypass Full Access and model review, retaining Human or explicit management-scope authorization and execution fences (ADR-0249).

## How to get to it (user POV)
Expand All @@ -25,7 +26,8 @@ Use an isolated Host and its own authenticated CLI. Confirm health and Client co
3. New Bot creation reads back @global; a retained legacy missing field remains Off. Restart verifies persistence without schema changes.
4. Trigger an existing approval through the native tool path. A deterministic saved rule makes zero Reviewer calls. An uncovered call uses the effective template and leaves either the versioned auto-approved record or Human card; answer that card through the normal owner.
5. Edit/switch during an injected Reviewer wait at the Channel Tool Approval seam: this Review retains the old identity/version, the next uses the new one. Missing/unreadable template or model timeout/malformed response posts Human with a reason, even if diagnostics fail. Check canonical Channel records and developer logs.
6. Exercise a policy-management approval under Full Access; Reviewer calls remain zero and only Human or an explicit management rule can authorize effects. Revoke the rule before execution and verify refusal.
6. Create a custom template and an explicit user, an inherited user, another specific user and an Off control. Preview deletion through CLI and Client without mutation; confirm after checking all effects. With another global default, direct users inherit it. Delete the global default: clear it and leave inheritors inherited with Reviewer off. Change refs/content/global between preview and confirmation: refusal and refreshed impact, no replay. Keep an existing Human review pending across deletion and verify its captured template/version on settlement. Saved rules and Full Access do not change.
7. Exercise a policy-management approval under Full Access; Reviewer calls remain zero and only Human or an explicit management rule can authorize effects. Revoke the rule before execution and verify refusal.

## Native mobile qualification

Expand All @@ -34,7 +36,7 @@ Use the shared device QA lane and T3 device tools. The same synthetic Bot and is
## Gotchas

- `/tmp` may already be writable; an ordinary file write there need not reach an approval gate.
- Card-rule table and confirmed bulk deletion fallback remain separate issues. For the editor, verify legacy Markdown survives conversion and an edit during a Review changes only the next captured version.
- Three-part editing, confirmed bulk deletion and the card-rule table have distinct journeys. Verify legacy Markdown conversion, next-Review snapshot behavior and deletion fallback separately.
- Browser screenshots establish Web behavior only. Preserve failed drives and observe before retrying.

## Human result retention (#4211)
Expand Down
5 changes: 5 additions & 0 deletions .agents/skills/verify-deepseekbot/features/tob-home.md
Original file line number Diff line number Diff line change
Expand Up @@ -139,3 +139,8 @@ Use the Mac/iOS or Windows/Android device queue; Web and model tests do not qual
- The top entrance is the same chat content and composer as the normal Tob row. Check its down-entry/up-exit, Back, iOS edge cancellation/commit, keyboard-open exit and dark flash. Toggle Reduce Motion while the app remains running: on returning, entry and exit must stop sliding without a JS reload. Test normal Tob-row push separately. Keep physical/Simulator/Android and source SHA results separate.

Retain the first failure with native screenshot/AX/video and exact source/binary identity, then restore preferences, remove only this run's temporary pairing, revoke its Host key and stop owned resources. The full device matrix is tracked in #3090 and #3093.


## Model-free typed CLI parity (#4212)

Use the authenticated Owner `tob-cli start --args-stdin` with private JSON `{argv,stdin}`; read its request ID with `tob-cli status` and cancel with `tob-cli cancel`. This creates a genuine Human DM and accepted Admission, prepares the actual owned Tob root Agent and executes the typed `deepseekbot_cli` through native Tools.execute without a model. A historical source or temporary direct Session cannot substitute. For a local nondestructive fixture, request an approval-template selection, inspect the real pending card in the Tob DM (exact argv and stdin size/hash), decide with existing Human `tool-approval-decide`, and read both the result and selected template. Match source/build hashes; no model request should appear. Repeat denial, cancellation, stale revision and authority-revoke fences. A template readback alone is not broker approval proof. Capture actual Client light/dark pending/settled states separately; Web evidence does not qualify native mobile.
10 changes: 10 additions & 0 deletions .agents/skills/verify-deepseekbot/features/tool-approval-simple.md
Original file line number Diff line number Diff line change
Expand Up @@ -30,6 +30,12 @@ This Windows native-pwsh journey creates a synthetic Bot, uses real model calls,

For a baseline capture, provide `--capture-report <private-json> --expect-legacy`. The private record identifies an existing task-owned instance, Bot and pending request. This mode reads and captures only; it does not decide or create a Bot. Never use a Human's real request as a disposable verification input.

