fix(pipeline): align subagent-mode suite fallback with CLI mode

[christso]

Summary

Follow-up to #1148. Two connected changes:

1. Align subagent-mode suite fallback with CLI mode (code)

Before this PR, pipeline input and pipeline run resolved the suite (evalset) directory name from suite.metadata?.name only. If the eval.yaml had no name field, suite.metadata was undefined and the artifact layout skipped the suite segment entirely — writing to <run-dir>/<test-id>/... directly.

CLI mode (agentv eval, via artifact-writer.ts:buildArtifactSubdir consuming test.suite) already falls back through metadata.name → eval-file basename (stripping .eval.yaml) → 'eval'. This fallback is applied by the loaders (yaml-parser.ts:317-324, jsonl-parser.ts:165) and attached to every test as test.suite.

Fix: switch pipeline input:137 and pipeline run:161 to read tests[0]?.suite. Both modes now produce the same <run-dir>/<evalset>/<test-id>/ layout. pipeline bench and pipeline grade consume manifest.suite (written by input/run), so they pick up the change automatically — no consumer edits needed.

Red/green evidence (eval.yaml with no name field):

Before:

/tmp/testout/
├── manifest.json  ← no "suite" field
└── test-01/
    └── code_graders/

After:

/tmp/testout/
├── manifest.json  ← "suite": "no-name"
└── no-name/       ← derived from no-name.eval.yaml basename
    └── test-01/
        └── code_graders/

Regression test: apps/cli/test/commands/eval/pipeline/input.test.ts — "falls back to eval file basename for suite directory when name is absent".

2. Doc cleanups in agentv-bench skill

Three small clarifications that came up while reviewing #1148:

• grader.md:21 — stale bench-dir example (results/export/ → results/runs/) and clarify the evalset segment is always present (after the fix in BbEval TypeScript Migration #1).
• grader.md Field Descriptions — split into fields consumed by pipeline bench (score, assertions[]) vs kept for traceability (summary, claims, user_notes_summary, eval_feedback). Per bench.ts:86-107 only the first two are merged. Also removed orphaned execution_metrics/timing entries that fix(agentv-bench): fix grader subagent pipeline bugs #1148 dropped from the JSON example but not from the description list.
• SKILL.md Phase 1 — added a one-liner on how orchestrators detect tests that don't need Phase 2: check whether <test-id>/llm_graders/ has any .json files. pipeline input only populates it for llm-grader assertions (per input.ts:239-263).
• subagent-pipeline.md — updated artifact tree to reflect always-present evalset segment.

Test plan

• bun run build — passed (pre-push hook)
• bun run typecheck — passed (pre-push hook)
• bun run lint — passed (pre-push hook)
• bun run test — passed (pre-push hook)
• bun run validate:examples — passed (pre-push hook)
• Manual red/green UAT with no-name.eval.yaml — documented above
• Regression test added for the no-name fallback

🤖 Generated with Claude Code

[cloudflare-workers-and-pages]

Deploying agentv with    Cloudflare Pages

Latest commit:	d4c8f02
Status:	✅  Deploy successful!
Preview URL:	https://721eec9a.agentv.pages.dev
Branch Preview URL:	https://docs-grader-followup-stale-r.agentv.pages.dev

View logs