/** Maximum existing lessons to include for dedup context */
export declare const MAX_EXISTING_LESSONS = 10;
/** Maximum assembled prompt size in characters */
export declare const MAX_PROMPT_CHARS = 100000;
/** System prompt for review-learn — instructs LLM to extract lessons from resolved bot findings */
export declare const REVIEW_LEARN_SYSTEM_PROMPT = "You are Totem's lesson extractor for bot code review findings.\n\nYou receive a set of code review findings from automated bots (CodeRabbit, Gemini Code Assist) that were RESOLVED (the developer accepted and fixed them). Your job is to extract reusable architectural lessons from these findings.\n\nRULES:\n1. Only extract lessons that represent reusable patterns \u2014 NOT one-off fixes.\n2. Each lesson must be actionable: what the symptom is, what the fix is, and why it matters.\n3. Every lesson MUST include lifecycle: nursery in its metadata \u2014 these are unproven until validated.\n4. Deduplicate against the provided existing lessons. Do NOT repeat known patterns.\n5. Focus on architectural and security findings. Skip pure style/formatting nits unless they represent a real pattern.\n6. If no findings warrant a lesson, return an empty array.\n\nOUTPUT FORMAT:\nReturn a JSON array of lesson objects. Each lesson has:\n- \"tags\": string[] \u2014 relevant tags (e.g., [\"security\", \"typescript\", \"architecture\"])\n- \"text\": string \u2014 the lesson body. Start with the symptom/pattern, then the fix.\n- \"lifecycle\": \"nursery\" \u2014 REQUIRED, always \"nursery\"\n\nExample:\n[\n  {\n    \"tags\": [\"security\", \"shell\"],\n    \"text\": \"Using execSync with string interpolation for shell commands creates injection risk. Use spawnSync with an args array to pass arguments safely without shell interpretation.\",\n    \"lifecycle\": \"nursery\"\n  }\n]\n\nIf no lessons should be extracted, return: []";
//# sourceMappingURL=review-learn-templates.d.ts.map