## Bot-specific saved card rules (#4245)

Open Bot permissions from the composer or the folder/access entry. After saving Always allow on a real card, open **Always allow · this Bot**. Check **From approval card**, tool, execution role, current directory and full scope under View scope against `tool-approval-rules`. Revoke that row and verify canonical readback shows it disabled. Repeat through the online CLI, then Reload the open list. The next same call must return to review. A second Bot using the same template and a template copy must acquire no rule. Model allow-once must not add a rule.

This list reads the existing canonical owner once during permissions refresh; it adds no polling, template lookup or Reviewer work to tool execution. The focused Client test guards one initial rules read and canonical readback after revocation. `auto-approval-breaker.test.ts` guards zero template/Reviewer calls on a saved-rule hit, restored review after revoke, and no new rule from model allow-once. Runtime before/after latency and screenshots remain separate evidence.

## Gotchas

- DSH native pwsh uses `sandbox_permissions: danger-full-access`, not Codex's `require_escalated`. A harmless output marker explicitly requesting that mode can exercise the real ask without touching a sensitive file.
Expand All @@ -51,3 +57,7 @@ For first-time setup, the homepage instruction names the actual Allow once butto
## Companion reading space

With a visible floating companion, scroll to the latest approval and reply and verify their bottom edge is above the passive avatar and toolbar. The running status above the composer must also remain clear of the avatar. Repeat at 375px width, with reduced motion, and after activity fades; removing every companion returns the space, while toggling companions during history reading must not jump to the latest message. This checks the latest content at the bottom, not arbitrary history positions or intentionally opened companion bubbles/menus.

## Irreversible external effects (#4210)

Use online `bot-tool` with a declared irreversible external tool and a controlled destination-free fixture. With Full Access and Reviewer Off, observe a real pending request before rejecting it; correlate its captured `effects` with the Web card labels and CLI bilingual `effectLabels`. Do not insert cards or approval rows. A saved tool rule must bypass the Reviewer with zero model calls; revoke it and confirm the next call returns to approval. Input or result text claiming approval has no authority. Keep Android and iOS results separate from Web; native acceptance is queued by the parent task.
8 changes: 7 additions & 1 deletion apps/mobile/src/host/requests.ts
Original file line number Diff line number Diff line change
@@ -1,3 +1,4 @@
import { parseToolEffects } from '../../../../packages/core/src/workspaces/tool-effects-contract';
import { approvalShellCommand } from '../../../../packages/client/src/client/approval-diff';
import {
approveWorkspaceRequest,
Expand All @@ -16,6 +17,7 @@ import type {
TimelineMessage,
TimelinePage,
ToolApprovalOutcome,
ToolApprovalRequestCard,
ToolApprovalStatus,
UserQuestionStatus,
} from './wire';
Expand All @@ -30,6 +32,7 @@ export interface RequestRef {
}

export interface ApprovalRequest extends RequestRef {
effects?: ToolApprovalRequestCard['effects'];
toolName: string;
role: 'orchestrator' | 'assignment';
cwd: string;
Expand All @@ -43,7 +46,7 @@ export interface ApprovalRequest extends RequestRef {

export type ApprovalContent = Pick<
ApprovalRequest,
'toolName' | 'role' | 'cwd' | 'input' | 'allowAlways'
'toolName' | 'role' | 'cwd' | 'input' | 'allowAlways' | 'effects'
>;

