AI Skill Report Card
Generating AI Skills
Quick Start15 / 15
Input: "I'm a senior tech recruiter, I source candidates for engineering roles at startups, screen for culture fit and technical depth, always personalize outreach."
Output:
Identity: You are an expert technical recruiter specializing in engineering roles at high-growth startups.
Process:
1. Source candidates using the MATCH framework (Motivation, Aptitude, Technical depth, Culture fit, Hunger)
2. Screen resumes and portfolios for relevant technical depth
3. Conduct outreach referencing candidate's actual work/projects
4. Run structured interviews assessing culture fit and technical skill
5. Manage full-cycle hiring from source to offer
Quality Standards:
- Every candidate interaction is personalized. No generic templates.
- Always reference their actual work.
Recommendation▾
Add an example with a very sparse/minimal input to demonstrate handling ambiguous or thin descriptions, per the 'don't ask clarifying questions' pitfall.
Workflow14 / 15
Progress:
- Extract the core role/identity from the raw description
- Identify or invent a memorable framework/acronym for the process (if one exists implicitly, name it)
- Break the process into 3-7 concrete, ordered steps
- Pull out explicit quality standards/rules the person mentioned or implied
- Draft the four sections: Identity, Process, Quality Standards, (optional) Knowledge Areas
- Tighten language — remove filler, keep it directive and specific
- Self-grade the output (see Grading below) and revise weak sections
1. Extract Identity
One sentence, third person, starts with "You are an expert/senior [role] who/specializing in [domain]." Avoid vague titles — be as specific as the input allows (e.g., "engineering roles at high-growth startups" not just "recruiting").
2. Build the Process
- Look for any named methodology in the input; if none exists, synthesize a short acronym-based framework from the key steps mentioned (this makes the skill memorable and structured).
- List steps in the actual order they'd be executed, numbered.
- Each step should be an action verb + specific detail, not generic ("Source candidates using the MATCH framework" not "Find good candidates").
3. Define Quality Standards
- Convert any "always/never/no generic X" statements from the input directly into rules.
- Keep to 2-4 crisp bullet points. These are the guardrails that prevent generic AI output.
4. Optional sections
Add "Knowledge Areas" or "Tools" sections only if the input mentions specific domain knowledge, tools, or terminology worth listing.
5. Grade the skill (internal check)
Score 0-100 based on:
- Clarity of identity (specific role, not vague) — 25 pts
- Actionability of process (concrete numbered steps, not platitudes) — 40 pts
- Enforceability of quality standards (specific, checkable rules) — 25 pts
- Conciseness (no filler) — 10 pts
If below ~85, tighten vague language and add specificity before finalizing.
Recommendation▾
Show a 'bad output' example (e.g., vague identity, generic process) alongside a corrected version to make quality contrast concrete rather than described.
Examples18 / 20
Example 1: Input: "I've done UX research for 8 years — interviews, synthesizing data, turning insights into design decisions." Output:
Identity: Expert UX researcher who conducts user interviews, synthesizes qualitative data, and turns insights into design decisions.
Process:
1. Write discussion guides from research questions
2. Run moderated interviews
3. Affinity map findings into themes
4. Translate themes into actionable design recommendations
5. Present insights with supporting evidence to stakeholders
Quality Standards:
- Every insight must trace back to specific participant quotes/behavior
- No recommendations without supporting data
Example 2: Input: "I review code and I don't take shortcuts, I look at architecture not just style." Output:
Identity: Senior code reviewer who evaluates architecture and design, not just surface-level style.
Process:
1. Read the full diff/PR context before commenting on any single line
2. Identify architectural issues (coupling, single responsibility violations, scalability risks) first
3. Flag style/nit issues separately and mark them as optional
4. Check for test coverage on new logic paths
5. Summarize with a clear approve/request-changes verdict and reasoning
Quality Standards:
- Never approve based on style alone; architecture issues take priority
- Every "request changes" must include a concrete suggested fix
Recommendation▾
Clarify how the internal self-grading step is applied in practice—since it's self-referential to this skill's own grading logic, an explicit before/after revision example would strengthen the workflow section.
Best Practices
- Default to 4-6 process steps — enough to be concrete, not so many it becomes a manual.
- Invent a named framework (acronym or short phrase) when the input implies a repeatable method but doesn't name one — this dramatically increases perceived structure/quality.
- Mirror the person's actual language/terminology from their input; don't genericize their vocabulary.
- Keep total output short enough to paste directly into a system prompt (under ~200 words for Identity+Process+Standards combined, unless the input demands more).
Common Pitfalls
- Don't write vague identities like "You are a helpful assistant who knows about X." Be role-specific and senior-level.
- Don't turn quality standards into more process steps — standards are constraints/rules, not sequential actions.
- Don't pad with generic advice ("communicate clearly", "be professional") unless the input specifically implies it.
- Don't ask the user clarifying questions — always produce a complete skill from whatever input is given, even if sparse.