v0.6: jsonl response path + session memory + global CLI + cron hardening

.env.example

@@ -20,6 +20,22 @@ SUMMARIZER_MODEL=claude-sonnet-4-6
 SCROLLBACK_LINES=500
 INTERMEDIATE_INTERVAL=10
 
+# Boot notification — when set to 1 (default), LiteClaw sends a one-shot
+# "🚀 LiteClaw ready" message to your Telegram chat after the daemon
+# finishes initializing. Set to 0 to disable.
+BOOT_NOTIFY=1
+
+# OpenClaw-style memory layout. Per-day transcripts, per-day markdown
+# summaries, rolling strategic compact, and a synthesized startup primer
+# all live under LITECLAW_DIR.
+# LITECLAW_DIR=~/.liteclaw
+# PRIMER_RECENT_TURNS=20
+
+# Pin the cwd Claude Code launches in so `claude --continue` consistently
+# resumes the same project bucket under ~/.claude/projects/<encoded-cwd>/.
+# Defaults to $HOME if unset.
+# CLAUDE_CWD=/Users/you
+
 # File transfer
 STAGING_DIR=~/liteclaw-files
 

.github/ISSUE_TEMPLATE/bug_report.md

@@ -0,0 +1,48 @@
+---
+name: Bug report
+about: Something isn't working as expected
+title: "[bug] "
+labels: bug
+assignees: ''
+---
+
+## What happened?
+
+<!-- A clear, one-sentence description of the bug. -->
+
+## Environment
+
+- **OS**: <!-- macOS 14 / Ubuntu 22.04 / WSL2 / etc. -->
+- **Python**: <!-- python3 --version -->
+- **tmux**: <!-- tmux -V -->
+- **Claude Code CLI**: <!-- claude --version -->
+- **LiteClaw commit**: <!-- git rev-parse --short HEAD -->
+
+## How to reproduce
+
+1. ...
+2. ...
+3. ...
+
+## Expected vs actual
+
+- **Expected**: ...
+- **Actual**: ...
+
+## Logs
+
+<details>
+<summary>Relevant lines from <code>/tmp/liteclaw_run.log</code></summary>
+
+```
+(paste log excerpt here — redact BOT_TOKEN if present)
+```
+
+</details>
+
+<!-- If the issue is about a cron: also paste the relevant entry from
+~/.liteclaw/cron-error-capture.md -->
+
+## Additional context
+
+<!-- Screenshots, unusual config in .env (redact secrets), anything else. -->

.github/ISSUE_TEMPLATE/feature_request.md

@@ -0,0 +1,34 @@
+---
+name: Feature request
+about: Suggest an improvement or new capability
+title: "[feat] "
+labels: enhancement
+assignees: ''
+---
+
+## The problem
+
+<!-- What friction, limitation, or missing capability prompted this? -->
+
+## Proposed solution
+
+<!-- Your preferred shape of the fix/feature. Concrete > abstract.
+     Example: "/recall skill <name> that surfaces all turns touching a given
+     skill" — not "better recall". -->
+
+## Alternatives considered
+
+<!-- Other approaches you thought about and why they're worse. -->
+
+## Scope / impact
+
+- [ ] Config-only (new env var, toggle)
+- [ ] Single-file Python change in `liteclaw.py`
+- [ ] Touches shell scripts (`start.sh`, `setup.sh`, `bin/liteclaw`)
+- [ ] Touches storage layout (`~/.liteclaw/` or `~/.claude/projects/...`)
+- [ ] New Telegram command
+- [ ] Needs external integration (API proxy, bot webhook, etc.)
+
+## Additional context
+
+<!-- Screenshots, links, related issues. -->

