templates/.claude/skills/reasoning-sandwich/SKILL.md
Template for pre-reasoning → action → post-verification model allocation
npx skillsauth add baekenough/oh-my-customcode reasoning-sandwichInstall this skill globally with one command. Works with Claude Code, Cursor, and Windsurf.
3 of 9 scanners reported clean
Some scanners were skipped, did not run, or reported a non-clean status. Review each row below.
A model allocation pattern that wraps implementation actions with stronger-model reasoning phases. The "sandwich" structure ensures complex tasks get proper analysis before and verification after the core action.
[Pre-reasoning] → stronger model (opus)
├── Analyze requirements
├── Identify edge cases
└── Define success criteria
[Action] → balanced model (sonnet)
├── Implement solution
├── Generate code/content
└── Execute plan
[Post-verification] → balanced or lighter model (sonnet/haiku)
├── Verify against criteria
├── Check for regressions
└── Validate completeness
| Phase | Recommended Model | Rationale | |-------|------------------|-----------| | Pre-reasoning (analyze/plan) | opus | Complex architectural reasoning, edge case detection | | Action (implement/generate) | sonnet | Optimized for code generation, balanced cost | | Post-verification (review/test) | sonnet or haiku | Structural verification, checklist validation |
| Scenario | Apply Sandwich? | Reason | |----------|----------------|--------| | New feature implementation | Yes | Needs analysis → code → verification | | Bug fix with clear root cause | No | Direct action sufficient | | Architecture decision | Yes | Heavy pre-reasoning, lighter action | | Batch file edits | No | Mechanical action, no reasoning needed | | Security-sensitive changes | Yes | Extra verification phase critical |
This pattern is used by:
structured-dev-cycle — stages map to sandwich phasesevaluator-optimizer — generator/evaluator model selection guidancedeep-plan — research (pre) → plan (action) → verify (post)| Anti-pattern | Problem | Fix | |-------------|---------|-----| | Opus for everything | Wasteful, slow | Reserve opus for reasoning-heavy phases | | Haiku for planning | Insufficient depth | Use opus for complex analysis | | Skipping verification | False completion risk | Always include post-verification phase |
development
Generate and maintain a persistent codebase wiki — LLM-built interlinked markdown knowledge base (Karpathy LLM Wiki pattern)
development
Use the project wiki as RAG knowledge source — search wiki pages to answer codebase questions before exploring raw files
tools
Analyze task trajectories to propose reusable SKILL.md candidates from successful patterns
data-ai
hada.io RSS feed monitoring for AI agent/harness articles with automated /scout analysis