Summary

Closes #189

Adds support for follow_up_prompts in eval YAML task definitions, enabling multi-turn evaluation scenarios where each follow-up reuses the same session and workspace.

Working as Linus (Backend Developer)

Changes

Core

• internal/models/testcase.go: Add FollowUps []string to TestStimulus (yaml: follow_up_prompts)
• internal/execution/engine.go: Add WorkspaceDir to ExecutionRequest for workspace reuse
• internal/execution/copilot.go: Skip setupWorkspace when WorkspaceDir is set
• internal/execution/mock.go: Support workspace reuse and SessionID passthrough
• internal/orchestration/runner.go: Add executeFollowUps() with result aggregation (events, tool calls, usage, duration)

Tests

• 3 YAML parsing subtests (present, omitted, single)
• 5 orchestration tests (no follow-ups, single, multiple, error mid-sequence, full benchmark)

Documentation

• schemas/task.schema.json: Add follow_up_prompts property
• site/src/content/docs/guides/eval-yaml.mdx: Add follow-up prompts guide section
• site/src/content/docs/reference/schema.mdx: Add schema reference entry

Example

inputs:
  prompt: "Create a Python function that reads a CSV file"
  follow_up_prompts:
    - "Add error handling for missing files"
    - "Write unit tests for the function"

How it works

1. Initial prompt executes normally, creating a workspace and session
2. Each follow-up reuses the same WorkspaceDir and SessionID
3. Results are aggregated: events/tool calls appended, duration summed, last output wins
4. If any follow-up fails, remaining are skipped and the run is marked as error
5. Graders evaluate only the final state after all prompts complete