Run panel: auto-approve operator-invoked tools

[RhysSullivan]

Invoking a tool through the Run/Test panel paused on approval-gated tools and dead-ended on a "This tool requires approval (a policy gates it)" message, with no way to actually run it from the panel. But the operator clicking Run is itself the approval, so the panel should just run the tool.

What changed

autoApprove is threaded through the execute path. When set, the execution engine runs the inline accept-all handler instead of intercepting the first elicitation as a pause, so an approval-gated tool runs to completion:

• executeWithPause(code, { autoApprove }) in the engine
• POST /executions accepts an optional autoApprove
• the Run panel sends autoApprove: true

Safety is preserved: block policies fail before any elicitation, so this never bypasses a hard block. The MCP host path is unchanged and still pauses for the model to approve.

Before / after

Same tool, same Require approval · npmdl.* policy badge. Before, the panel dead-ends; after, it returns the real result (HTTP 200, real download count).

Before

After

Tests

• New cross-target e2e scenario e2e/scenarios/run-panel-auto-approve.test.ts: drives the same POST /executions endpoint the panel uses against a tool gated by its own requiresApproval annotation. Without autoApprove the call pauses and the side effect does not happen; with autoApprove it runs to completion and the side effect lands. Green against a live selfhost instance.
• New engine unit test in tool-invoker.test.ts: the same eliciting tool that pauses without autoApprove runs straight to completion with it.
• Existing pause/resume suites still green (the default, non-autoApprove path is untouched).

[cloudflare-workers-and-pages]

Deploying with    Cloudflare Workers

The latest updates on your project. Learn more about integrating Git with Workers.

Status	Name	Latest Commit	Preview URL	Updated (UTC)
✅ Deployment successful! View logs	executor-marketing	b0bb4a8	Commit Preview URL Branch Preview URL	Jun 28 2026, 06:28 PM

[cloudflare-workers-and-pages]

Deploying with    Cloudflare Workers

The latest updates on your project. Learn more about integrating Git with Workers.

Status	Name	Latest Commit	Updated (UTC)
✅ Deployment successful! View logs	executor-cloud	b0bb4a8	Jun 28 2026, 06:30 PM

[github-actions]

Cloudflare preview

Torn down — the PR is closed.

[pkg-pr-new]

Open in StackBlitz

@executor-js/cli

npm i https://pkg.pr.new/@executor-js/cli@1183

@executor-js/config

npm i https://pkg.pr.new/@executor-js/config@1183

@executor-js/execution

npm i https://pkg.pr.new/@executor-js/execution@1183

@executor-js/sdk

npm i https://pkg.pr.new/@executor-js/sdk@1183

@executor-js/plugin-file-secrets

npm i https://pkg.pr.new/@executor-js/plugin-file-secrets@1183

@executor-js/plugin-graphql

npm i https://pkg.pr.new/@executor-js/plugin-graphql@1183

@executor-js/plugin-keychain

npm i https://pkg.pr.new/@executor-js/plugin-keychain@1183

@executor-js/plugin-mcp

npm i https://pkg.pr.new/@executor-js/plugin-mcp@1183

@executor-js/plugin-onepassword

npm i https://pkg.pr.new/@executor-js/plugin-onepassword@1183

@executor-js/plugin-openapi

npm i https://pkg.pr.new/@executor-js/plugin-openapi@1183

@executor-js/codemode-core

npm i https://pkg.pr.new/@executor-js/codemode-core@1183

@executor-js/runtime-quickjs

npm i https://pkg.pr.new/@executor-js/runtime-quickjs@1183

executor

npm i https://pkg.pr.new/executor@1183

commit: b0bb4a8

[greptile-apps]

Greptile Summary

This PR threads an autoApprove flag from the Run/Test panel through the HTTP API into the execution engine, so clicking Run is treated as the human approval rather than pausing on a requiresApproval-annotated tool.

• Adds autoApprove?: boolean to the ExecuteRequest schema and handler, routes it to executeWithPause, and inside the engine the autoApprove path short-circuits the pause-queue entirely by calling runInlineExecution with an acceptAllHandler (() => Effect.succeed({ action: "accept" })).
• block policies are unaffected because they reject before any elicitation fires; the new path is strictly narrower than bypassing a block.
• New unit test verifies the same eliciting tool that pauses without the flag completes straight through with it; new e2e scenario drives the HTTP endpoint end-to-end and asserts the side effect (policy write) lands only on the auto-approved call.

Confidence Score: 4/5

Safe to merge. The core execution change is a clean short-circuit: autoApprove routes through the existing inline path with an accept-all handler, leaving the pause-queue path completely untouched. Block policies fire before any elicitation and are unaffected.

The engine and API changes are correct and well-tested, with both a unit test and an e2e scenario proving the before/after contract. The only loose end is in tool-run-panel.tsx: the paused branch and its UI message were not updated to match the new reality where autoApprove: true is always sent, leaving stale dead code with misleading advice. This does not affect runtime behavior under normal conditions but would give operators wrong guidance in an unexpected edge case.

packages/react/src/components/tool-run-panel.tsx — the paused result branch and its UI message should be updated or removed.

