end-to-end-testing

name	end-to-end-testing
description	Use when running the Rundown end-to-end test runbook and reporting structured feedback on the workflow.

End-to-End Testing

Run the end-to-end test through Rundown state. Do not improvise. Start:

rundown run rundown:end-to-end-test

Flow

• Parent, final feedback, end-to-end-test/write-file.runbook.md, and end-to-end-test/review-and-collate.runbook.md run locally.
• end-to-end-test/review-file.runbook.md and end-to-end-test/collate-files.runbook.md are delegated by the review wrapper.

Only nested review and collation runbooks are delegated. If the wrapper completes before collate, report a defect; never delegate collation from the parent.

Operating Rules

• Follow the active prompt. Use default JSON output; never add --text (it is human-only output, not part of the agent protocol).
• Treat rundown run, rundown pass, rundown fail, rundown claim, and rundown collect output as the next context. Parent-side pass/fail/collect carry --run <rd_…> (capture the run id from rundown run output, echoed as runbookId on every event); child-side commands carry --claim-id <claim_id>.
• Use rundown status only to recover orientation after an error or interruption.
• Pass only after the current step is complete; fail when it cannot be completed as written.
• Preserve state on errors. Do not prune, clear, or delete .rundown.

The DELEGATE step auto-issues a claim token (delegations in rundown status). Claim it:

rundown claim <token>                  # returns claim_id; dispatch the child to a subagent

Drive the claimed child like any runbook: advance each step with rundown pass --claim-id <claim_id> / rundown fail --claim-id <claim_id>. Prompted child steps need this claim-id transition to advance. On a delegation-exposed run, a bare rundown pass/rundown fail is refused with ACTOR_CONTEXT_REQUIRED regardless of claim state (exposure is sticky). A --run-targeted parent advance is additionally refused with OPEN_DELEGATED_CHILDREN while a claimed child is open.

When the child's final step completes it auto-resolves its parent substep and the parent auto-aggregates and advances. rundown pass/fail --claim-id is idempotent on a resolved child, so it also confirms or overrides a child you stop early.

Artifacts

Use rendered {{ path Alias }} paths.

Expected artifacts:

• PlanSchemaPath
• PlanPath
• ReviewSchemaPath
• ReviewPath
• ReviewPaths
• CollatedReviewPath
• FeedbackPath

Do not rediscover paths or add artifact-only steps. Missing/wrong variables and write-one/check-another mismatches are defects.

Feedback

Feedback must be JSON matching review.schema.json. Use empty items when there are no findings.

Capture only concrete issues: unclear prompts, wrong delegation boundary, missing/wrong artifact variables, schema mismatches, required manual inference, failed commands with observed output.