Build Shared CI Workflow for Agent Runs

[FScholPer]

Goal

Provide reusable workflow/action(s) for agent execution and validation.

Why

Developers need one standard pipeline shape across repos with clear required checks and artifact outputs.

Tasks

• Add reusable workflow for the mandatory Lane A path.
• Integrate stages: build -> test -> policy -> artifact publish.
• Normalize failure reasons and status names across repos.
• Document adoption steps for maintainers.
• Add optional Lane B integration hooks that never affect merge eligibility.
• Add a lightweight candidate validation step before any expensive benchmark or full task-set run.

Required Pipeline Outputs

• Policy decision artifact
• Test summary artifact
• Traceability/quality metrics artifact
• Run metadata (commit, lane, tool versions)
• Validation status artifact for the cheap pre-benchmark candidate check

Docs-as-code pilot tasks

• Implement workflow job sequence using existing traceability scripts.
• Upload metrics and gate outputs as artifacts.
• Expose check names intended for branch protection.
• Add a fast harness validation step: import candidate, instantiate harness, run one tiny task spec, verify expected trace filenames.
• Reuse the pilot's small summary-first trace layout so failed tasks can be inspected without replaying a whole run.

Done When

• At least 2 pilot repos run the shared workflow end-to-end.
• Required check names are stable and documented.
• Lane B can be disabled without breaking merge-critical checks.
• Malformed candidates fail in the cheap validation stage before they consume benchmark time.
• The pilot workflow emits enough structured trace metadata for agents and maintainers to inspect failures from an index before drilling into per-task artifacts.

Parent: #2852