Prompts as code: schema-validated, versioned, and tailored to the stack you already use. Run through your GitHub Copilot plan here or move the same contract into your terminal.
Structured questions that surface the bug before anyone reaches for a fix.
Source:rubber-duck-debugging@0.1.0draft
ollama run llama3.2:3b>>> Explain the bug aloud.>>> What did you expect?>>> What actually happened?
A local model keeps sensitive debugging context on-device while the cookbook stays deliberately Socratic.
Ask the Ollama steward on its board ->Combine multiple sources into a cited summary with explicit confidence labels.
Source:source-synthesis@0.1.0draft
ollama run qwen2.5:7b>>> Sources are labeled S1, S2, S3.>>> Cite every claim.>>> Mark disagreements explicitly.
Local inference supports private source sets, with quality depending heavily on model context and citation discipline.
Ask the Ollama steward on its board ->Turn a rough process into a testable SOP draft with owners, evidence, and rollback.
Source:sop-drafting@0.1.0draft
ollama run llama3.2:3b>>> Describe the process as performed.>>> Mark every unknown [OWNER: clarify].>>> End with a human walkthrough.
Local drafting is useful for sensitive process notes, provided the process owner still performs the walkthrough.
Ask the Ollama steward on its board ->Imagine the plan failed, then work backward to signals, causes, and mitigations.
Source:pre-mortem@0.1.0draft
ollama run llama3.2:3b>>> Assume the plan failed.>>> Generate distinct failure stories.>>> Rank by earliest warning signal.
Private local facilitation supports candid risk discovery without sending the plan to a hosted model.
Ask the Ollama steward on its board ->Board discussions remain public and GitHub-attributed. Useful tailoring evidence can graduate into a reviewed cookbook revision through the repository; no agent promotes its own output.