diff options
Diffstat (limited to 'CLAUDE.md')
| -rw-r--r-- | CLAUDE.md | 8 |
1 files changed, 8 insertions, 0 deletions
@@ -85,3 +85,11 @@ Prefer Write over cumulative Edits for nontrivial new code. Small functions (und - **Warn at module load when an external tool path is configured but missing.** Calling `cj/executable-find-or-warn` (from `system-lib.el`) at `:config` time emits a `display-warning` if `prettier` / `pyright` / `pandoc` / etc. isn't on PATH, instead of letting the first format-on-save or LSP-attach fail with a confusing mid-edit error. Pattern in use: `modules/prog-webdev.el` (prettier), `modules/prog-python.el` (pyright). (`pattern` — 2026-05-16) - **ghostel F-key / prefix bindings need `ghostel-keymap-exceptions` + a rebuild, not just `ghostel-mode-map`.** In semi-char mode ghostel forwards every key not in `ghostel-keymap-exceptions` to the pty, and `ghostel-semi-char-mode-map` (rebuilt from that list, and outranking the major-mode map) wins. So binding F9 / F12 / C-; in `ghostel-mode-map` alone is silently dead inside agent/terminal buffers — the key reaches the shell, not Emacs. Fix: add the key to `ghostel-keymap-exceptions` AND call `ghostel--rebuild-semi-char-keymap` (`add-to-list` updates the list but not the already-built map). `term-config.el` (C-;, F12) and `ai-term.el` (F9 family) do this in their `with-eval-after-load 'ghostel`. This is the opposite of vterm, where binding in `vterm-mode-map` sufficed. (`gotcha` — 2026-06-05) + +- **Rulesets-owned changes propagate by edit-local + send-copy + explanatory note.** A bug or enhancement that belongs to a rulesets-owned synced file (a workflow under `.ai/workflows/`, a skill, a rule under `.claude/rules/`, a script under `.ai/scripts/`) is handled by editing the local copy so it's usable now, then sending rulesets a copy of the edited file plus an explanatory note — a local edit alone is overwritten on the next template sync, so the canonical update is what makes it durable. The note covers: how the problem was hit, what outcomes the change should alter, any implementation recommendations, and any follow-up instructions (e.g. send a note back with more info). Send notes with the inbox-send script (`inbox-send rulesets --file <path>`). Offer the change proactively when it would help. (`pattern` — 2026-06-12) + +- **Manual-verification handoff: VERIFY child, close the originating task.** When a task's code is complete and the only thing left is Craig's hands-on check, don't leave the whole task stuck in DOING. File the manual check as a `VERIFY` child under the "Manual testing and validation" parent task, then close the originating task itself (DONE for a top-level task, or the dated-log rewrite for a sub-task, per `todo-format.md`). The VERIFY child carries the residual human check and surfaces in the agenda until Craig confirms; the implementation task is no longer held open by it. Replaces the prior habit of leaving the fixed task DOING with a `*** TODO` manual-test note under the parent. (`pattern` — 2026-06-13) + +- **`make test` runs with no `package-initialize` — defuns inside a `use-package :config` are void there.** The Makefile's `EMACS_TEST` is `emacs --batch --no-site-file --no-site-lisp` with no `package-initialize`, so elpa packages never load and a `use-package` block whose package isn't found never runs its `:config`. Any `defun` nested inside that `:config` is unbound under `make test` / `make test-file`. The per-edit PostToolUse hook *does* initialize packages, so such defuns load there — a test can pass on save under the hook yet fail `make test`. To unit-test logic that lives in a `:config` block, extract it into a top-level defun outside `use-package` (the `cj/dwim-shell--empty-dirs-command` / `cj/dwim-shell--dated-backup-command` pattern) and test that; keybindings or mode-wiring that must stay in `:config` get live-daemon verification instead. (`gotcha` — 2026-06-13) + +- **Don't `cl-letf`-mock C primitives in tests — it triggers a native-comp trampoline rebuild that fails under `--batch`.** Mocking a primitive like `buffer-modified-p`, `file-exists-p`, or `kill-buffer` via `cl-letf`/`fset` makes native-comp try to compile and load a trampoline `.eln`, which errors under `emacs --batch` (`native-lisp-load-failed "file does not exists" .../subr--trampoline-*.eln`, often after a "Redefining 'X' might break native compilation of trampolines" warning). Don't mock the primitive: drive the real state instead (a `make-temp-file` fixture so `file-exists-p` is true for real, `insert`/`set-buffer-modified-p` for modified state, `buffer-live-p` to detect a kill), or extract the decision logic into a pure helper and test that. Mocking ordinary Lisp functions (`y-or-n-p`, `save-buffer`, `info`) is fine — the trap is specific to subrs. (`gotcha` — 2026-06-13) |