28 manually invoked frameworks, applied to one concrete unknown
CC Thinking Skills
Choose a Method, Check the Evidence
Turn an unclear problem into an analysis with inputs, a procedure, stopping conditions and checks. Route to a method or directly investigate a fault, assess a commitment, anticipate failure, find a bottleneck or understand a user’s job.
Hosts named by the author
- Claude Code
- GitHub Copilot
- Codex / Cursor
- GitHub stars
- 1,679
- Version
- 1.0.0
- Checked
Use one method and complete one check
Start with a small real problem. Invoke a known match directly and use the router when the fit is unclear. Judge the result by the assumption exposed, observation performed or decision changed, rather than the number of frameworks named.
Pick a task and get started
Install in the project or material workspace, confirm loading, then choose a task below. Filenames and thinking-results/ output paths are editorial conventions to replace with real locations; the examples contain no completed checks.
When the problem is still unclear, decide whether a framework helps and which mechanism addresses the unknown.
Install the catalog and select skills
Run the author’s installation command in a terminal, selecting supported hosts and the skills you need. Return to the assistant and explicitly invoke a thinking- entry; start with thinking-model-router when the fit is unclear.
Read the author’s setup guidenpx skills add tjboudreaux/cc-thinking-skillsUse the correct name for your installation
After Skills CLI or directory installation, ask the assistant to confirm it can read thinking-model-router/SKILL.md. If you installed only that entry, its selected skills must also be available. For the Claude Code plugin, run /plugin marketplace add tjboudreaux/cc-thinking-skills, then /plugin install thinking-skills@thinking-skills-marketplace. Adding the marketplace alone does not enable the plugin. Its exact router ID is thinking-skills:thinking-model-router; leaf skills use the same thinking-skills: prefix. The author also documents project .agents/skills/ and Claude .claude/skills/ locations; use the directory your host supports.
Read author installation and invocation instructionsGive your AI this first task
These editorial prompts follow the pinned source. They were not run through a model. Replace the inputs and paths; the results explain expected deliverables and inspection.
Invoke thinking-model-router. Read decision-brief.md, which should contain the decision, real options, known facts and sources, time and resource limits, and what happens if the choice must be reversed. Confirm the material and its gaps; a brief is not automatically verified evidence. Choose NONE or one primary skill by mechanism. Consider up to three complementary skills only when the primary leaves distinct material questions open. Return outcome, route IDs, each unique question, domain, problem type, constraints, rationale and exit conditions. If no framework helps, state the next concrete task. For this round, save the routing note to thinking-results/route.md without committing to a solution. If you return a bare slug, also give its exact callable name for this host; the Claude Code plugin uses thinking-skills:thinking-<slug>. List the inputs the selected skill needs next.
Check what you receive
- A NONE, single-skill or explicitly divided multi-skill route, with a reason and a condition for abandoning it. NONE is a valid answer.
- Check that every name belongs to the current 28-skill catalog and that roles do not duplicate one another. Routing does not complete the underlying analysis.
- Continue by naming the returned skill explicitly: “Use [exact skill name] on these materials within the agreed scope, and deliver its evidence and verification record.” If the route is NONE, proceed with the concrete task directly.
More about this skill
This collection contains 28 independent Agent Skills. Each specifies when it fits, when it does not, how to proceed, what to deliver and when to stop. Fault localization, systemic prevention, risk discovery and product choice use different mechanisms.
thinking-model-router first asks whether a framework is useful, normally returning NONE or one primary skill. It selects up to three only for distinct unresolved questions. You can directly invoke a known match, such as constraints, reversibility or pre-mortem. thinking-model-combination adds roles, sequence and a rule for resolving disagreement.
All 28 current entries are configured for manual invocation. The author’s audit grants none a replicated improvement verdict that clears its performance gates. The portfolio-v1 study made no model calls because prerequisites failed, so its effect was unmeasured. Examine usefulness on your own task without treating installation or historical numbers as an accuracy guarantee.
Sources and verification
Sources checked
Read the pinned README, all 28 skills, plugin and license files, author audit, complete evidence registry and relevant study records. This was a source and document review; no skill installation, model call, author evaluation or example outcome was tested.