Repository navigation
feat: core-concept /hey csinterview and /hey morecs follow-up interviews - #23
Merged
Merged
Conversation
The additional-question batch is now requested with /hey whatelse; the old name is no longer recognized. Help text, hints, error messages, READMEs, plan and examples are updated, and the parser test asserts the old name is rejected. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Paired controller/runner release 25312d1 with protected backups, an isolated rehearsal on a restored production copy and a compatible rollback controller. Records verification evidence and the rollback command. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Rewrite the interview prompt around a senior-interviewer persona and a ranked core-concept bar (fundamentality > relevance > cost of not knowing > transferability). Questions must defeat both a candidate who lacks the concept and one who can only recite its definition, ask one layer below the definition, and avoid look-up details and name-picking. The prompt adds a candidate -> select -> self-review procedure, repo-agnostic good/bad examples and a teaching-key contract. - csinterview returns two questions; fewer require a shortfall reason - interview keys may explain in up to five sentences (2000 chars) - wrong-option notes may bold up to two phrases with **...**; segments are escaped first and rendered as <b>, dropping unmatched or clipped markers - a wrong bare letter leads with the stored note for that option; reasoned grading must name the missed concept - interview prompts receive labelled priorQuestions - demo fixture replaces the name-picking anti-pattern Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
/hey morecs anchors on the latest csinterview or morecs batch of the revision and asks up to four questions one layer below it: three follow-ups plus one new interview question when an uncovered core concept qualifies, otherwise four follow-ups. Every 면접 question of the anchor batch needs a follow-up unless the model explains why; fewer than four require a shortfall reason, and nothing is padded. - output adds kind (followup|interview) and anchor (label or null); code checks anchors against the anchor batch, coverage, at most one new interview question and duplicates across the revision history - follow-ups are numbered 꼬리N, new interview questions continue 면접N; morecs batches store their labels and show "↳ 면접2에서 이어짐" in question, grade and answer-sheet comments - answer accepts 꼬리N; showanswer recognizes 꼬리-only threads - separate five-run limit; no interview batch points to csinterview - GMS, Codex and Claude register the morecs schema; the runner admits the phase; the merge gate is unchanged (it reads only Q questions) - legacy kindless batches and five-question interview batches keep their numbers and can anchor morecs Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Update both READMEs, the implementation plan and command help for the two-question csinterview bar and the morecs anchor, numbering, limit and grading rules. The demo adds a morecs batch anchored on the demo interview pair with a mixed 꼬리/면접 grade and thread reveal; the public examples are regenerated from it. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Real GMS runs needed 109 s for the first four-question batch and timed out three times at 120 s once the anchor batch held four questions. Leases are heartbeated during command steps and n8n polls without a total cap, so only the call budget limited it. csinterview keeps 120 s. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…questions Round-2 real GMS outputs showed follow-ups naming an anchor whose concept they did not deepen, new interview questions solvable by one complexity formula, and an answer resting on an assumed OS facility. The shared exclusions now name the last two, and morecs requires the anchor to be the question the follow-up deepens and a new interview question to beat the remaining follow-ups. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Independent scoring of round-3 real outputs found coreDepth 3.73 and falling grounding (3.62 -> 2.56 -> 2.31) across morecs rounds: later follow-ups added subsystems the change never has (replication, partitions, MAC encoding, version stores), repeated judgments under other anchors and filled distractors with formal variants. Follow-up assumptions may now change only one condition of a mechanism the change uses; otherwise a new core interview question is preferred. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The prompt said a location only had to be a readable line, but the validator accepts only lines inside the diff hunks. Round-5 real runs cited workflow.ts:90/94 and controller.ts:113 (outside their hunks), failing a morecs job on all three attempts twice. The prompt now states the rule and points to the nearest hunk line that uses the code. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Fixed rubric, independent per-question scores, raw outputs for every round, run statistics and the iteration log. Final after+morecs scores: coreDepth 4.11, nonNiche 4.88, learningValue 4.98, 0 ambiguous and one bare-definition question; the new csinterview alone meets every target. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The final independent scoring flagged s4 꼬리1 as bare-definition: its
goal sentence ('do not disclose beyond the target') was the correct
option. Both prompts now give this example, require the goal to state
only the outcome, and require every wrong option to claim it meets
that goal too; self-review checks both.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The prompt fixes and verification records were first committed on feat(be)/138-GMS_호출_횟수_절감 after another session switched the checkout. Their cherry-picks here have new SHAs; the docs now cite them and map the tested commits recorded in runs.json to these. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Round 7 (prompt 16cabe3) raised knowledge+application, distractors and grounding and kept coreDepth/nonNiche/learningValue above 4 with 0 ambiguous, but two questions still restate the answer in their stated goal, so the bare-definition target of 0 remains unmet. Adds round-7 outputs, scores and run records, and the branch reconciliation note. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Prompt wording alone left stems whose stated goal restates the answer (2 bare-definition flags in round 7). A string-overlap check was calibrated on 188 scored real questions and rejected: the 5 flagged questions were indistinguishable from good ones lexically. Each csinterview/morecs question now carries a required check: the option a definition-only candidate would pick, why, the stem sentence nearest the answer quoted verbatim, and a giveaway flag. The controller drops a question when the pick equals the key, the quote is not in the stem, or it declares a giveaway; survivors are kept with an explained shortfall, nothing is padded, and the check is not stored. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The enforced giveaway check (6e34bfa) produced 0 bare-definition and 0 ambiguous questions across 54 real outputs, with coreDepth 4.20, nonNiche 4.78 and learningValue 5.00. Adds round-8 outputs, scores, run records and the remaining weaknesses. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Every sample has a real before run (rounds 2 and 4) and a final after run with three chained morecs (round 8); the table shows each sample's averages so the comparison does not rely on pooled numbers alone. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
# Conflicts: # docs/operations.md # docs/progress.md
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
/hey csinterviewrewrite: 2 questions per run, on the most core CS concepts the change relies on. The prompt adds a senior-interviewer bar, ranks candidate concepts (fundamentality > relevance > cost of not knowing > transferability), and requires that both a candidate who lacks the concept and one who only knows its definition get the question wrong./hey morecs: up to 4 questions anchored on the latest csinterview/morecs batch of the revision.꼬리N) plus 1 new interview question (면접N), or 4 follow-ups when no new core concept qualifies.↳ 면접2에서 이어짐. Questions are deduped against the revision history, with 5 runs per revision.<b>, and unmatched or clipped markers are dropped.check(definition-only pick, verbatim stem quote, giveaway flag). The controller drops failing questions instead of failing the batch.morecs, and the runner admits the phase.answeraccepts꼬리N, andshowanswerrecognizes 꼬리-only threads.main: brings in the GMS usage logging ([S15P31E102-138] GMS 요청 사용량 로깅 #22).mainalso gains the n8n orchestration and earlier csinterview work that production already runs.Verification
npm test192/192 andnode scripts/integration.mjs71/71 on the merge commit, and examples match the demo.꼬리2 B 면접3 C, 꼬리-only reveal threads, bold rendering at every clip position, the giveaway check, and legacy batches. Isolated mutations of the new rules were all caught.gpt-6.1-solon 4 commits of this repo, with "before" (old prompt) and "after" plus 3 chained morecs per sample, scored by an independent reviewer against a rubric fixed before generation.docs/verification/csinterview-morecs-2026-10-26/README.md.Not verified
Deploy note
The runner must accept the
morecsphase before the controller sends it, so replace both together.🤖 Generated with Claude Code