Audit: Modernize MCP, Copilot cloud agents, GitHub agent automation and CI controls #2455

Objective

Track and remediate the MCP, automation, CI and security findings from the October 2026 review. Findings are audit hypotheses requiring reproduction on current default branch before changes; do not assume prior reported versions remain current.

Workstreams

• Inventory live MCP servers, scopes, secrets and agent tools; separate local and cloud configs and apply least-privilege allowlists.
• Validate bespoke MCP client against official SDK and protocol conformance, including stateless transport and real tools/list and tools/call.
• Recheck gh-aw version against current release; upgrade, regenerate locks and run compile, actionlint, zizmor and poutine.
• Audit CI for continue-on-error, deterministic installs, Ruff F821, duplicate jobs and concurrency; make required checks genuinely blocking.
• Verify and enforce documented coverage threshold (90% if still policy).
• Audit npm/Python/Bandit/Trivy gates and exception baselines; block unaccepted high/critical findings.
• Review Dependabot auto-merge authority, branch rulesets, immutable action SHAs, agent permissions and execution protections.
• Review issue agent automation controls: confidence, suggestions, and approval boundaries; do not confuse issue approvals with security controls.
• Split oversized governance workflows only after measuring duplicate triggers and preserving protections.
• Attach test logs, diffs, and links to PRs for each completed workstream.

References

• https://github.blog/changelog/2026-07-23-github-mcp-server-supports-the-next-mcp-specification/
• https://github.blog/changelog/2026-07-23-agent-automation-controls-in-github-issues-in-public-preview/
• https://docs.github.com/en/copilot/concepts/agents/cloud-agent

Acceptance

All relevant changes implemented via reviewed PRs, checks passing with logs, and private settings verified by authorized repository administrators. No permission or security-policy changes without explicit review.