Changelog

Release notes and version history for CodeMantis.

v1.5.0

May 2026

Per-Task Agent Routing

  • Per-task agent routing — main_chat, spec_writer, assistant, and help can each be routed to Claude Code or Codex independently (Settings > Agents)
  • 7-day usage-split card — per-agent token counts and estimated cost breakdown by agent
  • Sparse defaultAgentByTask settings map — categories you don't touch follow your primary agent
  • Self-Drive note: still inherits the session's agent (not a per-task category) to keep evidence-vocabulary consistent within a single autonomous run

v1.3.0

May 2026

OpenAI Codex as a First-Class Agent

  • OpenAI Codex is now a first-class agent alongside Claude Code
  • Project Picker shows an Agent Picker when both `claude` and `codex` are installed
  • Sessions carry an `agent_id`; the input toolbar adapts per agent
  • Codex Policy Pill (sandbox × approval-policy) replaces the Mode Selector for Codex sessions
  • Welcome Screen recognizes both agents — either alone is enough to start using CodeMantis
  • `codex login` / `~/.codex/auth.json` for OAuth against your ChatGPT Plus/Pro/Business subscription
  • Refreshed Assistant model lineup: Gemini 3.x preview, GPT-5.5 family, Opus 4.7 / Sonnet 4.6 / Haiku 4.5

v1.2.0

April 2026

AgentAdapter Abstraction

  • Internal refactor: every Claude path moved through a pluggable AgentAdapter trait
  • Streaming protocol, session lifecycle, and tool-approval flow converge on a single interface
  • Foundation for v1.3.0's Codex support — additional agents become a config, not a rewrite

v1.1.11

April 2026

SpecWriter Capability Handshake + Self-Drive Evidence Vocab

  • SpecWriter capability handshake — probes the project for real capabilities (Supabase keys, BrowserMCP, LLM creds) and live-fires them before writing
  • Coverage tab — automatic UI-completeness audit shown after spec generation
  • AUDIT-PATCH — "Patch spec & re-audit" asks the model for a splice patch instead of regenerating from scratch
  • Self-Drive evidence vocabulary — orchestrator parses Claude's verify replies against a typed vocab
  • Loop guard — bounded retry on repeat-pattern failures
  • Capability gating — verify items tagged `capability=<id>` auto-resolve as N/A when the capability is absent

v1.1.10

March 2026

Preflight, Mission Control, and Encryption at Rest

  • Preflight / Mission Control — per-project capability gate with a green/yellow/red PreflightTray at the top of the workspace
  • Mission Control setup wizard for every API key, secret, and tool a project needs
  • SpecWriter auto-writes `preflight.yaml` on spec finalization
  • Self-Drive refuses to start with unsatisfied blockers
  • AES-GCM encryption of AI provider API keys at rest
  • Graduated wake recovery — post-suspend wake re-attaches live sessions and reloads if needed

v1.1.8

March 2026

CLI Session Crash Recovery

  • Persistent CLI session id — force-quit sessions are resumable
  • Crash recovery on relaunch — pick up where you left off

v1.0.0

March 2026

Initial Release

  • Three-panel layout with chat, activity feed, and file browser
  • SpecWriter — AI-powered specification writing with verification checklists
  • 10+ project templates with CLAUDE.md generation
  • Multi-AI assistants (OpenAI, Google Gemini, Anthropic)
  • Integrated preview browser with viewport presets
  • MCP server management with 15+ templates
  • Integrated terminals and Monaco editor
  • 6 themes (Midnight, Ocean, Ember, Dawn, Sand, Arctic)
  • AI-powered changelog generation
  • Slash commands and CLI overlay
  • Auto-updates via GitHub Releases
  • Context meter with usage warnings
  • Session history and error recovery

For the full release history, visit GitHub Releases.