Releases: askalf/agent
v2.9.10
v2.9.9 — Description update
Updated npm description to match current capabilities.
v2.9.8 — README fix
Fix: README.md now properly included in npm package.
v2.9.7 — README & Docs
Complete README rewrite with nervous system integration docs, organism branding, command reference, architecture diagram.
v2.9.6 — Nervous System Support
Agent CLI now participates in the fleet's nervous system.
- Receives and displays agent messages with urgency levels
- Handles incident alerts from the immune system
- Emits signals after every execution (success/stuck/urgency)
- Enhanced heartbeat with memory and uptime data
- Fleet-wide signal broadcast awareness
v2.9.5 — Version sync
Version sync with AskAlf platform v2.9.5.
v2.6.0 — Hidden Windows Service, Smart Routing
Windows service now runs completely hidden via PowerShell Start-Process -WindowStyle Hidden. No visible terminal window.
v2.5.1 — Smart Execution Mode Routing
Auto-detects whether to use Claude CLI (AI reasoning) or direct shell ($0 cost) based on the task input. Server can also set mode explicitly.
v2.5.0 — Shell Fallback Execution
Shell Fallback
Devices without Claude CLI can now execute tasks directly via shell:
- Windows: PowerShell (
-NoProfile -NonInteractive) - Linux/macOS: bash (
-c)
When Claude CLI is installed, full AI execution is used (unchanged). When it's not, the task input is executed as a shell command directly.
This means any device can be useful immediately after connecting — run diagnostics, scripts, file operations — without needing Claude CLI installed.
Full Changelog: v2.4.8...v2.5.0
v2.4.8 — Autonomous execution permissions
Add --dangerously-skip-permissions to Claude CLI invocation so dispatched tasks can execute shell commands without interactive approval.