.github/workflows/ci.yml

@@ -0,0 +1,62 @@
+name: CI
+
+on:
+  push:
+    branches: [main, feat/*]
+  pull_request:
+    branches: [main]
+
+jobs:
+  syntax-check:
+    runs-on: ubuntu-latest
+    steps:
+      - uses: actions/checkout@v4
+
+      - name: Set up Python
+        uses: actions/setup-python@v5
+        with:
+          python-version: "3.12"
+
+      - name: Python syntax check (liteclaw.py)
+        run: python -c "import ast; ast.parse(open('liteclaw.py').read()); print('liteclaw.py syntax OK')"
+
+      - name: Bash syntax check (shell entry points)
+        run: |
+          bash -n setup.sh
+          bash -n start.sh
+          bash -n bin/liteclaw
+          echo "bash syntax OK"
+
+      - name: Install requirements (import check)
+        run: |
+          python -m pip install --upgrade pip
+          pip install -r requirements.txt
+          python -c "
+          import importlib
+          for mod in ('httpx', 'telegram', 'dotenv', 'apscheduler'):
+              importlib.import_module(mod)
+              print(f'import {mod}: OK')
+          "
+
+      - name: Static import check on liteclaw.py
+        run: |
+          # `python -c "import liteclaw"` would try to spin up the bot; we only
+          # want to verify the module parses and top-level imports resolve.
+          python -c "
+          import ast, sys
+          src = open('liteclaw.py').read()
+          tree = ast.parse(src)
+          top_imports = [n for n in tree.body if isinstance(n, (ast.Import, ast.ImportFrom))]
+          print(f'{len(top_imports)} top-level imports parse cleanly')
+          "
+
+  lint-ruff:
+    runs-on: ubuntu-latest
+    continue-on-error: true  # linting is advisory; don't fail the build while codebase catches up
+    steps:
+      - uses: actions/checkout@v4
+      - uses: actions/setup-python@v5
+        with:
+          python-version: "3.12"
+      - run: pip install ruff
+      - run: ruff check liteclaw.py --output-format=github || true

