VAREXIS-INDEX ยท Free evidence

Approval recovery and tool access boundaries

Six source-pinned decisions on local approval opt-in, restored-state integrity, partial approval, hosted MCP standing decisions and Playwright shared-context/file-access configuration. Documentation is not live authority, execution verification or new licensed integration support.

Catalogue additions are available free through the exact HTTP JSON links below. Enhanced natural-language discovery is not yet qualified for these records. They are outside MCP search, MCP preview and paid saved retrieval. No search qualification, executable capability or entitlement is added.

Complete exact-ID manifest · Full catalogue · Existing MCP coverage

Local shell and patch approval opt-in

Does setting an OpenAI Agents SDK on_approval callback require approval for local shell or patch calls?

In the pinned Python SDK documentation, local ShellTool and ApplyPatchTool default needs_approval to False. An on_approval callback alone does not enable approval. The callback runs only for calls that require approval and lack an existing decision; configure needs_approval separately.

Documentation reviewed 2026-11-08. Primary source.

Boundaries
  • Do not count a callback declaration as an active approval gate without the effective needs_approval policy and existing decision state.
  • A needs_approval=True call without an existing decision can invoke the callback; not every callback is bypassed.
  • Buyer implementation, live runtime enforcement, settled payment, verified deletion or certification are not established.
  • No expanded capabilities or licences for existing INDEX executables.

Complete record JSON · Exact free preview JSON · Source review

RunState integrity before approval recovery

Do schema validation and context_override authenticate an OpenAI RunState snapshot submitted by a reviewer?

The pinned Python SDK guidance says RunState.from_json and from_string restore execution state without authenticating the snapshot or its submitter. Schema checks, tool-call fingerprints, strict_context, context_override or removing approval records do not make untrusted state safe. The application must verify complete integrity, ownership and replay protection before deserializing a client-carried snapshot.

Documentation reviewed 2026-11-08. Primary source.

Boundaries
  • Use server-owned pending state or application-verified complete snapshots; refuse to infer reviewer authority from IDs, context replacement or a valid schema.
  • A complete integrity-protected snapshot bound to an authorised user/run with replay controls can satisfy application checks; those checks neither encrypt its contents nor prove external success.
  • Buyer implementation, live runtime enforcement, settled payment, verified deletion or certification are not established.
  • No expanded capabilities or licences for existing INDEX executables.

Complete record JSON · Exact free preview JSON · Source review

Shared HTTP browser context boundary

Does Playwright MCP shared-browser-context isolate browser state between connected HTTP clients?

The pinned Playwright MCP README defines --shared-browser-context as reusing the same browser context across all connected HTTP clients. A separate MCP client connection therefore does not establish a separate browser context when this setting is enabled.

Documentation reviewed 2026-11-08. Primary source.

Boundaries
  • Inspect effective launch configuration and browser identity before treating client connections as isolated users; connection count does not establish tenant isolation or live access.
  • The flag does not describe distinct independently operated server processes or separately verified context allocation; not all Playwright clients necessarily share state.
  • Buyer implementation, live runtime enforcement, settled payment, verified deletion or certification are not established.
  • No expanded capabilities or licences for existing INDEX executables.

Complete record JSON · Exact free preview JSON · Source review

Partial approval batch execution boundary

Do unresolved OpenAI Agents SDK approvals prevent every call in the same interruption batch from continuing?

The pinned Python SDK documentation permits resolving only some pending approvals. On resume, resolved calls can continue while unresolved calls remain in interruptions and pause the run again. A remaining approval interruption is not evidence that no call from the batch has progressed.

Documentation reviewed 2026-11-08. Primary source.

Boundaries
  • Track each pending decision and resulting call separately; an interruption list is not an atomic all-or-nothing execution gate or proof of zero side effects.
  • An application can require all decisions before calling Runner; that extra policy is not inferred from the SDK interruption mechanism.
  • Buyer implementation, live runtime enforcement, settled payment, verified deletion or certification are not established.
  • No expanded capabilities or licences for existing INDEX executables.

Complete record JSON · Exact free preview JSON · Source review

Hosted MCP sticky approval identity

Can an OpenAI hosted-MCP always-approve decision transfer to a same-named tool on another server label?

The pinned Python SDK identifies a hosted-MCP sticky decision by the combination of server_label and tool name. An always-approve decision does not approve a same-named tool on another label. A persistent always-approve or always-reject decision requires both identity fields to be non-empty.

Documentation reviewed 2026-11-08. Primary source.

Boundaries
  • Bind standing decisions to the documented server-label/tool tuple; labels do not authenticate the remote operator, version, endpoint or effective permissions.
  • The tuple does not distinguish changed implementations presented under the same label and name, and does not describe local MCP approval identity.
  • Buyer implementation, live runtime enforcement, settled payment, verified deletion or certification are not established.
  • No expanded capabilities or licences for existing INDEX executables.

Complete record JSON · Exact free preview JSON · Source review

Unrestricted file access and workspace roots

Does Playwright MCP keep its default workspace and file-URL restrictions after allow-unrestricted-file-access is enabled?

The pinned README says --allow-unrestricted-file-access permits files outside workspace roots and unrestricted file:// navigation. By default the documented file access is restricted to workspace roots, or the working directory when no roots are configured, and file:// navigation is blocked.

Documentation reviewed 2026-11-08. Primary source.

Boundaries
  • Do not infer default workspace/file-URL restrictions after this flag is enabled. OS permissions, browser sandboxing and other tools require separate evidence.
  • The flag neither grants OS privileges nor proves every path readable. Default roots do not prove an OS sandbox or constrain unrelated tools.
  • Buyer implementation, live runtime enforcement, settled payment, verified deletion or certification are not established.
  • No expanded capabilities or licences for existing INDEX executables.

Complete record JSON · Exact free preview JSON · Source review