chore: version packages

[builder-io-integration]

This PR was opened by the Changesets release GitHub action. When you're ready to do a release, you can merge this and the packages will be published to npm automatically. If you're not ready to do a release yet, that's fine, whenever you add more changesets to main, this PR will be updated.

Releases

@agent-native/core@0.49.26

Patch Changes

• e63c360: Add durable hosted MCP ask_app tasks with ask_app_status polling.

• 8726f38: Teach all app agents that provider shortcut actions are not capability limits and that broad provider searches, joins, classifications, and absence claims should use provider API staging, saved responses, staged-dataset queries, or sandboxed code with explicit coverage reporting. Ensure lean, A2A, and MCP ask-agent registries include run-code when code execution is enabled, and give sandboxed code generic providerRequest/providerFetchAll helpers for broad paginated provider corpus work.

• e63c360: Support provider API pagination cursors sent through JSON request bodies.

• e63c360: Make PR visual recap robust to plan-app deploy-propagation windows. The recap
  CLI ships to npm independently of the plan-app server, so a recap can run after
  the new CLI is live but before the matching action routes have propagated to
  every cold-start server instance:

  • create-visual-recap publish now retries a transient 404 (the route 404s on a
    not-yet-updated instance) instead of failing the recap outright.
  • The live block-reference fetch (get-plan-blocks) now retries transient
    404s/timeouts before falling back to bundled instructions, so the agent
    authors against the current block vocabulary instead of stale tags.

• e63c360: Anchor overlapping code annotation hover cards to the right edge.

@agent-native/dispatch@0.10.4

Patch Changes

• 1d0f069: Add staged provider API responses for Dispatch so broad provider searches can save bounded corpora and reduce them outside chat context.
• 8726f38: Teach all app agents that provider shortcut actions are not capability limits and that broad provider searches, joins, classifications, and absence claims should use provider API staging, saved responses, staged-dataset queries, or sandboxed code with explicit coverage reporting. Ensure lean, A2A, and MCP ask-agent registries include run-code when code execution is enabled, and give sandboxed code generic providerRequest/providerFetchAll helpers for broad paginated provider corpus work.

[github-actions]

Visual recap — skipped

The visual recap job did not run for this pull request. This is informational only and does not block the PR.

Recap skipped for 4aa5d1e: bot author (type=Bot).

[builder-io-integration]

Builder reviewed your changes — looks good ✅

Review Details

Changesets Release Review

This is an automated version bump PR generated by the Changesets release action. The PR consolidates 6 changesets into two package releases:

• @agent-native/core 0.49.25 → 0.49.26 (5 changes: durable MCP ask_app tasks, provider API pagination, PR recap resilience, provider corpus guidance, code tooltip anchoring)
• @agent-native/dispatch 0.10.3 → 0.10.4 (2 changes: staged provider API responses, provider corpus guidance)

Validation results:

• ✅ All 6 changesets properly consumed and deleted
• ✅ CHANGELOG.md entries accurately reflect consumed changesets
• ✅ Version bumps correctly patch-level (semantic versioning)
• ✅ package.json versions match CHANGELOG headings
• ✅ No orphaned changesets or duplicate entries
• ✅ JSON syntax valid

Risk assessment: Low — pure metadata changes with no code impact. The PR is ready for merge and npm publish.

🧪 Browser testing: Skipped — PR only modifies package metadata and CHANGELOGs, no frontend/backend code impact