28 portable Agent Skills for structured reasoning in Claude Code, GitHub Copilot, Codex, Cursor, and other compatible tools

Claude Code Thinking Skills is a public catalog of Agent Skills. Use it for decisions, diagnosis, systems, risk, strategy, and innovation. Each skill gives an agent a clear procedure. Use a skill when its mechanism fits the task.

Jump to why use this catalog, quick start, use a skill, skill catalog, evidence, repository layout, checks and contribution, questions, or license.

Use these skills when a problem needs a clear reasoning frame. The catalog covers diagnosis, decisions, creation, risk, execution, product, and strategy.

Read the catalog audit before making performance claims.

npx skills add tjboudreaux/cc-thinking-skills

Install all skills without prompts:

npx skills add tjboudreaux/cc-thinking-skills --all

/plugin marketplace add tjboudreaux/cc-thinking-skills
/plugin install thinking-skills@thinking-skills-marketplace

git clone https://github.com/tjboudreaux/cc-thinking-skills.git
mkdir -p /path/to/project/.agents/skills
cp -R cc-thinking-skills/skills/* /path/to/project/.agents/skills/

Claude Code can also load the same directories from ~/.claude/skills/ or .claude/skills/.

Start with the router when the right frame is unclear:

Use thinking-model-router to choose the right framework for this problem.

Invoke a leaf skill when the match is clear:

Use thinking-scientific-method to localize this bug.
Use thinking-reversibility to classify this architecture decision.
Use thinking-pre-mortem to stress-test this launch plan.
Use thinking-theory-of-constraints to find the binding bottleneck.

The Claude Code plugin uses thinking-skills:thinking-model-router as the exact router ID. The router can return NONE, one skill, or up to three complementary skills.

The 28 active skill IDs match the directories under skills/

• thinking-model-router; choose NONE, one frame, or a few complementary frames
• thinking-model-combination; sequence distinct models when one model leaves a blind spot

• thinking-scientific-method; rank causes and test the cheapest discriminator
• thinking-five-whys-plus; find systemic causes after localization
• thinking-kepner-tregoe; compare IS/IS-NOT defects and must/want options
• thinking-systems; map flows, feedback, delays, and high-impact points
• thinking-map-territory; resolve conflicts between docs and live behavior
• thinking-cynefin; classify the cause-and-effect domain
• thinking-socratic; expose hidden requirements with direct questions

• thinking-reversibility; separate easy-to-undo choices from costly commitments
• thinking-opportunity-cost; compare a choice with the top forgone use of resources
• thinking-probabilistic; forecast with base rates, ranges, and belief updates
• thinking-second-order; trace delayed effects, incentives, and feedback
• thinking-steel-manning; build the strongest opposing case before deciding
• thinking-bounded-rationality; set a good-enough threshold and stop the search
• thinking-circle-of-competence; check the evidence boundary before answering
• thinking-lindy-effect; weigh durability while checking for domain change

• thinking-first-principles; separate physics from convention and rebuild from basics
• thinking-triz; resolve apparently incompatible design requirements
• thinking-via-negativa; remove harmful or nonessential elements first
• thinking-thought-experiment; test controlled what-if scenarios when real tests cost too much
• thinking-jobs-to-be-done; understand the progress users hire a product to make
• thinking-effectuation; act from available means under uncertainty

• thinking-pre-mortem; turn failure paths into fixes and stop checks
• thinking-red-team; review authorized attack paths with reproducible findings
• thinking-margin-of-safety; size buffers against error and breach cost
• thinking-theory-of-constraints; find and manage the binding throughput constraint
• thinking-ooda; act while a time-pressured situation changes

The catalog audit reports:

• 28 active skills, all manual-only
• no automatic-retain verdict
• portfolio-v1 with zero model calls and an unmeasured result
• a provisional thinking-scientific-method row at +4.0 percentage points

The +4.0-point row is below the +5-point utility margin and has evidence gaps. Treat it as directional evidence, not an accuracy claim. See the decision-ready audit and analysis/evidence.json

skills/          Agent Skill source
scripts/         skill validation
evals/           structural, routing, and outcome evals
analysis/        evidence registry and audit
.claude-plugin/  Claude Code metadata

Study artifacts live under evals/studies/. External datasets follow their license limits.

Validate skill structure:

node scripts/validate-skills.js

Run the local structural gate:

EVAL_RUN=local node evals/run-structural.js

Run routing evaluation with the authenticated droid CLI:

EVAL_RUN=local node evals/run-routing.js

Read the evaluation harness guide and CONTRIBUTING.md before changing the catalog.

To add a skill, create skills/thinking-{name}/SKILL.md with Agent Skills YAML front matter, a precise trigger, a non-trigger boundary, a procedure, and checks. Run validation and the relevant evaluation before submitting.

Use thinking-model-router when the fit is unclear. Use a leaf skill when the task already names its mechanism.

Yes. Invoke any skill directly when its trigger matches.

No. The evidence supports structured procedures, not a guaranteed accuracy gain.

MIT License. See LICENSE for the full text.