fix(agent): overly broad check for secrets protection

[icecrasher321]

Summary

Remove overly broad check for secrets for agent handler. And opaque input clearing.

Type of Change

• Bug fix

Testing

Tested manually

Checklist

• Code follows project style guidelines
• Self-reviewed my changes
• Tests added/updated and passing
• No new warnings introduced
• I confirm that I have read and agree to the terms outlined in the Contributor License Agreement (CLA)

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

1 Skipped Deployment

Project	Deployment	Actions	Updated (UTC)
docs	Skipped		Aug 8, 2026 3:22am

[cursor]

PR Summary

High Risk
Changes core secret provenance, logging, and provider egress boundaries across the block executor, Agent, and Mothership handlers; incorrect registry forking could leak secrets into logs or drop required tracking on errors.

Overview
Narrows resolved-secret handling so URLs, resource IDs, and control fields keep normal request semantics unless a field is proven model-visible. Integration docs and commands drop the centralized opaqueModelInput hard-rejection path and instead describe request.modelInput projection for serialized external model content and authenticated internal routes for raw bytes.

Block executor splits provenance registries: log-facing inputs use resolver projections only, successful outputs use a handler-narrowed registry, and failures/stream errors can use errorResolvedSecretTraceRegistry so echoed credentials in provider errors are tracked without reactivating projected prompts or selectors in block output provenance. Streaming uses a separate diagnostic registry for client onStream error projection.

Agent handler stops treating customToolId, skillId, and response-format names as broad structural secret violations; private selectors are settled for display while lookup still uses resolved values, response-format names are aliased for the provider, and tool-result provenance can persist without reactivating private selectors. Mothership applies similar skill-selector display settlement and separates error vs result provenance; conversation IDs are not treated as secret-bearing output content.

Generic tool handler no longer folds opaqueModelInput into its provenance path collection. Pi Babysit reduces blanket scrubPiSecrets on round prompts, diagnostics, and changed-file lists (scrubbing remains on selected error paths).

^{Reviewed by Cursor Bugbot for commit f09556c. Configure here.}

[greptile-apps]

Greptile Summary

The PR narrows resolved-secret protection to fields that actually cross model, persistence, or diagnostic boundaries while preserving ordinary opaque inputs and internal lookup identifiers.

• Refactors Agent and Mothership input projection and provenance handling.
• Updates executor display/error projection and streaming provenance propagation.
• Removes broad opaque-input rejection from ordinary tool inputs and adds focused regression coverage.
• Synchronizes integration authoring and validation guidance with the revised boundary model.

Confidence Score: 5/5

The PR appears safe to merge.

No blocking failure remains; the previously reported Mothership identifier disclosure was invalidated because the authenticated internal route resolves the identifier for lookup and excludes it from the external model payload.

Important Files Changed

Filename	Overview
apps/sim/executor/handlers/agent/agent-handler.ts	Refines model-visible projection, private selector settlement, provider diagnostics, tool provenance, and memory persistence boundaries.
apps/sim/executor/handlers/mothership/mothership-handler.ts	Preserves private skill IDs for authenticated internal lookup while projecting only model-visible labels and metadata.
apps/sim/executor/execution/block-executor.ts	Projects resolver-recorded secrets from displayed inputs and carries dedicated provenance through streaming and error paths.
apps/sim/lib/uploads/utils/model-input.ts	Revises upload model-input classification and projection behavior for inline content versus ordinary resource locators.
apps/sim/tools/index.ts	Removes centralized opaque-input rejection behavior that was applied too broadly to ordinary tool request values.

_{Reviews (2): Last reviewed commit: "address comments" | Re-trigger Greptile}

[icecrasher321]

bugbot run

[icecrasher321]

bugbot run

[icecrasher321]

@greptile

[icecrasher321]

bugbot run

[cursor]

✅ Bugbot reviewed your changes and found no new issues!

Comment @cursor review or bugbot run to trigger another review on this PR

^{Reviewed by Cursor Bugbot for commit f09556c. Configure here.}