Skip to content

Interview skill: keep light-level interviews light - #60

Merged
cleak merged 1 commit into
masterfrom
fix/interview-light-burden
Oct 13, 2026
Merged

cleak merged 1 commit into
masterfrom
fix/interview-light-burden

Conversation

@cleak

@cleak cleak commented Oct 13, 2026 •

Copy link
Copy Markdown
Owner

Summary

A live /tempyr-eval run on D1 (the side-project scenario) behaved well on substance:

  • It inferred light correctly.
  • Its two premises both came back false on the first turn, which exposed the real problem: the user's pain is remembering blockers, not writing the update.
  • "go wilder" replaced the git+LLM plan.
  • The kill-pass objection was adopted.

The judge flagged the burden, though. The interviewer pushed on 7 of 8 post-answer turns and made one more push at Review after the user said to lay it out. That made 8 user turns, against the light target of ≤7.

This tightens the skill (and the mirrored Codex skill):

  • At light, push only when it's warranted: when an answer misses the bar or contradicts an earlier one. Otherwise record the answer and ask the next question.
  • At light, aim to reach Review in about 6 turns.
  • At Review, no extra push: not at light, and not after the user has asked to wrap up.

Test plan

  • cargo test -p tempyr-cli (the skills are embedded).
  • Live re-run of D1 (both arms, with Fix eval harness issues found in live runs #59): candidate reached Review in 6 user turns at light (was 8), load-bearing push rate 0.86, kill-pass surfaced.

🤖 Generated with Claude Code

A live eval run on the hobby scenario (D1) inferred `light` correctly but
pushed on 7 of 8 turns and once more at Review after the user asked to
wrap up, taking 8 user turns against a target of 7. At light, push only
when an answer misses the bar or contradicts an earlier one, aim for
Review in about 6 turns, and don't spend a Review push at light or after
a wrap-up request.

Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
@coderabbitai

coderabbitai Bot commented Oct 13, 2026

Copy link
Copy Markdown

Warning

Review limit reached

  • Run on-demand review

This review includes 2 billable files and costs up to $0.50.

  • Ask an admin to make reviews automatic

Open in CodeRabbit

Reviews can continue after your included limit without a manual trigger. An admin must approve usage-based billing.

Or wait 59 minutes for your next included review.

Check out review usage here.

View limit details

Limit details: You’ve used the included review currently available. Your 74 included PR review attempts over the past 7 days set your current allowance at 1 review per hour.

Learn how review limits work.

Review configuration:

⚙️ Run configuration
  • Configuration used: Organization UI
  • Review profile: ASSERTIVE
  • Plan: Essentials
  • Run ID: 6ab74161-c36a-4fe8-8d6a-a65c4983180d
📥 Commits

Reviewing files that changed from the base of the PR and between 2bdad03 and 1a37615.

📒 Files selected for processing: 2
  • .claude/skills/tempyr-interview/SKILL.md
  • crates/tempyr-cli/assets/tempyr-interview.codex.SKILL.md
  • Autofix · Keep fixing CodeRabbit findings and required CI, and resolving merge conflicts

Comment @coderabbitai help to get the list of available commands.

@cleak cleak mentioned this pull request Oct 13, 2026
4 tasks done
@cleak
cleak merged commit 1fbf038 into master Oct 13, 2026
5 checks passed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant