primelisted

Load targeted knowledge from `.agents/knowledge/` to inform the current session. ## Process ### 1. Determine Focus Area If the user provided a `focus` argument, use it. Otherwise, infer from: - The current git branch name - Recent git log (last 3-5 commits) - Any files currently modified (git diff --name-only) ### 2. Load and Filter Knowledge Read all `.jsonl` files in `.agents/knowledge/`: ``` .agents/knowledge/ ├── patterns.jsonl ├── gotchas.jsonl ├── decisions.jsonl ├── anti-patterns.jsonl ├── codebase-facts.jsonl └── api-behaviors.jsonl ``` For each entry, check relevance by matching: - `tags` against the focus keywords - `affectedFiles` against files being modified - `type` against the work type (e.g., "bug fix" → prioritize gotchas + anti-patterns) ### 3. Prioritize by Relevance Sort matched entries: 1. **High confidence + direct file match** → Always show 2. **High confidence + tag match** → Show if ≤10 results 3. **Medium confidence + file match** → Show if ≤15 results 4. **Everything else** → Skip (available via direct read if needed) ### 4. Present Knowledge Brief Output a compact brief: ``` ## Knowledge Brief: [focus area] ### Must-Know (directly relevant) - 🔴 [anti-pattern or gotcha]: ... - 🟢 [pattern]: ... ### Context (good to know) - 📐 [decision]: ... - 📦 [codebase-fact]: ... ### Caution - ⚠️ [gotcha]: ... --- *N entries loaded from knowledge base. Last reflection: [date]* ``` **Rules:** - Keep the brief under 30 lines — this is a primer, no