Turn Interview ASR into Dual-Perspective Reviews with talk-review

Where this fits

You just finished an AI product interview and dropped the raw ASR transcript into your agent. You want an honest post-mortem, not a generic "you did great" summary. talk-review reads the transcript, asks for the linked resume and experience materials, and produces two grounded reviews from the same evidence: one for your improvement, one simulating how a recruiter evaluates you.

Why agents benefit

• Faithful evidence separation: splits interviewer and candidate utterances without guessing, and flags uncertain ASR segments instead of silently "fixing" them into facts.
• Dual-perspective output: generates a candidate-side improvement plan and a recruiter-side evaluation from the same evidence base, so both views stay consistent.
• Role-specific lens: for AI product, AIGC, LLM, Agent, model/data platform, and AI+industry roles, it loads a dedicated AI PM review framework that diagnoses business-product-tech chain, metric definitions, project maturity, personal ownership, and terminology-substituted-for-facts.
• Precise gap routing: identified gaps are routed back to the relevant skill (resume, experience-deepthink, etc.) for targeted follow-up, not generic advice.
• Material routing discipline: auto-reads uniquely matched private materials, asks only when ambiguous, and refuses formal evaluation when required materials are missing.

Practical scenarios

Post-interview debrief after an AI product round

You just finished a first-round interview for an Agent product role. You upload the ASR, the skill asks for your resume and experience materials, then returns a candidate-side review that points out you used "multi-agent orchestration" without explaining your actual ownership — plus a recruiter-side note showing how that reads as terminology replacing facts.

Reviewing a stored transcript from a model platform role

You have a transcript from a data platform interview saved weeks ago. The skill auto-associates the matching resume and experience materials from private space, and produces a retroactive evaluation that helps you decide whether to reapply or move on.

Preparing for a second round

Before round two, you run the skill on your first-round transcript. The recruiter-side evaluation reveals the interviewer probed metric definitions you glossed over. You now know exactly which numbers to prepare.

Add it to your agent workflow

Install with:

npx skills add riwonswain-ovo/OfferLoop --skill talk-review

Then trigger it naturally: "复盘刚才的面试" or "分析这份面试转写". The skill reads references/review-rubric.md, recruiter-analysis.md, role-evidence-review.md, and the voice contract before producing output — so it runs with a consistent rubric every time.

Compare before adopting

Before adopting, compare against generic interview-feedback skills. Check three things: (1) whether the skill preserves ASR uncertainty instead of fabricating clean text; (2) whether the dual-perspective output is actually grounded in the same evidence; (3) whether the AI PM lens is specific enough for your role type. Also check maintenance freshness — this repo is actively updated, and the skill's material-routing discipline is a real differentiator.

Why it is worth tracking

This skill is worth tracking because it solves a real pain point — turning noisy ASR transcripts into structured, dual-perspective feedback — with a disciplined evidence model. With 13 stars and an active repo, it's early but focused. Evaluate it when you do regular interview prep or want to add structured interview analysis to your agent workflow. You can follow its progress on OpenAgentSkill.