Important Files Changed

Filename	Overview
packages/core/execution/src/engine.ts	Adds acceptAllHandler and an early-return branch in startPausableExecution that routes through runInlineExecution when autoApprove is set, bypassing the pause queue entirely. Logic is clean, the typed error channel is preserved, and block policies still fire before elicitation.
packages/core/api/src/executions/api.ts	Adds autoApprove: Schema.optional(Schema.Boolean) to ExecuteRequest. Minimal, correct schema change with a clear comment.
packages/react/src/components/tool-run-panel.tsx	Adds autoApprove: true to the execute payload. The existing paused result branch is now dead code (the server never returns paused when autoApprove: true is sent), but the stale UI message in that branch remains.
e2e/scenarios/run-panel-auto-approve.test.ts	New cross-target e2e scenario that drives the HTTP endpoint directly, proves the tool pauses without autoApprove and completes with it, and cleans up via Effect.ensuring. Well-structured and reads as a spec.
packages/core/execution/src/tool-invoker.test.ts	Adds a unit test that confirms the same eliciting tool that pauses without autoApprove runs to completion with it. Good complementary coverage to the e2e test.

Sequence Diagram

%%{init: {'theme': 'neutral'}}%%
sequenceDiagram
    participant Panel as ToolRunPanel (React)
    participant API as POST /executions
    participant Engine as ExecutionEngine
    participant Invoker as ToolInvoker
    participant Tool as Approval-gated Tool

    Panel->>API: "{ code, autoApprove: true }"
    API->>Engine: "executeWithPause(code, { autoApprove: true })"
    Note over Engine: autoApprove branch: skip pause queue
    Engine->>Engine: runInlineExecution(code, acceptAllHandler)
    Engine->>Invoker: "execute(code, { onElicitation: acceptAllHandler })"
    Invoker->>Tool: invoke tool
    Tool-->>Invoker: ElicitationRequest (requiresApproval)
    Invoker->>Engine: acceptAllHandler(ctx)
    Engine-->>Invoker: "{ action: "accept" }"
    Tool-->>Invoker: tool result
    Invoker-->>Engine: ExecuteResult
    Engine-->>API: "{ status: "completed", result }"
    API-->>Panel: "{ status: "completed", text, structured, isError }"

Loading

%%{init: {'theme': 'base', 'themeVariables': {"darkMode": true, "background": "#0d1117", "primaryColor": "#21262d", "primaryTextColor": "#e6edf3", "primaryBorderColor": "#8b949e", "lineColor": "#8b949e", "textColor": "#e6edf3", "edgeLabelBackground": "#161b22", "actorBkg": "#21262d", "actorBorder": "#8b949e", "actorTextColor": "#e6edf3", "actorLineColor": "#8b949e", "signalColor": "#8b949e", "signalTextColor": "#e6edf3", "noteBkgColor": "#373320", "noteBorderColor": "#d4a72c", "noteTextColor": "#f0e6c0", "labelBoxBkgColor": "#21262d", "labelBoxBorderColor": "#8b949e", "labelTextColor": "#e6edf3", "loopTextColor": "#e6edf3", "activationBkgColor": "#30363d", "activationBorderColor": "#8b949e"}}}%%
sequenceDiagram
    participant Panel as ToolRunPanel (React)
    participant API as POST /executions
    participant Engine as ExecutionEngine
    participant Invoker as ToolInvoker
    participant Tool as Approval-gated Tool

    Panel->>API: "{ code, autoApprove: true }"
    API->>Engine: "executeWithPause(code, { autoApprove: true })"
    Note over Engine: autoApprove branch: skip pause queue
    Engine->>Engine: runInlineExecution(code, acceptAllHandler)
    Engine->>Invoker: "execute(code, { onElicitation: acceptAllHandler })"
    Invoker->>Tool: invoke tool
    Tool-->>Invoker: ElicitationRequest (requiresApproval)
    Invoker->>Engine: acceptAllHandler(ctx)
    Engine-->>Invoker: "{ action: "accept" }"
    Tool-->>Invoker: tool result
    Invoker-->>Engine: ExecuteResult
    Engine-->>API: "{ status: "completed", result }"
    API-->>Panel: "{ status: "completed", text, structured, isError }"

Loading

Comments Outside Diff (2)

1. packages/react/src/components/tool-run-panel.tsx, line 236-245 (link)

   Stale dead-code branch: with autoApprove: true always in the payload, the server's startPausableExecution routes through runInlineExecution and can only ever return status: "completed". The paused branch here is therefore unreachable in normal operation, and if it were somehow hit (e.g. a schema version mismatch strips autoApprove), the message "adjust the policy to run it directly" would be wrong advice — the panel is already sending autoApprove: true, so adjusting a policy would not unblock the call. Either remove the branch or update the message to reflect the new reality.

   Note: If this suggestion doesn't match your team's coding style, reply to this and let me know. I'll remember it for next time!

2. packages/react/src/components/tool-run-panel.tsx, line 374-376 (link)

   The "paused" UI message is now stale. Since the panel always sends autoApprove: true, a requiresApproval-annotated tool runs straight through, and a block-policy tool fails (returning completed with isError: true) rather than pausing. If this message is ever shown it means something unexpected happened, and "adjust the policy" is no longer the right next step.

_{Reviews (1): Last reviewed commit: "Format files flagged by oxfmt" | Re-trigger Greptile}