# Project Deep Dive

> Use when turning an internship, job, project, product, research effort, competition, community initiative, repository, document set, or case into an evidence-backed project dossier for career use.

- Skill: `zhanlincui/project-deep-dive` (Agent Skill, multi-file: 3 files)
- Install (CLI): `npx skillmds@latest add zhanlincui/project-deep-dive`
- Raw SKILL.md: https://api.skillmd.com/api/skills/zhanlincui/project-deep-dive/raw
- Safety review: pending
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: Research & Search
- Author: ZhanlinCui (https://skillmd.com/u/zhanlincui)
- Updated: 2026-09-17
- Page: https://skillmd.com/skills/zhanlincui/project-deep-dive

---


# Project Deep Dive

**REQUIRED BACKGROUND:** Use `career-state-protocol`. Every material fact, proposed addition, correction, and rejection follows its Claim rules.

## Goal

Turn raw experience into a top-tier, defensible project asset. Default to enrichment: use current industry and target context to find missing depth, then confirm what the candidate actually did. Enter source-only mode only when the user explicitly forbids additions or inference.

## Start with evidence

Classify the work as an experience hub, single project, subproject, research effort, competition, community system, or source conversion. Inspect available files, repositories, screenshots, transcripts, and linked materials before interviewing the user.

If a target JD or company is supplied, research it and build the target lenses first. Otherwise use the general AI excellence lenses from `career-state-protocol` without assuming a role.

## Deep-Dive Map gate

Before drafting the full dossier, present three to five recommended dimensions and ask the user to confirm the scope:

| Priority | Dimension | Target signal | Existing evidence | Facts to confirm | Next question |
|---|---|---|---|---|---|

Rank dimensions by target relevance, differentiation, evidence potential, and follow-up risk. Explain why each matters. Ask one question at a time after scope selection unless the user requests fast mode.

Read `references/deep-dive-map.md` for diagnostic rules.

## Deepening loop

For each selected dimension:

1. State what the current evidence supports.
2. Introduce the relevant Industry Benchmark.
3. Identify `candidate` claims and precise confirmation questions.
4. Separate personal ownership from team work.
5. Capture decisions, alternatives, mechanism, data, evaluation, result, attribution, limits, and learning where relevant.
6. Promote or reject claims through `career-state-protocol`.
7. Test two likely interviewer follow-ups before closing the dimension.

Industry knowledge can become a question, benchmark, or `Design Extension`. It cannot become claimed implementation. Keep prototype, proposal, experiment, shipped product, production system, and organization-wide adoption distinct.

## Output

Create or update `P###` using `references/project-dossier.md`. Include:

- concise spoken opening;
- problem, stakes, role, ownership, and hard decisions;
- product or system design at the evidence-supported depth;
- data, eval, reliability, safety, cost, and adoption where relevant;
- result, attribution, boundaries, failure, and learning;
- confirmed claims, remaining candidate claims, and evidence gaps;
- complete answers to likely follow-ups;
- Design Extension for credible work not actually delivered.

## Quality bar

Prefer concrete mechanisms and tradeoffs over stack lists. Translate metrics into definitions. Never turn an evaluation plan into measured impact or a team outcome into personal ownership.