Luci · Luci here. System up. What you need?
Luci
Luci here. System up. What you need?
You
Is there a way to always start with these defaults for ccgram - maybe a command or something. So I don't have to go through the directory and cli selector unless I want something different to the default?
Luci
Need context first. Let me check how ccgram launches.
Luci
Service has fixed workdir already. Your pain is the **new-session flow** — picking dir + CLI each time. That's ccgram config. Let me check.
Luci
Defaults already exist as env vars: `CCGRAM_PROVIDER` (default CLI), `CCGRAM_TMUX_START_DIR=/home/lucienne/workspace` (default dir). But selector still shows. Let me read ccgram source to see the new-session flow + whether a skip-selector command exists.
Luci
`ccgram.main` imported but no package dir visible — likely editable install. Let me find the real source.
Luci
Editable install. Source: `/home/lucienne/workspace/ccgram/src/ccgram/`. Let me find the new-session / selector flow.
Luci
`/new` flow lives in `handlers/topics/new_command.py` + `directory_browser.py`. Let me read them.
Luci
Got full architecture. Now check: does a fast-path / default-skip already exist? Grep the start-dir env + directory browser entry.
Luci
`CCGRAM_TMUX_START_DIR` not referenced in source — that env var may be dead. Let me check config.py for browser root + how the directory browser picks its starting dir.
Luci
Grep output mangled by rtk. Let me read the directory browser + callbacks directly to map the flow precisely.
Luci
Flow mapped. New-topic = dir browser → (worktree) → provider pick → mode pick → create. 3-4 taps every time. Let me find the actual create-window entrypoint + the configured default provider so my proposal is concrete.
Luci
`_create_window_and_bind` is the core. Let me read it to confirm the exact signature my proposed command would call.
Luci
Mapped it. Findings:
- **No quick-start exists.** Every new topic = full picker: dir browser → (worktree if git) → provider → mode. 3–4 taps each time.
- **Your `CCGRAM_TMUX_START_DIR=/home/lucienne/workspace` systemd env is dead** — grepped whole ccgram source, nothing reads it. Sets no picker default. Pure no-op.
- Defaults that *do* exist: provider already defaults `claude` (`CCGRAM_PROVIDER`), mode = standard. Just no way to skip the taps.
So: must build it. Core creator `_create_window_and_bind(path, provider, mode)` already exists — a quick path reuses it. Two real design calls before I build (this goes through dev-loop + `make check`):
Luci
Locked: **1-tap default button + `/setdefault`**, defaults `~/workspace · claude · YOLO`.
This is a ccgram feature (multi-file, `make check` gate) → mandatory dev-loop before any edit. Invoking now.
Luci
Dev-loop classified: **Tier 2**. Now branch + read the remaining edit targets precisely.
Luci
Now find where the Telegram slash-command menu (descriptions) is built, so `/setdefault` shows up there.
Luci
Have everything. Writing the plan, then implementing.
Luci
Plan clear. Let me check the existing callback-test harness (mocking style) before writing, then implement everything.
Luci
Harness understood. Writing plan doc, then implementing all files.
Luci
Now implement. Starting with config + store:
Luci
Now the store in `user_preferences.py` — add field, accessors, serialization:
Luci
Formatter ran. Now add the get/set accessors. Let me Read the favorites section to target precisely.
Luci
Now callback_data.py — add the new constants:
Luci
Now `directory_browser.py` — resolver + ⚡ button. Add import first:
Luci
Now wire the ⚡ row into `build_directory_browser`:
Luci
Now `directory_callbacks.py`: imports, `_handle_quick`, dispatch, register. Add imports:
Luci
Now add `CB_DIR_QUICK` to the dispatch and `@register`, plus the `_handle_quick` handler. First the dispatch branch:
Luci
Now add the `_handle_quick` handler after `_show_provider_picker`:
Luci
Now add `CB_DIR_QUICK` to the `@register` decorator list:
Luci
Both import and `@register` already have it (my earlier edit hit the register block). Let me verify:
Luci
Now the `/setdefault` command module. Check `window_query` cwd getter + `safe_reply` location:
Luci
`view_window().cwd` gives bound cwd. Writing the `/setdefault` module:
Luci
Now register `/setdefault` in registry.py + ensure the callback module is imported by `load_handlers`. Check load_handlers:
Luci
`load_handlers` must import the new module so `@register` runs. Add it to the topics import:
Luci
Now register the command in `registry.py` (import, CommandSpec, COMMAND_NAMES):
Luci
Now the Telegram slash-menu entry in `cc_commands.py`:
Luci
Update `/new` welcome text to mention the shortcut:
Luci
Now check for tests asserting exact command count/list that my additions would break:
Luci
`test_registry` derives from `COMMAND_NAMES` (which I updated consistently), so no break. Let me verify the relevant slice + check `test_life_task`:
Luci
Now the tests. Store test:
Luci
Now the quick-start callback test:
Luci
Now document in `ccgram/CLAUDE.md` — bot commands list + config keys:
Luci
Code complete. Now run `make check` (fmt + lint + typecheck + tests):
Luci
Those 10 lint failures are all in files I never touched (`mc_thread_tee`, `restart_trace`, `life_task` — Luci's local fork additions). Let me confirm they're pre-existing, not my regression:
Luci
Confirmed: identical 10 failures on base with my changes stashed — **pre-existing fork lint-debt** (`mc_thread_tee`/`restart_trace`/`status_cmd` local additions never got `# Lazy:` markers). My diff adds zero new violations. A red shared gate blocks every ccgram commit, so I'll annotate those in a **separate** commit (keeps the feature diff surgical). First, run the rest of the gate on my code:
