docs: backfill no-print decision

CLAUDE.md

@@ -38,8 +38,6 @@ Files skipped: binary files, empty files, `.git`/`.cache`/`.uv-cache` directorie
 
 ## Conventions
 
-- No `print()` in library/CLI source — it's a code smell here. Use
-  `sys.stdout.write(...)` / `sys.stderr.write(...)` with an explicit `\n` in the
-  format string. When fixing output bugs, amend the existing `write` call (add
-  `\n`, change destination) rather than swapping in `print()`. `print()` is fine
-  in tests, scratch scripts, and REPL examples.
+- No `print()` in library/CLI source — use `sys.stdout.write` /
+  `sys.stderr.write` with an explicit `\n`. Rationale + scope:
+  `planning/decisions/2026-06-26-no-print-in-source.md`.

planning/decisions/2026-06-26-no-print-in-source.md

@@ -0,0 +1,40 @@
+---
+status: accepted
+summary: No print() in library/CLI source — use sys.stdout.write / sys.stderr.write with an explicit newline.
+---
+
+# No print() in library/CLI source
+
+**Decision:** Source code never calls `print()`. CLI output goes through
+`sys.stdout.write(...)` / `sys.stderr.write(...)` with an explicit `\n` in the
+written string.
+
+## Context
+
+A proposed output fix swapped `sys.stdout.write(f"Fixing {filename}")` for
+`print(f"Fixing {filename}")` to add the missing trailing newline. The
+underlying bug was a missing `\n`, not the choice of API. Upstream prior art
+(`pre-commit-hooks`) uses `print()`, so the swap looked idiomatic.
+
+Options on the table:
+
+1. Replace the `write` call with `print()` (gets the newline "for free").
+2. Keep the `write` call and add the explicit `\n` to the format string.
+
+## Decision & rationale
+
+Take option 2. `print()` in production code paths is treated as a smell here:
+it hides the newline behaviour, mixes a convenience builtin into deliberate I/O,
+and obscures the destination stream. Keeping explicit `write` calls makes both
+the stream and the terminator visible at the call site, and the original bug is
+fixed in place by adding `\n` rather than switching APIs. The rule is about
+*source*: `print()` remains fine in tests, scratch scripts, and REPL examples.
+
+This is the active behaviour in [`cli`](../../architecture/cli.md) — the
+`Fixing <file>` line is emitted via `sys.stdout.write`.
+
+## Revisit trigger
+
+Reopen if the project adopts a logging framework or a CLI/output library that
+supplies its own print-style helper, making hand-rolled `write` calls the
+inconsistent choice.