docs: auto-update for merged PRs (2026-09-02)

[aheritier]

Documentation changes

Brings docs/features/evaluation/index.md up to date with the evaluation
features merged into main over the last 36 hours. The assertions
(code-based grading) and --repeat pass@k/pass^k features landed across
several PRs but were never documented; the skills-related PRs and the
StartableToolSet backoff PR already carried their own doc updates and needed
no further changes.

• Eval Criteria: document the new assertions field and the supported
  assertion types (contains, not_contains, equals, starts_with,
  ends_with, regex, cost_threshold, tool_called).
• Scoring Metrics: add the Assertions row now that assertions are
  actually run and scored (not just accepted by the schema).
• New "Repeat Metrics (pass@k / pass^k)" section: explain what --repeat
  now prints and what the two metrics measure.
• Regression gate: note that assertion rate is now one of the aggregate
  metrics compared against --baseline, and that pass@k/pass^k are reported
  as informational (non-gating) deltas.

One commit per source PR:

• feat: add assertions schema to EvalCriteria for code-based grading #4088 — assertions schema on EvalCriteria
• feat: wire assertion runner into runSingleEval execution pipeline #4100 — assertion runner wired into the eval pipeline
• feat: expose pass@k and pass^k metrics for --repeat runs #4092 — pass@k / pass^k metrics for --repeat
• feat: include assertions and pass@k in baseline regression comparison #4103 — assertion rate & pass@k in baseline regression comparison

PRs reviewed with no doc changes needed

• feat: add verify script runner for post-agent outcome verification #4089 (verify script runner) — fully removed again by feat: include assertions and pass@k in baseline regression comparison #4103, never documented, no net doc change.
• feat: upgrade relevance judge prompt with rubric, CoT, and anti-bias rules #4091 (judge prompt rewrite) — internal prompt change, not user-facing/documented behavior.
• fix(tui): stabilize slash command description alignment #4090, ci: run ci and codeql checks on pull requests to any base branch #4075 — no doc-relevant changes.
• feat: safely gate embedded skill commands #4093, fix(skills): let inline skills override discovered ones #4094, fix(skills): reject whitespace in inline skill names #4095, feat(skills): expand embedded commands in inline skills #4096 (skills) — already updated docs/features/skills/index.md in the same PR; verified accurate against final code.
• fix(#4060): add bounded jittered backoff to StartableToolSet retry path #4062 (RAG/toolset backoff) — already updated docs/tools/rag, docs/tools/lsp, docs/tools/mcp, docs/community/troubleshooting in the same PR; verified accurate against final code.
• docs: update CHANGELOG.md for v1.130.0 #4101 — CHANGELOG-only PR, already docs.