diff --git a/SKILL.md b/SKILL.md
new file mode 100644
index 0000000..b6b239f
--- /dev/null
+++ b/SKILL.md
@@ -0,0 +1,41 @@
+---
+name: python-performance-adrs
+description: >
+  Python performance ADRs and rules.
+  Load this when writing or refactoring performance-sensitive Python code
+  (CLI tools, data pipelines, DB/async boundaries, JSON, etc.).
+---
+
+# Python Performance ADRs (Skill)
+
+Use this skill when:
+- Implementing new Python components.
+- Refactoring hot paths, imports, async boundaries.
+- Touching JSON, DB, collections, string formatting, or error handling.
+
+## How to use
+
+1) Always load:
+   - RULEBOOK.md (concise rules, always valid)
+2) For deeper context, load only the relevant paper:
+   - papers/async-overhead.md
+   - papers/collection-membership.md
+   - papers/json-serialization.md
+   - papers/exception-flow.md
+   - papers/string-formatting.md
+   - papers/memory-slots.md
+   - papers/import-optimization.md
+   - papers/database-patterns.md
+
+## Rules (summary)
+
+- Prefer sync over async unless you truly need concurrency.
+- Use sets for membership checks on large collections.
+- Use orjson for high-volume JSON where possible.
+- Avoid expensive exception paths in hot loops.
+- Use f-strings for formatting.
+- Use __slots__ in high-volume objects.
+- Be cautious with many imports in CLI / startup paths.
+- Treat SQLite writes as a potential bottleneck.
+
+All details are enforced via RULEBOOK.md; never reinvent these patterns.