Simplify v1 CLI plumbing

[xeophon]

Overview

This makes the v1 command-line tools smaller and more consistent. It removes code that was doing the same work twice, uses information that is already saved with each run, and fixes debug runs being recorded as failures when their command succeeded.

What changed

Evaluation and resume

• Resume now uses the task hash already stored in each saved episode instead of recalculating it every time.
• Runs created before task hashes were stored still resume correctly by calculating the hash from their saved task data.
• Missing hashes use that fallback, while malformed list or object hashes are treated as invalid rows and rerun safely.
• Saved and current configurations are compared directly instead of hashing both configurations first.
• Saved JSON configurations are read through Pydantic directly, and the already calculated run path is reused throughout command startup.
• Removed an unused evaluation environment-id shortcut and a one-use resume-message helper.

Debug and validation

• A successful debug command now produces a successful trace instead of being saved as failed.
• Debug runs save their resolved configuration as configs/debug.json, matching the other commands.
• Gold-answer and setup-only validation now share one execution path instead of maintaining two nearly identical implementations.
• Resume loaders use one JSON parser and safely skip malformed result rows.

Environment scaffolding and command configuration

• The environment name is now required by the configuration schema instead of being checked manually later.
• Removed unreachable per-file skip logic, ignored return values, and temporary aliases from environment scaffolding. The existing top-level overwrite protection and --force behavior remain unchanged.
• Removed unused eager command-configuration exports and an unused default argument from command id resolution.

Dashboards and saved output

• The evaluation dashboard builds its static header and runtime description once instead of rebuilding them four times per second.
• Progress display work is reused within each refresh, and terminal arrow-key handling is disabled when the dashboard is showing logs because paging is inactive there.
• Dashboard sizing now uses the same Rich console as the live display.
• Replay dashboard state is typed directly as replay progress, removing a fallback lookup that could never be needed.
• Saved-output descriptions now consistently refer to configs/<command>.json, traces.jsonl, and run.dir rather than the old TOML paths or display name.

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 9b15b4ff21

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[macroscopeapp]

Approvability

Verdict: Not approved

Macroscope's review found this PR not approvable — This PR combines a mostly mechanical CLI refactor with changes to persisted debug status, resume rollout selection, and the exported configuration API. Those effects cross durable-run and compatibility boundaries, so the changes merit human review.

^{You can add or adjust custom eligibility rules. Learn more.}