From thinking-frameworks-skills
Builds a properly structured Claude Code skill (SKILL.md, resource files, evaluation rubric) from extracted components and a synthesis verdict. Use when you have extracted components and need to materialize them as a skill.
How this skill is triggered — by the user, by Claude, or both
Slash command
/thinking-frameworks-skills:skill-constructionThe summary Claude sees in its skill listing — used to decide when to auto-load this skill
Builds a Claude Code skill (the `SKILL.md` file plus any resource files and the evaluation rubric) from a structured set of extracted components and a synthesis verdict. The penultimate step in the skill-creator workflow.
Builds a Claude Code skill (the SKILL.md file plus any resource files and the evaluation rubric) from a structured set of extracted components and a synthesis verdict. The penultimate step in the skill-creator workflow.
The skill is invoked by the calling agent (typically the skill-creator agent) once synthesis-application has produced a GO or GO_WITH_GAPS verdict on the extracted components. NO_GO sends the agent back upstream.
- [ ] Step 1: Assess complexity level (1-3) from extracted content volume
- [ ] Step 2: Plan resource files (or decide SKILL.md alone is enough)
- [ ] Step 3: Draft SKILL.md frontmatter (name, description, triggers)
- [ ] Step 4: Draft SKILL.md body (workflow checklist, key patterns, guardrails)
- [ ] Step 5: Draft each resource file (focused topic, < 500 lines, WHY/WHAT structure)
- [ ] Step 6: Construct an evaluation rubric (10 criteria, 1-5 scale)
- [ ] Step 7: Output the constructed skill files
extracted_components: structured terms / propositions / arguments / solutions from component-extractionsynthesis_output: the GO / GO_WITH_GAPS verdict from synthesis-application plus its gap-fill recommendationstarget_dir: where the skill should be written (typically skills/{skill-name}/)purpose_context: e.g., skill_extraction_from_methodology (the standard case)Match structure to content volume. Don't over-engineer; don't under-engineer.
| Level | Steps in workflow | Estimated lines total | Structure |
|---|---|---|---|
| 1 | 3-5 | < 300 | SKILL.md + rubric only |
| 2 | 5-8 | 300-800 | SKILL.md + 1-3 resource files + rubric |
| 3 | 8+ | 800+ | SKILL.md + 4-8 resource files + rubric |
Decide based on:
YAML frontmatter:
name: lowercase-hyphens, ≤64 chars, no XML, no reserved words ("anthropic", "claude")description: third-person, both what the skill does AND when to use it, including trigger keywords. ≤1024 chars. This is the discovery surface — Claude uses it to decide whether to invoke the skill.Body sections (typical order):
# {Skill Name} — title + one-paragraph framing## Workflow — checklist the calling agent can copy## {Domain-specific sections} — patterns, decision logic, key concepts## Common patterns — 2-4 named patterns with usage context## Guardrails — must-nots and anti-patterns## Related — pointers to adjacent skills (one level deep only)Each resource file:
## Why X then ## What to do structure (the WHY activates context; the WHAT gives concrete instructions)Two grouping approaches:
key-concepts.md, decision-framework.md, examples.mdAlways include resources/evaluators/rubric_{skill_name}.json. 10 criteria, each scored 1-5. Common criteria:
Target average ≥ 3.5 before shipping.
## Skill Construction Output
### Complexity assessment
Level: {1 | 2 | 3}
Rationale: {1-2 sentences}
Structure: {file list}
### Files created
- {target_dir}/SKILL.md ({line count} lines)
- {target_dir}/resources/{file}.md ({line count} lines) — {topic}
- ...
- {target_dir}/resources/evaluators/rubric_{name}.json (10 criteria)
### Self-evaluation against the rubric
- Average score: {X / 5}
- Below-3 criteria: {list}
- Recommended next-pass focus: {1-2 specific improvements}
### Gaps from upstream synthesis-application
{If the synthesis verdict was GO_WITH_GAPS, restate the gap list and how the construction addressed each (filled / deferred / surfaced as a TODO in the SKILL).}
Conform to Anthropic's agent skills authoring best practices:
ServerName:tool_name.The extracted components come from a methodology guide or framework doc. Workflow steps in SKILL.md mirror the methodology's steps. Resources cover decision logic, examples, and templates.
Less common but valid. The extraction yields propositions and arguments more than steps. SKILL.md becomes a "how to apply this paper's framework to a downstream task" guide. Resources cover the paper's framework and worked examples.
The "extracted components" come from an existing skill that's grown too large or has overlap with another. The construction step writes a leaner version, deferring resources to focused files.
synthesis-application returned NO_GO, return upstream — building a skill from incomplete components is wasted work.name. "claude", "anthropic" — silently breaks discovery.component-extraction — produces the input this skill consumes.synthesis-application — runs immediately before this skill; its GO / GO_WITH_GAPS / NO_GO verdict gates this skill's invocation.evaluation-rubrics — the rubric construction in this skill's Step 6 mirrors the rubric design patterns there.skills/skill-creator/SKILL.md invokes this skill as its Step 5.npx claudepluginhub lyndonkl/claude --plugin thinking-frameworks-skillsGuides creation of effective skills that extend Claude's capabilities with specialized knowledge, workflows, and tool integrations. Includes principles for concise context and appropriate levels of freedom.
Guides creation of new Claude Code skills from scratch, covering SKILL.md anatomy, YAML frontmatter best practices, trigger phrases, and optional resources like scripts. Triggers on 'create a skill', 'new skill', 'scaffold skill'.