docs(eval): clarify eval authoring contracts

[christso]

Summary

Eval authors now get the corrected contract shape before they draft repo-state self-evals: assertion strings can carry the grading contract without a duplicate criteria field, expected_output is framed as a golden/reference answer, and historical repo-state tests are shown with a real workspace.repos[] commit pin.

The agent-facing skill data now teaches and reviews the same conventions, including the issue #18 pattern of using shorthand assertions instead of redundant named llm-grader blocks.

Related handoff: https://github.com/EntityProcess/agentv-beads/issues/18

Verification

• git diff --check
• bunx biome check apps/web/src/content/docs/docs/evaluation/eval-cases.mdx apps/web/src/content/docs/docs/evaluation/eval-files.mdx apps/web/src/content/docs/docs/guides/eval-authoring.mdx skills-data/agentv-bench/references/eval-yaml-spec.md skills-data/agentv-eval-review/SKILL.md skills-data/agentv-eval-writer/SKILL.md
• No eval YAML files were changed, so eval schema validation was not applicable.

Additional repo-local CLI readback was attempted but blocked by local build setup: bun apps/cli/src/cli.ts skills get agentv-eval-writer hit a stale @agentv/core export, and bun run build could not proceed because tsup is not installed in this worktree.

---

[cloudflare-workers-and-pages]

Deploying agentv with    Cloudflare Pages

Latest commit:	a0c90cc
Status:	✅  Deploy successful!
Preview URL:	https://36296e5d.agentv.pages.dev
Branch Preview URL:	https://docs-eval-authoring-skill-da.agentv.pages.dev

View logs

[christso]

Coordinator review follow-up:

I found one docs consistency issue before merge: apps/web/src/content/docs/docs/tools/validate.mdx still said validation required criteria, while this PR correctly updates the eval/test docs to say criteria is conditional when expected_output, assertions, or turns are present.

Fixed on the PR branch in a0c90cc5 (docs(eval): align validate field requirements). Local verification after the fix:

• git diff --check
• bunx biome check on the touched docs and skill-data files

Waiting for refreshed GitHub Actions before merge.