Changelog

v0.9.4

Watch auto-reconnect backoff + Pilsner REST token + test coverage expansion.

watch --reconnect-delay <sec>: configurable initial reconnect backoff (default 2s, doubles each attempt).
watch --reconnect-max-delay <sec>: cap on reconnect backoff (default 60s).
watch --max-reconnect default changed from 5 → 10.
watch --json now emits a {"type":"reconnecting","attempt":N,"delay_secs":D,"reason":"..."} NDJSON line on each reconnect attempt.
KakaoCredentials carries an optional rest_token field for the ~138-char Cache.db bearer token required by talk-pilsner.kakao.com endpoints.
resolve_base_credentials() auto-populates rest_token from Cache.db on startup — best-effort, non-blocking if Cache.db is locked.
REST client selects the correct token per endpoint: rest_token for pilsner, oauth_token for katalk.
auth_flow: Cache.db copy timeout no longer propagates as an error — gracefully falls back to Ok(None) so commands work even when KakaoTalk is locking the directory.
Added 14 new unit tests: 11 for export.rs (ExportFormat parsing, resolve_author, JSON/CSV/TXT format) + 3 JSON output parsing tests (doctor, auth-status, cache-stats). 257 tests total, zero warnings.

Local message cache integration.

watch now persists incoming messages (MSG and SYNCMSG packets) to a local SQLite cache.
read merges local cache with LOCO-fetched messages, deduplicating by logId. Displays [read] Merged N messages from local cache when cache contributes new messages.
read also back-fills the cache with LOCO results so future reads benefit from previously fetched history.
New get_messages(chat_id, limit) API on MessageDb for ordered message retrieval.
3 new integration tests for get_messages (20 total message_db tests).

Edit command + completion promise.

edit <chat_id> <log_id> <message>: edit messages via LOCO REWRITE command. Returns -203 on macOS (dtype=2 platform restriction) — implemented for future Android support.
--completion-promise: global flag that prints [DONE] to stdout after any command completes successfully. Designed for LLM agent integration where the caller needs a reliable completion signal.
E2E verified: MemoChat + Christine (send/completion-promise pass, edit blocked by macOS -203 as expected).
240 tests passing, zero clippy warnings.

React command + SYNCACTION handler.

react <chat_id> <log_id>: add reaction to a message via LOCO ACTION command (type=1 = like).
SYNCACTION push handler in watch: real-time reaction events from other users now appear in watch output.
CI fix: version integration test now uses env!("CARGO_PKG_VERSION") instead of a hardcoded string.
240 tests passing, zero clippy warnings.

Delete, mark-read, and protocol capture.

delete <chat_id> <log_id>: delete a message via LOCO DELETEMSG. Works in DM chats (creates feedType:14 deletion marker). Requires -y for non-interactive use.
mark-read <chat_id> <log_id>: mark messages as read up to a logId via NOTIREAD (fire-and-forget).
watch --capture: output unknown LOCO push packets as structured NDJSON for protocol reverse engineering.
SYNCDLMSG handler: delete events from other users now trigger hooks/webhooks in watch mode.
SYNCREWR handler: edit event placeholder (pending protocol confirmation).
probe --capture-pushes: extended timeout (10s) idle listen mode for capturing push packets.
Open-chat safety checks for delete (same --force override as send).

Stability, security, quality, tests, and performance.

230+ tests across unit, integration, and CLI smoke tests.
Security hardening: zeroize sensitive memory, validate outbound messages, enforce HTTPS for remote webhooks.
Error model: OpenKakaoError with is_retryable() for smarter reconnect decisions across all command paths.
Recovery state persistence: auth-status shows failure history, cooldowns, and last recovery source.
Performance: rate limiting for unattended sends, hook/webhook execution, and open-chat reads.

Documentation overhaul.

Comprehensive website docs update for v0.7.1 features: MemoChat usage, LOCO error codes, watch state recovery, timestamp validation, rate limiting guidance, encryption fragility warning, and changelog page.
Legacy command naming cleanup: loco-read, loco-chats → read, chats across all examples.
Added missing --unattended --allow-watch-side-effects flags to automation examples.
Expanded credential loading order with detailed descriptions and token type guidance.
LOGINLIST sync documentation in protocol pages (commands.mdx, connection.mdx).

LOGINLIST chatLog sync for faster reads.

read now passes the target chatId in the LOGINLIST chatIds/maxIds arrays during connection. The server returns recent messages (including received ones) in the login response's chatLog field, making the initial read faster and more complete.
New sync_chat_ids field on LocoClient (Vec<(i64, i64)>) allows commands to request specific chat data from LOGINLIST before the first SYNCMSG round-trip.
Messages from LOGINLIST chatLog are automatically deduplicated against SYNCMSG results by logId.
Cache.db-free auto-relogin via email_cmd with 3-tier fallback for credential recovery.

Polish, Error Model, Output Consistency.

OpenKakaoError adopted across watch, send, read, and loco_helpers paths with is_retryable() support for smarter reconnect decisions.
Profile module split: profile.rs (1784 lines) refactored into a 5-file profile/ directory (mod.rs, hints.rs, graph.rs, probe.rs, app_state.rs).
--json output added for send, send-file, and watch (NDJSON streaming).
Dead code cleanup: zero #[allow(dead_code)] remaining, removed unused variants and functions.
All 165+ tests passing, zero clippy warnings.