v2.4.5 — The Thinking Skills Release

Two thinking skills, one clear boundary. meow:problem-solving ships seven non-default techniques for when you're stuck on approach. meow:sequential-thinking gains three diagnostic-framework references for when you're stuck on cause. Both sit on distinct axes and cross-reference each other explicitly.

Highlights

• New skill meow:problem-solving — seven strategic-unsticking techniques (simplification cascades, collision-zone thinking, meta-pattern recognition, inversion, scale game, first principles, via negativa). Dispatch table routes by stuck-symptom; explicit boundary reroutes debugging requests to meow:sequential-thinking.
• meow:sequential-thinking extended with three diagnostic frameworks — five-whys-plus.md, scientific-method.md, kepner-tregoe.md. Core workflow unchanged; new references load on demand when a specific methodology fits.
• Curation over absorption — audit of 39 third-party thinking frameworks (cc-thinking-skills) landed a total of five (two in problem-solving, three in sequential-thinking). The remaining 34 stayed out by design.

The Boundary

A long-standing gap in the meowkit skill catalog: what do you load when you're "stuck" but not debugging?

You're stuck on...	Skill	Trigger example
Cause (why is this broken?)	meow:sequential-thinking	"This test is failing and I don't know why"
Approach (which way do I build this?)	meow:problem-solving	"I've implemented this 5 ways and it still feels wrong"

Both descriptions state the boundary explicitly so LLM triggering stays precise.

New Skill: meow:problem-solving

Seven techniques, each with a crisp trigger symptom:

Technique	Red Flag (trigger phrase)	Win Shape
Simplification Cascades	"Just need to add one more case..."	10× (things deleted), not 10%
Collision-Zone Thinking	"I've tried everything in this domain"	Emergent properties from metaphor
Meta-Pattern Recognition	"This problem is unique"	Pattern in 3+ domains = universal
Inversion Exercise	"There's only one way to do this"	Valid inversion reveals context-dependence
Scale Game	"Should scale fine" — no evidence	Production-ready before production
First Principles	Everyone accepts a constraint blindly	Rebuild from fundamentals
Via Negativa	"Adding X to fix problem from Y"	Remove to surface the real issue

Dispatch

user stuck on approach
   ├→ match symptom → apply ONE technique → resolved ✓
   ├→ not resolved → try a combination
   ├→ actually stuck on cause → route to meow:sequential-thinking
   └→ actually a multi-perspective trade-off → route to meow:party

Anti-patterns (the Gotchas section)

• Tool-stacking — running 3 techniques at once hides which one helped.
• Debugging misroute — "my code is broken" is not problem-solving territory.
• Premature abstraction — collapsing 2 instances; meta-pattern rule requires 3+ domains.
• Invalid inversion — "trust all user input" is a security hole, not a valid inversion.
• First-principles tourism — listing assumptions ≠ challenging them.

Sequential-Thinking: Diagnostic Frameworks

Three optional methodologies when general hypothesis-testing isn't structured enough:

Reference	Use when
five-whys-plus.md	Post-mortems, recurring problems, human-error investigations — bias guards + stopping criteria
scientific-method.md	A/B tests, performance investigations, production incidents — falsifiable prediction discipline
kepner-tregoe.md	Multi-system bugs, contested root causes — IS/IS-NOT matrix (SA/PA/DA/PPA)

All three adapt the same output contract as the existing hypothesis-testing.md template — they're methodologies feeding the same table, not parallel workflows.

Curation Decisions (Why Only Five, Not 39)

A full audit of cc-thinking-skills' 39 frameworks ran against the existing catalog. Rejection reasons:

• Already covered — inversion, first-principles variants duplicated by the included techniques
• Overlaps existing meow skill — pre-mortem / red-team → meow:cso; steel-manning → meow:party; debiasing → meow:review; socratic → meow:elicit
• Out of scope — decision-making (regret-minimization, reversibility), personal-decision (circle-of-competence), product-strategy (jobs-to-be-done, effectuation), estimation (fermi, lindy)
• Too obvious — occams-razor, second-order, cynefin (a capable LLM already does these)

The audit found one borderline INCLUDE (thinking-archetypes — Senge's 7 system-dysfunction patterns). Tech-lead rejected on YAGNI grounds: no user-reported gap, dispatch-table precision preferred over catalog breadth. Deferred to future releases if evidence emerges.

Files Changed

Area	Files
New skill	.claude/skills/meow:problem-solving/SKILL.md + 9 references (when-stuck, 5 claudekit techniques, first-principles, via-negativa, attribution)
Skill extensions	.claude/skills/meow:sequential-thinking/SKILL.md (References section expanded) + 3 new refs
Registry	.claude/agents/SKILLS_INDEX.md (new row)
Docs	website/reference/skills/problem-solving.md, sequential-thinking.md, index.md, skills-index.md, changelog.md, guide/whats-new/v2.4.5.md

Attribution

• claudekit-engineer/problem-solving v2.0.0 — five core techniques + dispatch pattern
• microsoft/amplifier — original insight-synthesizer agent lineage (MIT, commit 2adb63f)
• cc-thinking-skills — first-principles, via-negativa, and the three diagnostic frameworks (MIT)

Upgrade

npx mewkit upgrade

Fresh install: npx mewkit init.