diff --git a/lazy-tdd/SKILL.md b/lazy-tdd/SKILL.md
new file mode 100644
index 0000000..7db4f04
--- /dev/null
+++ b/lazy-tdd/SKILL.md
@@ -0,0 +1,50 @@
+---
+name: lazy-tdd
+description: Lazy TDD / test-driven debugging — use when a bug, goof, regression, or "that didn't work" is discovered in code you wrote; when fixing a failing behavior; when refactoring; or when deciding whether new code needs tests. Code that works the first time may ship untested; a discovered goof means you owe a test that reproduces it before (or with) the fix.
+---
+
+# Lazy TDD
+
+From Dan Connolly's
+[2006 resolution: no more undocumented, untested code](https://www.madmode.com/2006/advogato_entry0044.html)
+(MadMode, Mar 2006):
+
+> It's best to write the tests first … For personal projects, I'm not
+> disciplined enough to write the tests first every time; if the code I
+> write works the first time, I sometimes let myself get away with it.
+> But I'm doing pretty well about doing test-driven debugging, at least.
+> I write tests for any code that doesn't work the first time. And I
+> write tests when I refactor and change things. The confidence to make
+> changes that comes from having tests in place is very freeing.
+
+## the deal
+
+1. **Got it right the first time? No test owed.** Writing tests first is
+   still best, but it's not required here. Don't pad a change with tests
+   nobody asked for.
+2. **Goof discovered? You owe a test.** Any bug in code you wrote — found
+   by you, the operator, CI, a reviewer, or a runtime failure — earns a
+   regression test. No fix ships without it.
+3. **Refactoring or changing behavior? Write tests** covering what you're
+   about to touch, so the change is made with confidence.
+
+## paying the debt (test-driven debugging)
+
+1. Write a test that reproduces the goof, in the project's existing test
+   framework and style, next to related tests.
+2. Run it; **watch it fail** for the reason you expect. A test that
+   passes before the fix proves nothing about the fix.
+3. Fix the code.
+4. Run it; watch it pass. Run the neighboring suite too.
+5. Commit the test with the fix (same commit, or test first).
+
+Name the test for the behavior, not the bug ticket: "rejects negative
+amounts", not "fix for goof #3".
+
+## when a test is genuinely impractical
+
+If the goof can't reasonably be captured in an automated test (e.g. it
+lives in config, a one-off script, or untestable I/O with no seam), say
+so explicitly, explain why, and suggest the seam that would make it
+testable (see [ocap-discipline](../ocap-discipline/SKILL.md) for
+injecting I/O). Don't silently skip the debt.