fix: patterns mode is iterative — keep extracting until done

Process: Discovery → Deepening → Edge cases → Cross-reference.
Repeat until scanning yields no new patterns.
Target: 50-200+ patterns for a language stdlib, 15+ per topic.

SKILL.md

@@ -242,13 +242,39 @@ Output: `<language>-patterns` or `<ecosystem>-patterns` repo.
 codebase?" Filter everything through: "If I were writing new code
 in this language/ecosystem, what rules does this source teach me?"
 
-**`patterns/<topic>.md`** — one file per topic area:
-- Error handling patterns
+**This is iterative, not one-shot.** Keep extracting until you've
+identified ALL patterns the source demonstrates. A first pass finds
+the obvious ones. Second pass greps for variations and edge cases.
+Third pass finds the patterns that break. You're done when scanning
+the source no longer reveals new rules.
+
+**Process:**
+1. **Discovery pass** — scan the source by topic area, identify every
+   distinct pattern (aim for 15-30+ per topic in a large codebase)
+2. **Deepening pass** — for each pattern, grep for 5-10 real usages
+   across the codebase. Note variations. Find the best example.
+3. **Edge case pass** — find where each pattern DOESN'T apply.
+   Grep for violations — are they bugs, or legitimate exceptions?
+4. **Cross-reference pass** — which patterns interact? Which ones
+   conflict? Document the decision framework for choosing between
+   competing patterns.
+
+Repeat until scanning the source yields no new patterns. A language
+stdlib should produce 50-200+ patterns across all topics.
+
+**Output structure — one file per topic:**
+
+`patterns/<topic>.md` — topics include:
+- Error handling
 - Naming conventions
 - Concurrency patterns
 - Testing patterns
 - Interface/protocol design
 - Module organization
+- Documentation conventions
+- Performance idioms
+- Configuration patterns
+- Extension/plugin patterns
 
 Each pattern entry:
 - Name (short, linkable heading)
@@ -270,6 +296,10 @@ Each pattern entry:
 structure, TODO culture, and project history unless they directly
 inform HOW to write code. Focus on patterns a user should copy.
 
+**Done criteria:** You've scanned every major directory in the source.
+No new patterns emerge from further grep/read. Each topic file has
+15+ patterns with real examples. Edge cases are documented.
+
 ---
 
 End all output files with `<!-- PATTERN_COMPLETE -->` sentinel.