CHANGELOG.md

@@ -1,5 +1,48 @@
 # Changelog
 
+## v0.6.1 (2026-04-23) — Daemon-restart resilience, /tell-summarizer, community hygiene
+
+### Added
+- **Daemon-restart resilience** — user replies no longer disappear when the daemon is restarted mid-poll.
+  - `~/.liteclaw/pending_replies.json` records every user message whose placeholder ("⏳ 작업 중…") was sent but not yet answered (user_msg_id, placeholder_msg_id, chat_id, jsonl offset, sent timestamp, target).
+  - `_resume_pending_on_boot` runs from `Application.post_init`. For each pending entry it either (a) reads the session jsonl and edits the placeholder with the now-completed answer, (b) spawns a background task to finish polling, or (c) marks the placeholder as "재기동으로 유실됨" if older than `PENDING_MAX_AGE_SEC` (default 900 s).
+  - `run_polling` now passes `stop_signals=(SIGINT, SIGTERM)` explicitly so `pkill liteclaw` triggers PTB's graceful drain. Hard kills / crashes are covered by the persistent-pending path.
+  - New env: `PENDING_MAX_AGE_SEC` (default 900).
+- **`/tell-summarizer` command** (alias `/tellsum`) — runtime control over the summarizer system prompt. `/tellsum <instruction>` appends, `/tellsum show` displays, `/tellsum clear` resets. Persists in-memory only.
+- **Community hygiene** — `CONTRIBUTING.md`, `.github/workflows/ci.yml` (AST parse + `bash -n` + requirements import smoke + advisory ruff lint), `.github/ISSUE_TEMPLATE/{bug_report,feature_request}.md`, shields.io badges in both READMEs.
+
+### Changed
+- **`SUMMARIZE_PROMPT`** gained a **PRESERVE-AS-IS** block — numbered / lettered option lists ("B / C / D 선택") must reach Telegram verbatim. The summarizer previously kept "Pick one" while dropping the option bodies.
+
+---
+
+## v0.6.0 (2026-04-20 → 2026-04-21) — jsonl response path, session memory, global CLI, cron hardening
+
+(Merged via https://github.com/breaktheready/liteclaw/pull/1)
+
+### Added
+- **Global `liteclaw` CLI** — `liteclaw start|stop|restart|status|logs|attach` from any directory. `setup.sh` installs a symlink at `~/.local/bin/liteclaw`.
+- **Pinned Claude Code session via `--session-id <uuid>`** — `start.sh` allocates / adopts a stable UUID in `~/.liteclaw/sessions.json.liteclaw_session_id` and always resumes the same conversation, unaffected by other Claude Code windows in the same cwd.
+- **JSONL response path** (Phase A) — responses are lifted from Claude Code's own session log (`~/.claude/projects/<encoded-cwd>/<id>.jsonl`) instead of scraping the tmux pane. No ANSI chrome, no scrollback truncation, no summarizer over-compression. Automatic fallback to pane-scrape on jsonl failure.
+- **OpenClaw-style memory layout** (`~/.liteclaw/`) — per-day transcripts, daily markdown digests (LLM-compacted), rolling strategic summary, startup primer. Legacy `~/.liteclaw-history.jsonl` auto-migrates.
+- **Compact session-id alias (`sid`)** in transcripts — integer index into `sessions.json.history[]`, ~70 % shorter per row than embedding the full UUID.
+- **`/recall session [uuid]`** — session-scoped recall.
+- **Boot-ready Telegram ping** — one-shot "🚀 LiteClaw ready" after init, rate-limited to 5 min.
+- **`~/.liteclaw/cron-error-capture.md`** — forensic log of failed cron runs (job config + pane snapshot).
+
+### Fixed
+- **`is_idle_prompt` false-positive** on Claude Code UI chrome ("· Claude Max", "· Share Claude Code…") that jammed the cron busy-wait at 120 s. Removed `·` and `*` from `_SPINNER_CHARS`, anchored `_ACTIVITY_PATTERNS` to line start.
+- **Cron trust prompt** auto-accept in `_run_cron_job` for unattended project dirs.
+- **APScheduler weekday indexing** — `1-5` was Tue-Sat instead of Mon-Fri.
+- **Mid-poll status edits** leaking "Thinking (Crystalizing)…" into Telegram. Gated behind `SHOW_POLLING_STATUS` (off by default).
+- **`_followup_edit` overwriting delivered messages** with stale status edits when jsonl path had already returned a complete turn. Now skipped when `_jsonl_delivered_complete` is set.
+
+### Changed
+- Placeholder text "📤 Sent. Waiting for response…" → "⏳ 작업 중…".
+- `setup.sh` plants sensible `~/.tmux.conf` defaults when missing (mouse on, history-limit 50000).
+
+---
+
 ## v0.5.0 (2026-04-17) — CLI Mirror, Draft Streaming, Reasoning Lane & Skill System
 
 ### Added

CLAUDE.md

@@ -250,9 +250,10 @@ Ensures response delivery even when infrastructure is degraded.
 ## Auto-Recovery Mechanisms
 
 ### API Proxy Recovery
-- On Tier 1 failure: runs `docker compose up -d` in the proxy directory
+- On Tier 1 failure: runs `docker compose up -d` in the proxy directory (Linux hosts)
 - Retries connection after restart
 - Notifies user via Telegram on successful recovery
+- **macOS note**: proxy runs as LaunchAgent, not Docker — recovery would no-op. launchd's `KeepAlive` handles crash restarts automatically. See [MAC-OPS.md](./MAC-OPS.md).
 
 ### Session Auth Recovery (401)
 - Detects "401" + auth keywords in tmux pane output
@@ -350,10 +351,11 @@ Direct `tmux send-keys -l` crashes on quotes, newlines, and special characters.
 
 ### Infrastructure: API Proxy
 LiteClaw's Tier 1 summarizer depends on an OpenAI-compatible API proxy.
-- If down: auto-recovery attempts `docker compose up -d`
+- If down: auto-recovery attempts `docker compose up -d` (Linux/auto-recovery path only)
 - Manual check: `curl -s http://localhost:3456/v1/models`
-- Recommended proxy: [mattschwen/claude-max-api-proxy](https://github.com/mattschwen/claude-max-api-proxy) — Docker compose, reuses Claude Max subscription
+- Recommended proxy: [mattschwen/claude-max-api-proxy](https://github.com/mattschwen/claude-max-api-proxy) — reuses Claude Max subscription
 - **Degraded without proxy** — Tier 2/3 still work but quality drops
+- **macOS**: Docker cannot access the user keychain where Claude stores OAuth tokens. Run the proxy as a LaunchAgent instead — see [MAC-OPS.md](./MAC-OPS.md).
 