Luci
Need the full ruff list to separate mine from pre-existing:
Luci
Convention is `_context` for unused. And my additions tipped two complexity limits (`config.__init__` 62>60, `build_directory_browser` 11>10). Fixing all — extract helpers + rename unused params. Config first:
Luci
Now `directory_browser.py` complexity — fold the branch into a helper:
Luci
Now fix unused `context` params in `setdefault_command.py` (rename to `_context`) and the unused `pytest` import + `await_args` typecheck in the test:
Luci
Now the test fixes — remove unused `pytest`, guard `await_args`:
Luci
Re-check my files for ruff + typecheck:
Luci
My files are clean (ruff + pyright). Now run the actual tests for my code:
Luci
Happy-path test fails — `_create_window_and_bind` not called. Likely `provider_registry.is_valid("claude")` returns False without bootstrap. Let me check:
Luci
`is_valid` needs bootstrap registration — unit test must patch `provider_registry`. Fix the happy-path test:
Luci
All 14 pass. Now the pre-existing lint-debt — annotate those 10 untouched lazy imports so the shared gate goes green. Let me view them and add `# Lazy:` markers:
Luci
These are Luci-local additions missing `# Lazy:` markers. Let me read each region and annotate precisely:
Luci
main.py has redundant stdlib aliases (`import sys as _sys` when `sys` is already module-level). Let me check the top imports — hoisting the redundant ones is cleaner than interleaving markers:
Luci
I'll keep this behavior-preserving — pure `# Lazy:` markers, zero logic change (this is unrelated code I'm not testing). Block 1:
Luci
lint-lazy now annotated. But `make lint` also runs `ruff check`, and `make typecheck` — both have **more** pre-existing debt in untouched files (hermes.py, life_task.py, mc_thread_tee.py — actual type bugs). I won't fold a fork-wide cleanup into a feature. Let me (1) confirm I broke no existing tests via the full unit suite, and (2) prove the ruff/typecheck debt pre-exists:
Luci
5 failures — triage needed. One is clearly mine (`setdefault_command.py` layering invariant). Let me check that invariant + confirm the others pre-exist via stash:
Luci
`setdefault_command.py` correctly trips the F2 layering invariant — a NEW handler touching `user_preferences`/`thread_router` must be a *deliberate* allow-list entry (same as `directory_callbacks.py`, which sets prefs + reads thread bindings identically). Adding it with justification:
Luci
Now verify the other 4 failures (process_detection ×2, status_cmd ×2) pre-exist on base — stash my work and run just those:
Luci
Confirmed — all 4 fail identically on base (hermes-related fork debt, not mine). My only regression was the layering invariant, now fixed. Re-run full suite to confirm delta = 0:
Luci
Delta = 0 regressions (4971 pass, only the 4 pre-existing hermes failures). Now confirm lint-lazy is green and the remaining ruff/typecheck debt is all pre-existing (not mine):
Luci
lint-lazy now **green**. typecheck debt is mc_thread_tee only (my test error fixed). Let me confirm the 2 main.py ruff errors are the pre-existing `:163` ones, not from my comment edits:
Luci
All clean on my side: **lint-lazy GREEN** (I fixed 10→0), and every remaining ruff/typecheck failure is pre-existing fork debt in untouched files (`hermes.py`, `life_task.py`, `mc_thread_tee.py`, `run_bot` complexity) — none mine. Let me check what the formatter touched before staging, so I commit cleanly:
Luci
The `ruff format` reformatted ~26 unrelated fork files — scope noise I won't commit. Reverting the format-only files, keeping just my functional edits:
Luci
Clean — only functional edits remain. Now commit. Feature commit first (the deliverable), then a separate chore commit for the lint-lazy annotations:
Luci
Both committed. Now the council step (dev-loop 2.6) — dispatching an adversarial reviewer on the feature diff:
Luci
Council: 2 IMPORTANT + 1 latent, all legitimate. Fixing all three.
**Fix 1** — `_handle_quick` error branches leave stale state. Clear browse+worktree state before validation so all exits are covered:
Luci
**Fix 2 + 3** — rename `dir` → `directory` (builtin shadow + convention) and guard empty string:
Luci
Update test call sites (the kwarg `dir=` → `directory=`; the stored `"dir"` key stays):
Luci
Now re-verify: ruff + pyright + the affected tests after council fixes:
Luci
All green (ruff, pyright, 192 tests). Committing the council fixes:
Luci
Feature committed in 3 clean commits. Now set the Luci-specific runtime defaults in `.env` (this is what makes `~/workspace · yolo` actually take effect — not committed to the repo). Read it first:
Luci
Now restart `ccgram.service` to load the new code + config, and verify it comes up clean:
Luci
ccgram restarted clean — **30 bot commands** now (was 29, +1 for `/setdefault`), monitor + polling up, no errors/tracebacks. (The "bot was kicked" warnings are the pre-existing `luci-persistent` window issue, unrelated.) Final check — confirm config resolves the new defaults:
Luci
Config resolves exactly as requested: **`~/workspace · claude · yolo`**. Now finalize — merge the feature branch into ccgram's main so the working tree settles (the editable install runs from the tree). Check branch + merge:
Luci
Two mainlines: `main` (upstream-tracking) and `luci-main` (the local fork with mc_thread_tee/life_task/etc). Let me confirm which the feature branched from + which the service should track: