chore: Add Claude Fix some bugs.

docs/context/archive-rules.md

@@ -0,0 +1,16 @@
+# Archive Rules
+
+## Raw File Archival
+
+After creating a Source Document Summary for any raw file:
+1. Move the raw file to `docs/archive/`
+2. Record the move in the source summary's header: `Archived From: [original path]`
+3. Do not read from `docs/archive/` unless the user explicitly says "go back to the original [filename]"
+
+## Summary Lifecycle Rules
+
+1. **Session handoffs expire**: After a new handoff is written, the previous handoff moves to `docs/archive/handoffs/`. Only the latest handoff stays in `docs/summaries/`.
+2. **Decision records persist**: Decision records (DR-*) stay in `docs/summaries/` permanently — they are institutional memory.
+3. **Source summaries persist**: Source document summaries stay until the project ends — they replace raw documents.
+4. **Analysis summaries**: Keep only the latest version. If re-run, the new one replaces the old (archive the old one).
+5. **Maximum active summaries**: If `docs/summaries/` exceeds 15 files, consolidate older source summaries into a single `project-digest.md` and archive the originals.

docs/context/processing-protocol.md

@@ -0,0 +1,23 @@
+# Document Processing Protocol
+
+Use this whenever you need to process multiple documents or large files.
+
+## For 1-3 Short Documents (< 2K words each)
+
+Read sequentially. After each document, write a Source Document Summary (Template 1 from `templates/claude-templates.md`) to disk. Then proceed with work using summaries only.
+
+## For 4+ Documents OR Any Document > 2K Words
+
+**Step 1:** List all documents with file sizes. Present to user for prioritization.
+
+**Step 2:** Process each document individually:
+- Read one document
+- Extract into Source Document Summary format
+- Write to `./docs/summaries/source-[filename].md`
+- Release the document from active consideration before reading the next
+
+**Step 3:** After all documents are processed, read only the summaries to form your working context.
+
+**Step 4:** Cross-reference summaries for contradictions or dependencies. Note these explicitly.
+
+**Step 5:** Proceed with the actual task using summaries as your reference.

docs/context/subagent-rules.md

@@ -0,0 +1,18 @@
+# Subagent Deployment Rules
+
+## When to Use Subagent vs. Main Agent
+
+| Situation | Approach | Why |
+|-----------|----------|-----|
+| Reading/analyzing documents | Subagent | Keeps source content out of main context |
+| Research and competitive analysis | Subagent | Heavy reading, return summary only |
+| Writing deliverables | Main agent | Needs full decision-making context |
+| Schema/architecture design | Main agent | Needs holistic project understanding |
+| Code generation | Subagent | Isolated implementation, return result |
+| Review and QA | Subagent | Fresh perspective, no bias from writing |
+
+## Output Requirements
+
+Subagent output must conform to the Output Contracts in `templates/claude-templates.md`. No free-form prose returns.
+
+Optimal subagent return size: 1,000-2,000 tokens of structured summary. Longer returns consume main agent context without proportional benefit.