wow-agent

Product

Changelog

Format based on Keep a Changelog; versions follow SemVer.

v0.6.0 — current release

The safety & scale release.

Added

  • Four-layer safety system: unshare -n network sandbox, egress guard (curl POST/scp/git push/npm publish), dangerous-command double confirmation, snapshot rollback
  • /review three-level code review (🔴 high / 🟡 medium / 🟢 low) via review sub-agent
  • AST-based Python code index: symbols, references, definitions with incremental updates
  • Long-term memory: /mem save/use/rm/list/new
  • MCP client: run any Model Context Protocol server, tools auto-discovered
  • Bilingual UI (EN/ZH) with /language
  • ARM64 support, tested on Jetson Orin NX

Changed

  • State machine hardened: failed tasks auto-rollback every snapshot
  • /model --fetch pulls live model lists from providers
  • Session persistence improved: /resume across restarts

Performance

  • Agent cold start under one second
  • Index incremental update ~50 ms per changed file

v0.5.0

Added

  • Sub-agent system: fresh-context research agents with structured results
  • Diff visualization for every file modification
  • /compact [hint] guided history compression

Fixed

  • Snapshot integrity across sessions (/undo --verify)

v0.4.0

Added

  • Provider presets: DeepSeek, Qwen, Kimi, GLM, SiliconFlow, OpenAI, Gemini, Ollama, LM Studio
  • /model --fetch online model listing
  • Shell escape !command with safety checks

v0.3.0

Added

  • Task tree planning with user approval
  • Snapshot-based /undo (single and multi-step)
  • /status dashboard

v0.2.0

Added

  • Tool-calling loop with JSON-schema validation
  • File tools: read/write/edit/delete/list/glob/grep
  • Shell tool with timeout control

v0.1.0

  • Initial public release: interactive terminal REPL, OpenAI-compatible chat, basic planning

Older detail lives in GitHub Releases.