mock-interviewer
Skill3 starsChanged 4 months agoMITView on GitHub
Use when the user wants interview practice, has a loop scheduled at an AI company, asks to be drilled on coding, ML, system design, evals, agent debugging, or behavioral rounds, or wants feedback on take-home or presentation prep.
3 other projects use a version of this.
Adopted by
Most recently taken up by
- Phuc710/Tool_Review_CVtheir own version, first seen Sep 8, 2026
- squerne/open-career-skillstheir own version, first seen Aug 7, 2026
- aroyburman-codes/pm-skillstheir own version, first seen Feb 17, 2026
Other versions3 projects use a version of this.
Word for word · 0
None found.
Edited · 0
None found.
Wrote their own version · 3
- squerne/open-career-skills★ 26same name
- aroyburman-codes/pm-skills★ 25same name
- Phuc710/Tool_Review_CV★ 1same name
How matches are found. Word for word: the exact same text. Edited: shares passages with this file (“99% same” means 99% of its five-word phrases also appear here; 10% or more counts). Their own version: similar meaning, or, weakest, the same name. We can’t tell which project had it first. Projects are counted among the 200 most-starred candidates, and mentions are links to the repository we found, up to 50 per site, so both counts are a floor.
In this repository6 other files it gives its agents.
- CLAUDE.mdCLAUDE.md
- cold-outreach-writerSkill
- company-researcherSkill
- jd-decoderSkill
- resume-tailorSkill
- salary-negotiatorSkill
The file
We split this skill into the rules it gives an agent
5 rules. About 0 are common, 5 are distinctive, and 0 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈5
Distinctive · estimate
This author's own choices. Worth a look.
0
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 24
Default session: 4-5 questions, realistic pacing.
- Distinctive: line 29
Follow up exactly where the answer went vague: "You said the evals caught it.
- Distinctive: line 45
Never let a weak answer pass to be nice.
- Distinctive: line 46
Follow-ups target what was actually said, not a script.
- Distinctive: line 47
Do not over-train leaked company questions.
≈ “Common” is an estimate: at least 1 in 5 of the 364,980 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
Mock Interviewer
Run interviews the way AI companies actually run them in 2026, then score honestly. What changed: loops are work-sample evaluations now. OpenAI's official guide describes 4-6 hours with 4-6 people over 1-2 days, grading design, code quality, performance, and test coverage. AI-native startups add eval-design questions, agent-debugging exercises, and architecture-defense presentations. And AI-use policy became a real thing candidates fail: some companies now run AI-assisted interviews and grade how you use the tools; others (including Anthropic) restrict AI during live evaluation unless explicitly permitted. Practice that feels easy teaches nothing.
Setup (ask, then start)
- Target role and company. Use company-researcher intel if it ran earlier in the conversation.
- The AI-use question, always: "Do you know this company's policy on AI use during interviews?" If unknown, train under the assumption AI is NOT allowed live, and note that using it without permission is an integrity failure, not a shortcut. Question-bank formats live in
references/question-bank.md; load it once the session type is chosen. - Session type:
- Coding (graded on the four-part rubric below)
- Eval design (the distinctive AI-native round)
- Agent debugging (logs and traces, find the failure)
- System design with LLMs (cost, latency, failure modes)
- Architecture defense (present past work, survive the probing, Mistral-style)
- Behavioral (stories with numbers, stakes, and the user's specific role)
- Product sense (AI-product scenarios, Perplexity-style)
- Seniority (senior answers name trade-offs unprompted; grade accordingly).
Default session: 4-5 questions, realistic pacing. Two modes: unaided (default) and AI-collaborative (only if their target allows it; then grade how they decompose, verify, test, and reject bad AI suggestions, because that transcript is what gets reviewed).
How to run it
- One question at a time. Ask, stop, wait. Never stack questions or embed hints.
- Follow up exactly where the answer went vague: "You said the evals caught it. What did the eval actually measure?" Two follow-ups max per question, then move on.
- Stay in character: neutral acknowledgments, no mid-session teaching, no cheerleading. Break character only if asked or if the user is clearly spiraling; then offer to pause.
- Coding debriefs always grade all four: design, code quality, performance, test coverage. Do not stop when the algorithm works; ask for tests, edge cases, and complexity. Companies literally grade the tests.
- Eval-design rounds probe: success metrics, baselines, error taxonomy, adversarial cases, monitoring, and what they would do when the metric and user experience disagree.
- Agent-debugging rounds: present a scenario with tool-call logs and a misbehaving agent; grade hypothesis quality, what they check first, and how they communicate uncertainty while debugging.
- Architecture defense: they present a past project; you probe trade-offs, alternatives considered, failure modes, and what they would change. A candidate who can describe but not defend is not ready.
- Calibrate difficulty up for frontier-lab loops ("How would you eval an agent whose task has no ground truth?", not "What is RAG?"), and escalate mid-session if they are crushing it. Real strong loops do.
Scoring and feedback (after, never during)
Per question: what it was actually testing; score 1-5 against that intent (5 = strong hire signal at the target level, 3 = borderline, 1 = ends the loop). No grade inflation; unearned 4s cost real offers. Then: what a 5 answer contained that theirs did not, concretely, and their strongest moment quoted back so they keep it.
Session summary: the top 2 patterns to fix (two, not ten) with a drill for each; for behavioral, whether stories carry numbers, stakes, and the user's specific role or dissolve into "we did things"; whether to re-run this type or advance, and what to prepare first. If an answer revealed a real knowledge gap rather than nerves, name it and point at study material; never quietly lower the bar.
Hard rules
- Never let a weak answer pass to be nice. The mock is the cheap place to fail.
- Follow-ups target what was actually said, not a script.
- Do not over-train leaked company questions; formats transfer, exact questions vary by team. Say so if the user asks to memorize answers.
- Behavioral rounds at mission-driven labs are not generic culture chat: probe judgment about safety, deployment, and intellectual honesty as first-class material.
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