 ### Startup Sequence
 

CONTRIBUTING.md

@@ -0,0 +1,85 @@
+# Contributing to LiteClaw
+
+Thanks for considering a contribution. LiteClaw is a small, opinionated
+single-file Python bridge — changes should match that shape.
+
+## Ground rules
+
+- **Keep the core a single file.** `liteclaw.py` intentionally ships as one
+  file so operators can read / fork / deploy with zero ceremony. Factor into
+  separate modules only when the piece is genuinely standalone (e.g., a new
+  executable under `bin/` or an isolated helper).
+- **No new Anthropic API key.** The project's reason-for-being is "reuse
+  your existing Claude Code CLI session." A patch that introduces a hard
+  dependency on `ANTHROPIC_API_KEY` or a specific vendor SDK will be closed.
+- **Fallbacks are non-negotiable.** Anything that relies on an external
+  service (summarizer proxy, jsonl path, pyannote, etc.) must degrade
+  gracefully, not crash, when the external piece is unavailable.
+- **Prefer environment variables over new CLI flags.** LiteClaw is
+  configured via `.env`. New knobs should land in `.env.example` with a
+  one-line description and a sane default.
+
+## Project structure
+
+```
+liteclaw/
+├── liteclaw.py              # core daemon (~5k lines, single file)
+├── start.sh                 # tmux session + session-id pin
+├── setup.sh                 # one-shot install (deps, venv, CLI symlink)
+├── bin/liteclaw             # global CLI dispatcher
+├── .env.example             # all configuration knobs
+├── README.md / README_KO.md # user docs
+├── MAC-OPS.md               # macOS proxy/launchd notes
+└── DEVNOTES.md              # NOT shipped — session-scoped devnotes (gitignored)
+```
+
+## Local sanity checks
+
+Before opening a PR run at least:
+
+```bash
+python3 -c "import ast; ast.parse(open('liteclaw.py').read()); print('OK')"
+bash -n setup.sh start.sh bin/liteclaw
+```
+
+The GitHub Actions workflow `.github/workflows/ci.yml` runs the same checks
+plus a dependency import test on every push / PR.
+
+## Commit style
+
+- Small, logical commits. Split a feature from refactors from docs.
+- Title prefix: `[feat]`, `[fix]`, `[docs]`, `[chore]`, `[refactor]`.
+- Body: wrap at ~80 chars, explain the **why** and cite line numbers for
+  non-obvious changes.
+
+## Telegram-side behavior changes
+
+If your change affects what ends up in the user's Telegram chat (summarizer
+prompt, follow-up edits, jsonl path, etc.), include a short "user-visible
+diff" section in the PR description — ideally a before/after screenshot or
+transcript snippet. Regressions in Telegram UX have been subtle and hard to
+notice in code review alone.
+
+## Issues and discussion
+
+- Bug report → [template](.github/ISSUE_TEMPLATE/bug_report.md). Please
+  include `/tmp/liteclaw_run.log` excerpts (redact `BOT_TOKEN`).
+- Feature request → [template](.github/ISSUE_TEMPLATE/feature_request.md).
+  Concrete proposals move faster than abstract wishes.
+
+## Security
+
+If you find a security issue (credential leak, RCE surface, etc.), please
+do NOT open a public issue. Contact the author directly via GitHub
+(@breaktheready) with details.
+
+## Release flow
+
+Maintainer-only:
+
+1. Merge PRs into `main`.
+2. Tag `vX.Y.Z` on `main`.
+3. GitHub Release with notes summarizing the "What's new in vX.Y" section
+   from the README.
+
+Thanks for making LiteClaw sharper.