skills: agent skills home (devops-bench-review + run-parallel-evals + run-eval) by pradeepvrd · Pull Request #132 · gke-labs/devops-bench

pradeepvrd · 2026-06-25T17:52:44Z

Summary

A dedicated home for agent skills / guidelines so they can evolve independently of feature PRs.
Stacked on #131 (feat/bastion-matrix).

Three skills:

devops-bench-review (new) — review-only review of a PR or the current workspace across
four lenses: correctness; parallel-safety across the eval matrix axes (Task × Model ×
AgentConfig) — the emphasis, with a shared-state checklist and per-axis reasoning; task & stack
conventions; and docs conventions. It analyzes statically and may run unit tests / ruff
lint+format checks, but never runs benchmark evals or provisions infra.
run-parallel-evals — drives the full parallel matrix; harness-agnostic with an Antigravity
portability map and local/remote execution modes. (Relocated here so all skills sit together.)
run-eval (new) — drive a single Task × Model × AgentConfig run end to end (a 1×1×1
matrix); reuses run-parallel-evals' wrappers and recovery/reference files.

Each skill is a source dir under .agents/skills/<name>/ plus a .claude/skills/ discovery symlink
(both force-added, since .agents/.claude are in .git/info/exclude).

Stacking / dependencies

Base: feat(bastion): parallel Task×Model×AgentConfig eval matrix + Vertex auth #131 (feat/bastion-matrix) — provides docs/bastion.md and scripts/bastion/* that
run-parallel-evals / run-eval reference.
run-parallel-evals also references docs/parallel-evals.md, which lands in docs(parallel-evals): parallel evaluation runbook + known issues #126
(feat/parallel-eval-runs); the skill is fully wired once docs(parallel-evals): parallel evaluation runbook + known issues #126 is in the merge path.

The run-parallel-evals skill now lives in the standalone agent-skills PR (#132) so skills evolve independently of this feature branch. The docs it references (docs/parallel-evals.md, docs/bastion.md) and scripts/bastion/* remain here.

The run-parallel-evals skill (and its .claude/skills discovery symlink) now live in the standalone agent-skills PR (#132).

Stacked on #132 (skills/agent-skills). Each matrix combo provisions its own cluster; this makes every task collision-free under concurrent runs: - 6 manifest-gen tasks -> deployer: noop (no cluster); legacy factory honors noop - optimize-scale: new prebuilt/optimize-scale GKE stack + pre-seeded workload; matrix pins TARGET_DEPLOYMENT_NAME/NAMESPACE so both arms agree - deploy-hello-app: run-unique Artifact Registry repo name - per-run tofu stack-dir copy (both arms) removes the shared .terraform.lock race (resolves the 'Shared OpenTofu working directory' known-issue) - import + parallel-fix the merged complex/GKE tasks (#64 migration, #87 opa, #93 multi-region, #86 postgres/unhealthy/gitops, #76 debug-crashloop): per-run GitOps repo paths, dropped shared-SA container.admin (BYO creds), region-prefixed cluster names (avoid node-SA substr collision), unique task_id - cp-recovery documented as the kind-only exception (docs/bastion.md)