docs(fill): typed-input privacy stops at agent-device surfaces

Merged
TH
13h ago
+13-0

Summary

States the boundary the reporter proved in #3261: the privacy controls for typed input (fill --text-stdin, --record-as) keep the literal out of agent-device's own responses, diagnostics, and recorded scripts, but XCTest writes its own diagnostic logs and .xcresult bundle on iOS — on a simulator and on a physical device — and typed text can appear there. Per the maintainer decision on the issue, this is documentation only; prevention stays open research, so this uses Refs, not Fixes.

Touched files: 4 (one help line, its pinning test, and two docs pages). Same sentence in all three places. No setting is claimed to prevent it.

The privacy controls for typed input (--text-stdin and --record-as) keep the literal out of
agent-device's own responses, diagnostics, and recorded scripts. They do not rewrite what the
platform records: on iOS, XCTest writes its own diagnostic logs and its .xcresult bundle, and
typed text can appear in both, on a simulator or a physical device.

Validation

Docs and help only, so no runtime behavior changes and no device run applies.

  • pnpm check:affected --run --head 293c63e617c02977613577356fdc80126b373c43 — all runnable checks passed (format, lint, typecheck, layering, fallow, build, vitest-related, command-docs, docs-links, docs-prose).
  • Help-text pin (src/commands/schema/cli-help-topics.test.ts): fails on base (cli-help.ts reverted → AssertionError: The input did not match /keep the literal out of agent-device's own responses, diagnostics, and recorded scripts/), passes on head.
  • Website build and check:doc-links ran clean locally for the new /docs/security-trust link.

Unverified: whether a runner setting can keep the value out of the native logs (the issue's open research), and the same wording's fit for tvOS/visionOS/macOS targets, which this change deliberately does not name.

🤖 Generated with Claude Code