[Mukesh Dua (mukesh-dua)]

Summary

Align both Discovery deployment skills with the GA 2026-06-01 API while preserving explicit preview configurations, disable agent and tool confirmations in deployed agents, and fix validation conversation naming so Studio can select the session history and output assets. Also repair the CP2K 2024.2 image build by installing binutils, pinning the compatible LibXSMM release, and exposing its pkg-config files.

Type of change

• New agent (under agents/<name>/)
• Update to an existing agent
• New starter kit (under starter-kits/<name>/)
• Update to an existing starter kit
• Schema change (docs/schemas/**)
• Workflow / script change (.github/**)
• Documentation only
• Other (describe below)

Related issue / tracking

n/a

Schema impact

• No schema changes
• Schema changes are backward compatible
• Schema changes are breaking — add the breaking-change label and link the rollout plan below

Validation checklist

• Documentation (README.md, schemas) is updated to match the change.
• No hand-edits to .auto-registry/** (it is generated post-merge).
• No OS / editor artefacts committed (.DS_Store, .env, *.swp, .idea/, .vs/, etc.).
• If model-weight files are added, they are Git-LFS tracked, ≤5 GB each, and in an allowed format.
• No secrets, customer data, or internal-only URLs are committed.
• All Markdown links resolve.

Validation performed after merging the latest upstream main:

• python -m pytest .github\tests\test_agent_deployer_validation.py .github\tests\test_deployer_api_contracts.py -v -ra — passed: 38 tests and 41 subtests.
• git diff --check — passed.
• PowerShell parser checks using [System.Management.Automation.Language.Parser]::ParseFile(...) for both deployment runners — passed.
• Inline Python checks of both skill frontmatters and ast.parse(...) on eight embedded Python blocks — passed.
• Inline Python existence checks for relative Markdown link targets in the changed documentation — passed. External links and anchor targets were not exhaustively checked, so the all-links checklist remains unchecked.

Earlier live smoke tests of these changes:

• CP2K remote image build and deployment succeeded. An H2 PBE single-point calculation completed with return code 0, converged SCF, and nonempty output files.
• Coconut remote image build and deployment succeeded. Validation executed a real caffeine search, returned five results, and verified output artifacts of 1,940 and 415 bytes through a tool-generated receipt. The runner reported validation passed without confirmation requests.
• Live Studio rendering and independent downloads of the output assets were not verified.

The full repository test suite was not run locally; CI remains responsible for broader checks.

Reviewer notes (optional)

• Existing ignored configuration files and saved run configurations retain their explicitly selected API versions. New configurations default to GA.
• GA validation uses /openai/v1/responses and agent_reference; explicit preview validation uses /openai/responses?api-version=... and agent, including polling and resume.
• Starter-kit knowledge-base bindings use versionless IDs for GA. Preview continues to accept versioned IDs, and invalid or unresolved IDs fail explicitly rather than being silently rewritten.
• Deployed agents persist humanInTheLoop: Disabled and confirmation: Disabled for all bound tools, overriding explicit source settings. Source agent YAML remains unchanged. Nodepool capacity confirmation and unexpected platform-required approvals are not bypassed.
• Initial and retry validation conversations now use the investigation name as their display name. Existing conversations are not renamed. The starter-kit deployer does not create validation conversations.
• CP2K 2024.2 requires the libxsmmext and libxsmmnoblas components supplied by LibXSMM 1.17; its pkg-config files are located directly in the conda environment's lib directory.
• The fork's main was synced with upstream before this PR branch was prepared. Previously merged agent additions are not part of this PR's changes.