Skip to content

CHANGELOG.md

Latest commit

 

History

History
92 lines (70 loc) · 4.97 KB

CHANGELOG.md

File metadata and controls

92 lines (70 loc) · 4.97 KB
Raw

Changelog

v0.6.1 (2026-03-19) — Dual timeout for stuck agents

Fixes

  • Dual agent timeout replaces the old single 30-min timer:
    • Idle timeout (10 min default, AGENT_IDLE_TIMEOUT env var): reset on any stdout activity. An agent that produces no output for 10 minutes is considered stuck and killed. This catches the "hung waiting on API" case that was causing 30-40 min hangs.
    • Absolute ceiling (45 min default, AGENT_ABSOLUTE_TIMEOUT env var): never resets, regardless of any activity. Hard cap for runaway agents producing garbage output.
  • Both timeouts log their reason (idle vs absolute) to make post-mortems easier.

v0.6.0 (2026-03-19) — Reliability + remote control

Fixes

  • Infinite retry loop eliminated: scheduleRetry no longer resets retryCount after MAX_RETRIES. Previously, a group that hit max retries would silently reset the counter and retry forever.
  • Orphaned agent processes from previous runs are now killed on startup. PIDs are tracked in data/agent-pids.json and cleaned up on boot, preventing slot starvation and timeout cascades after a crash or forced restart.

New

  • /status command: shows active agents per group, queue depth (pending tasks + messages), and uptime. Available via Telegram.
  • /skills command: lists all installed skills with descriptions, read live from .claude/skills/. Available via Telegram.
  • Telegram command menu: setMyCommands() called at startup so all commands appear with descriptions when the user types /.
  • GroupQueue.getStatus(): exposes live queue state (active count, waiting groups, per-group task/message queues) for external consumers.

v0.5.5 (2026-03-18) — Remote control hotfix

Fixes

  • /reset now clears the pending task and message queue, not just kills the current agent. Previously, resetting would immediately drain the next queued item, making it impossible to stop a runaway queue remotely.
  • /reset reply updated to confirm both kill and queue clear.

New

  • /update command: pulls latest code from git, rebuilds, and restarts via launchd. Allows remote updates over Telegram without SSH access. This is a bootstrapped command — requires one manual git pull && npm run build to get it onto a running instance, after which all future updates can be done via Telegram.

v0.5.0 (2026-03-04) — Public Beta

First public beta release. GhostClaw is feature-complete and ready for testing.

New features

  • Mission Control dashboard with live status, chat management, task scheduling, soul editing, logs, and research tabs
  • Research tab with inline file editor (click to view/edit output, download links)
  • Ralph autonomous task loop — run multi-step research and task checklists overnight
  • Telegram file delivery — research output sent as documents after Ralph completion
  • Queue acknowledgment — "Got it, finishing a task first..." when messages arrive during tasks
  • Typing indicator stays alive during long agent responses
  • Voice transcription via ElevenLabs Scribe (replaced OpenAI Whisper)
  • Ghost personality throughout dashboard (loading quips, save feedback, footer rotation)

Fixes

  • Ralph task ID collisions on retry
  • Error alert spam from transient WhatsApp disconnections
  • Research tab no longer shows unrelated project files (README, CHANGELOG, etc.)

Cleanup

  • Removed experimental add-parallel skill
  • Consolidated Gmail skills (single add-gmail-agent skill)
  • Updated voice transcription skill code package to ElevenLabs
  • Updated all docs: README, CLAUDE.md, .env.example, SKILL.md files
  • Debug skill rewritten for direct process model (no more Docker references)

v0.2.1 (2026-03-03)

  • PID file lock prevents duplicate instances (fixes Telegram 409 and WhatsApp conflict errors)
  • Mission Control dashboard (built-in web UI)
  • Branded README with GhostClaw mark
  • Repo transferred to b1rdmania/ghostclaw

v0.2.0 (2026-03-03)

GhostClaw identity release. Full independence from NanoClaw naming and architecture.

  • Complete rename: all env vars, MCP servers, directories, package names now ghostclaw/GHOSTCLAW_*
  • Settings-based MCP server configuration — agents install MCP servers via standard settings.json, no code changes needed
  • Agent environment model: CLAUDE_CONFIG_DIR for session isolation, HOME left untouched so tools find credentials naturally
  • Reserved MCP server names, shape validation, source-of-truth global sync
  • Ralph autonomous task loop — run multi-step tasks overnight from a checklist
  • Telegram formatting (bold, italic, code blocks)
  • Morning briefing skill
  • Cleaned up stale NanoClaw assets, docs, and duplicate files

v0.1.1 (2026-03-01)

  • Gmail integration via MCP
  • Voice transcription (Whisper)
  • Heartbeat monitoring
  • Skills engine with security scanning

v0.1.0 (2026-02-28)

  • Initial fork from NanoClaw
  • Containers removed — agents run as direct Node.js child processes
  • Telegram as primary channel
  • WhatsApp group chat support
  • Scheduled tasks (cron, interval, one-shot)
  • Per-group personality system