Skip to content

Add Computer Use web form proof lane#1746

Merged
tpae merged 2 commits into
osaurus-ai:mainfrom
mimeding:codex/computer-use-web-form-proof
Jul 1, 2026
Merged

Add Computer Use web form proof lane#1746
tpae merged 2 commits into
osaurus-ai:mainfrom
mimeding:codex/computer-use-web-form-proof

Conversation

@mimeding

@mimeding mimeding commented Jun 28, 2026

Copy link
Copy Markdown
Contributor

Summary

  • Adds deterministic Computer Use web-form proof coverage for boring-form workflows.
  • Hardens the live Computer Use path so ambiguous same-label targets stop before typing.
  • Masks secure field values from model context, evidence, and value-change markers.
  • Routes resolved text input through element actions instead of coordinate fallback, and proves declined dangerous app launch does not call driver open.

Validation

  • git diff --check
  • OSAURUS_DISABLE_KEYCHAIN_FOR_TESTS=1 OSAURUS_TEST_ROOT=/tmp/osaurus-pr1746-core swift test --package-path Packages/OsaurusCore --filter 'ComputerUseEvidencePackTests|AgentViewTests|ScreenContextDistillerTests|MockMacDriverContractTests'
  • swift test --package-path Packages/OsaurusEvals --filter 'ComputerUseWebFormProofTests|ComputerUseLoopEvalTests'
  • RUN_EVALS=0 STRICT=1 scripts/evals/computer-use-evidence.sh
  • Final narrow rerun: OSAURUS_DISABLE_KEYCHAIN_FOR_TESTS=1 OSAURUS_TEST_ROOT=/tmp/osaurus-pr1746-agentview-final swift test --package-path Packages/OsaurusCore --filter AgentViewTests

Risk

  • This changes Computer Use safety behavior in bounded cases only: ambiguous targets, secure-field rendering, resolved text entry, and declined app-open guardrails.
  • Model-dependent live Computer Use evals are still opt-in; this PR includes deterministic local proof and artifact privacy validation.

@mimeding mimeding marked this pull request as ready for review June 29, 2026 01:09
@mimeding mimeding marked this pull request as draft June 29, 2026 11:44
@mimeding mimeding marked this pull request as ready for review June 30, 2026 04:31
@tpae tpae merged commit adfc50a into osaurus-ai:main Jul 1, 2026
5 checks passed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants