AI Skill Report Card

Extracting Skills From Documents

B+79·Sep 29, 2026·Source: Web
12 / 15

Given a document (e.g., a PDF on negotiation tactics, a markdown file of book notes on decision-making), run through six steps: skim it, analyze its structure, extract its components, synthesize actionable steps, construct the skill file, then validate it against a quality rubric.

Skill Creation Workflow
- [ ] Step 1: Inspectional reading (skim, classify, assess skill-worthiness)
- [ ] Step 2: Structural analysis (unity, parts, problems solved)
- [ ] Step 3: Component extraction (terms, propositions, arguments, solutions)
- [ ] Step 4: Synthesis (completeness check, actionable transformation, triggers)
- [ ] Step 5: Skill construction (write SKILL.md + resources)
- [ ] Step 6: Validation (score against rubric, refine)
Recommendation▾
Add more concrete examples with actual before/after text snippets showing extraction in action (e.g., a real excerpt transformed into terms/propositions/solutions), not just high-level input/output summaries
14 / 15

Step 1: Inspectional Reading

Skim the whole document before going deep:

  • Read title, headers, table of contents, intro/conclusion, index if present
  • Classify document type: theoretical (explains "what is") vs. practical (explains "how to")
  • Assess skill-worthiness: does this contain a repeatable process, decision framework, or methodology — or is it just narrative/opinion with nothing actionable? If it's not skill-worthy, say so and stop.

Output: a one-paragraph summary of what the document is, its type, and whether it merits a skill.

Step 2: Structural Analysis

  • State the unity: what single problem or goal does this document address? Write it as one sentence.
  • Enumerate the parts: outline major sections and how they relate to the unity.
  • Define the problems: what specific questions does the author answer in each part?

Step 3: Component Extraction

Go section by section and extract:

  • Terms: key vocabulary/concepts the author defines specially
  • Propositions: the author's key claims/assertions
  • Arguments: how propositions are supported (evidence, reasoning chains)
  • Solutions: concrete techniques, steps, frameworks, or decision rules offered

Prefer extracting verbatim techniques over paraphrasing — precision matters more than brevity here.

Step 4: Synthesis and Application

  • Completeness check: do you have terms, propositions, arguments, AND solutions? If solutions are missing, the document may not convert to an actionable skill — flag this.
  • Identify applications: what real tasks would this methodology help with?
  • Transform to actions: rewrite extracted solutions as imperative steps ("Do X, then check Y") rather than descriptive claims ("X leads to Y").
  • Define triggers: what user phrases or situations should invoke this skill?

Step 5: Skill Construction

  • Decide complexity: single SKILL.md file, or SKILL.md + supporting resource files (templates, checklists, reference tables)?
  • Write YAML frontmatter (gerund name, third-person description with triggers)
  • Structure body: Quick Start → Workflow → Examples → Best Practices → Common Pitfalls
  • If the source has reusable templates/checklists, extract them into separate files and reference them; don't inline everything into SKILL.md if it pushes past ~500 lines.

Step 6: Validation and Refinement

Score the drafted skill against these criteria (1–5 each, threshold ≥3.5 average):

  • Actionability: can someone follow this without re-reading the source document?
  • Specificity: concrete steps/examples vs. vague abstractions?
  • Trigger clarity: will Claude correctly invoke this skill when relevant?
  • Completeness: does it cover the source's core methodology, not just fragments?
  • Concision: is it as short as possible while remaining usable?

If below threshold, identify the weakest criterion and revise that section specifically.

Recommendation▾
Example 2 is good (shows a bad/non-actionable case) but a third example showing a partial-failure case (e.g., document with propositions but no solutions) would strengthen edge-case coverage
13 / 20

Example 1: Input: A PDF titled "The Eisenhower Decision Matrix" explaining urgent/important quadrants with examples. Output: A skill named prioritizing-tasks-eisenhower-matrix with a Quick Start showing the 2x2 grid, a Workflow for classifying tasks into quadrants, and Examples of real task classifications.

Example 2: Input: A memoir chapter with anecdotes about a CEO's career, no repeatable process. Output: Assessment at Step 1 states the document is narrative, not methodological — no skill created, explain why.

Recommendation▾
The validation rubric in Step 6 is abstract—show a worked example of scoring a sample skill against the criteria to make it more actionable
  • Always do Step 1 before deciding whether to proceed — don't assume every document deserves a skill.
  • Preserve the source's specific terminology in extracted terms; renaming concepts loses fidelity.
  • When extracting solutions (Step 3), separate "always do this" (rules) from "consider this" (heuristics) — they belong in different parts of the final skill (Workflow vs. Best Practices).
  • Keep the final skill decoupled from the source document — it should work for someone who never read the original.
  • Skipping the completeness check (Step 4) and building a skill from theory alone, with no actionable steps.
  • Copying the document's structure (chapters) instead of restructuring into skill format (Quick Start/Workflow/Examples).
  • Writing descriptions in first or second person instead of third person with explicit triggers.
  • Producing an overlong skill file when a shorter one with linked resource files would serve better.
0
Grade B+AI Skill Framework
Scorecard
Criteria Breakdown
Quick Start
12/15
Workflow
14/15
Examples
13/20
Completeness
14/20
Format
14/15
Conciseness
12/15