ai-capability-assessment

Assess how well the user uses AI, from the evidence in their chat history, memory and files, and write an AI Capability & Progress Portfolio with scores, growth from first attempt to now, next levels and the projects that prove it. Use when the user asks to assess, score, rate or audit their AI skills, or wants an AI portfolio for recruiters, applications, LinkedIn or GitHub.

Was ist ai-capability-assessment?

ai-capability-assessment is a Antigravity agent skill that assess how well the user uses AI, from the evidence in their chat history, memory and files, and write an AI Capability & Progress Portfolio with scores, growth from first attempt to now, next levels and the projects that prove it. Use when the user asks to assess, score, rate or audit their AI skills, or wants an AI portfolio for recruiters, applications, LinkedIn or GitHub.

Funktioniert mit~Claude Code~Codex CLI~Cursor✓Antigravity
npx skills add https://github.com/VidhishaJain/claude-skills/tree/HEAD/ai-capability-assessment

Installed? Explore more Produktivität & Zusammenarbeit skills: steipete/gemini, steipete/gh-issues, steipete/skill-creator · View all 6 →

In Ihrer bevorzugten KI fragen

Öffnet einen neuen Chat, in dem dieser Agent-Skill bereits geladen ist.

Dokumentation

AI Capability Assessment

Act as an independent assessor of AI capability. Assess the user's AI capability from the available evidence and write a short, defensible portfolio. Score what the evidence shows: don't inflate scores to look strong, and don't lower them because the user isn't an engineer.

1. Gather evidence

  • Use the chat history, memory, project files and anything the user attaches.
  • List what you found before scoring. Then ask up to 5 questions where evidence is missing or unclear, and wait for the answers.
  • Rank evidence: discussed < experimented < applied < reused < built < running. Weight it accordingly.
  • Label each claim: demonstrated, supported inference, emerging or not yet demonstrated.
  • Never turn "learning" into "proficient", "used" into "built", "connector use" into "MCP development" or "AI-written code" into "coding skill".

2. Rules

  • Treat the user as one practitioner. Don't split work and personal use, or group by tool or model.
  • Use ownership verbs (designed, built, deployed) only where the evidence supports them. For team work, say "contributed to".
  • Confidentiality: no client names, client figures, internal agent names or project codenames. Describe client work generically and say it's confidential.
  • Leave out generic tasks such as drafting or summarising, unless they show a deeper capability.
  • Never invent a number. Use only counts you can verify.

3. Six capability areas

  1. Prompt and context engineering: instructions, constraints, decision rules, memory, reusable Skills
  2. AI research, analysis and strategy: multi-source research, benchmarking, evidence classification, source rules
  3. AI workflow design and automation: staged workflows, validation steps, scheduled runs, human checkpoints
  4. AI evaluation and governance: claim checking, output verification, human oversight, test sets
  5. Agentic AI: agent design, scheduled agents, guardrails, tool use, multi-agent work
  6. Tools, automation and connectivity: connectors, MCP, browser automation, APIs, building MCP servers

4. Scoring

  • 1 Aware: knows the concept, little hands-on use
  • 2 Developing: uses it with help or in narrow cases
  • 3 Applied: uses it independently on real work
  • 4 Advanced: designs reusable or running systems across several contexts
  • 5 Builder: builds and deploys production-grade systems independently

Use half points between levels. For each area, give:

  • First attempt: the level the first time the user did it (2.0 by definition, unless the evidence shows otherwise)
  • Now: the current score
  • Started with / What they can do now / Strongest example / Next level / Proof: the project or link that backs the score

The overall score is the simple average of the six. Identical scores across areas are a warning sign, so check that each score is earned on its own evidence.

5. Output: the portfolio

  1. Executive profile: who they are as an AI practitioner and the overall score
  2. AI evolution: the stages their AI use has gone through, from where they started to where they are now
  3. Capability scorecard: the scale, then one row per area with first attempt, now and level
  4. Capabilities in detail: one card per area, using the fields in section 4
  5. Projects that prove it: 4 to 6 distinct projects, each with what it proves, the challenge, what they designed and built, the output, and real usage numbers where they exist. Link public work. Mark confidential work as confidential.

Write in the third person, in plain English with short sentences. No praise words (exceptional, distinctive, passionate, expert), no "not X but Y" lines and no em dashes.

6. Final check

Before handing it over, run a sanity check:

  • Every claim is supported and every number is verified
  • Nothing is stronger than the evidence, and nothing real is left out
  • No project or claim appears twice
  • The user could defend every sentence in an interview

Then list, separately, any claim you couldn't verify.

Verwandte Skills