[Mukesh Dua (mukesh-dua)]

Summary

Add a CPU-only R Statistics prompt agent with a reproducible R 4.5.3 / ggplot2 4.0.0 container, pinned dependencies, synthetic usage example, documentation, and container regression tests. Harden the Discovery agent deployer's validation so confirmation-only replies and connectivity responses cannot be mistaken for successful tool execution.

The deployer now correlates tool calls with outputs, requires measured execution receipts for code environments, detects pending approvals without self-approving, preserves the original prompt on retries, resumes the existing conversation when appropriate, and retains failed investigations and response snapshots for diagnosis.

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 on Windows after pulling the latest fork main:

• python -B -m unittest discover -s .github\tests -p test_agent_deployer_validation.py -v — passed, 21 tests.
• python -B -m unittest discover -s agents\r-statistics-agent\tools\r-statistics -p "test_*.py" -v — passed, 7 Linux-container integration tests against the previously built r-statistics:1.1.0 image.
• $Errors = $null; [System.Management.Automation.Language.Parser]::ParseFile((Join-Path (Get-Location) '.github\skills\discovery-services-agent-deployer\scripts\deploy-discovery-agent.ps1'), [ref]$null, [ref]$Errors) | Out-Null; if ($Errors.Count) { $Errors; exit 1 } — passed, no PowerShell parse errors.
• pwsh -NoProfile -File .github\skills\discovery-services-agent-deployer\scripts\deploy-discovery-agent.ps1 r-statistics-agent -WhatIfPlan — passed.
• python -B .github\scripts\validate_pr.py --changed-files $List --repo-root . --output $Result — passed for all staged files; $List and $Result were temporary files, removed after validation. The changed-file list was populated using git diff --cached --name-only. One nonblocking POL-019 warning: the publisher-email DNS check could not run because dnspython is not installed.
• git diff --cached --check — passed.

The full repository test suite and a complete external Markdown-link check were not run. No model weights or binary output artifacts are included.

Reviewer notes (optional)

• The image runs as UID/GID 10001, uses an immutable base-image digest, installs version-pinned dependencies at build time, and does not install packages at runtime.
• Only the Dockerfile, package manifest, and installer enter the image build context. Tests and synthetic examples remain source-only; edge-case checks are part of the existing test module.
• Human-in-the-loop and tool confirmation remain enabled in the contributed agent definition.
• Code-environment validation now requires a DISCOVERY_VALIDATION_RESULT= receipt in actual tool stdout plus a platform output-resource URI. This intentionally rejects deployments that previously passed on assistant text alone; the format and approval/resume workflow are documented in the deployer references.
• Local tenant/workspace configuration, deployment checkpoints, response snapshots, and generated outputs are excluded from this contribution.

[Mukesh Dua (mukesh-dua)]

Fixed the test collection failure by declaring requests==2.32.5 in .github/requirements-ci.txt. The new deployer regression tests import requests; the package was present locally but absent from the clean CI environment. Verified in a new Python 3.12 virtual environment: pip install -r .github\requirements-ci.txt succeeded; python -B -m pytest .github\tests\test_agent_deployer_validation.py -v -ra passed (21 tests); python -B -m pytest .github\tests --collect-only -q succeeded (907 tests, no collection errors). Full-suite execution remains for the rerun in CI.

[Mukesh Dua (mukesh-dua)]

Updated the exact dependency-set regression guard to include requests, preserving the existing strict pinning and workflow-manifest checks. Verified the entire suite in a disposable Linux container with Python 3.12, Node.js, and all dependencies installed from .github/requirements-ci.txt: python -B -m pytest .github/tests/ -q -ra -n 4 --dist worksteal -p no:cacheprovider — 907 passed, 11 subtests passed. git diff --check also passed. This supersedes the earlier collection-only verification.

[faranm-stack]

Approve

[yousefi-msft]

Approved