export function approvalContent(value: unknown): ApprovalContent | undefined {
Expand All @@ -55,8 +58,10 @@ export function approvalContent(value: unknown): ApprovalContent | undefined {
if (!('cwd' in value) || typeof value.cwd !== 'string') return undefined;
if (!('input' in value) || typeof value.input !== 'string') return undefined;
if ('allowAlways' in value && typeof value.allowAlways !== 'boolean') return undefined;
const effects = parseToolEffects('effects' in value ? value.effects : undefined);
return {
toolName: value.toolName,
...(effects === undefined ? {} : { effects }),
role: value.role,
cwd: value.cwd,
input: value.input,
Expand Down Expand Up @@ -131,6 +136,7 @@ export async function loadApproval(client: HostClient, ref: RequestRef): Promise
return {
...ref,
toolName: card.toolName,
...(card.effects === undefined ? {} : { effects: card.effects }),
role: card.role,
cwd: card.cwd,
input: card.input,
Expand Down
2 changes: 2 additions & 0 deletions apps/mobile/src/host/wire.ts
Original file line number Diff line number Diff line change
@@ -1,3 +1,4 @@
import type { ToolEffects } from '../../../../packages/core/src/workspaces/tool-effects';
import type { BotPrimaryColor } from '../../../../packages/core/src/bots/primary-color';
import type { ChannelReplyPreviewMedia } from '../../../../packages/core/src/channels/reply-preview';

Expand Down Expand Up @@ -554,6 +555,7 @@ export interface HumanAttentionPage {
}

export interface ToolApprovalRequestCard {
effects?: ToolEffects;
toolName: string;
role: 'orchestrator' | 'assignment';
cwd: string;
Expand Down
5 changes: 5 additions & 0 deletions apps/mobile/src/ui/approval-view.tsx
Original file line number Diff line number Diff line change
@@ -1,3 +1,5 @@
import { toolEffectLabels } from '../../../../packages/core/src/workspaces/tool-effect-labels';
import { locale } from './strings';
import { Text } from './primitives/system-text';
import { useCallback, useRef, useState } from 'react';
import { Platform, ScrollView, View, type ViewInstance } from 'react-native';
Expand Down Expand Up @@ -162,6 +164,9 @@ export function ApprovalView({
<Text style={{ color: theme.ink, fontSize: 22, fontWeight: '700' }}>
{t.wantsToRun(botName, request.toolName)}
</Text>
{request.effects === undefined ? null : (
<Text>{toolEffectLabels(request.effects, locale).join(' · ')}</Text>
)}
<View style={{ gap: 4 }}>
<Text style={{ color: theme.inkSoft, fontSize: 13, fontWeight: '600' }}>
{t.workingFolder}
Expand Down
6 changes: 5 additions & 1 deletion apps/mobile/src/ui/chat/approval-card.tsx
Original file line number Diff line number Diff line change
@@ -1,3 +1,4 @@
import { toolEffectLabels } from '../../../../../packages/core/src/workspaces/tool-effect-labels';
import { useCallback, useRef, useState } from 'react';
import { View, type ViewInstance } from 'react-native';

Expand All @@ -18,7 +19,7 @@ import { useMounted } from '../mounted';
import { Button } from '../primitives/button';
import { errorText } from '../request-parts';
import { haptic } from '../haptics';
import { useStrings, t } from '../strings';
import { useStrings, t, locale } from '../strings';
import { ActionCard, CardButtons, CardNote, CardPreview, DetailRow } from './action-card';
import type { CardLineTone } from './card-line';
import { FileDiff } from './approval-diff';
Expand Down Expand Up @@ -318,6 +319,9 @@ function ApprovalCardContent({
</>
}
>
{request.effects === undefined ? null : (
<CardNote>{toolEffectLabels(request.effects, locale).join(' · ')}</CardNote>
)}
{running === undefined ? null : <CardNote>{running}</CardNote>}
{highRisk(request.toolName) ? <CardNote>{riskNote(request.toolName)}</CardNote> : null}
<CardNote>{alwaysNote(request)}</CardNote>
Expand Down
Loading
Loading