Repository navigation
Releases
Releases · gutchapa/opencode-telegram
v1.0.11
v1.0.10
What's fixed
- Misleading fallback text: the 'agent timed out' note was baked into the general inability reply and showed up even when no agent had run (short-chat messages go direct to the local model). It now appears only in the task fallback, i.e. only when the opencode agent genuinely ran and hit the timeout.
- Hardened direct-Qwen call: history is sanitized before sending (valid roles + non-empty content, prompt capped below the context window) to prevent the intermittent HTTP 400; HTTP error bodies are now logged so any future failure is instantly diagnosable; transient errors (429/5xx) retry once.
- Agent timeout raised from 180s to 300s (still env-overridable via OPENCODE_TIMEOUT_MS) so real tasks on the local 9B model have room to finish.
Tested
- Direct-Qwen fast path (hello / time question / demo message) — verified no misleading 'timed out' text
- Agentic fallback chain (task → agent timeout → task fallback)
- Slash commands (/help)
- Bot restarted via launchd, single process, polling healthy
v1.0.9 — conversation history persistence
v1.0.9 — Conversation history persistence
Control OpenCode from your phone via Telegram: 70+ slash commands, run shell commands, approve permissions, manage sessions, and chat with your local AI assistant.
Highlights (1.0.0 → 1.0.9)
- Persist conversation history across bot restarts (v1.0.9)
- Fix
TASK_REstem matching; dropfind/locateshell false positives (v1.0.8) - Catch "attempt to" fake promises; stop "go ahead" from routing to the
gobinary (v1.0.7) - Inspection routing + fake-action fallback filter (v1.0.6)
- Route unknown binaries; drop
sayfrom shell whitelist (v1.0.5) - 70+ slash commands, session management, shell execution, permission approvals
Install
npm i -g gutchapa-opencode-telegram