Releases: vakovalskii/codbash
v7.5.1 — Neuraldeep visual identity + Pi hardening
What's new
🎨 Visual identity port (#225)
Codbash gets the neuraldeep look:
- Deeper dark palette:
#08090Cbg /#121418panels /#2A2D33borders - Neon green brand accent
#00FF88in dark, AA-safe#10804Ain light - Inter font + antialiased smoothing
- 180ms theme cross-fade on switch
accent-pulsekeyframe for primary CTAs (opt-in via.cta-pulseclass)data-revealCSS scaffold for fade-up animations (JS hook-up follow-up)prefers-reduced-motionrespected everywhere- Monokai theme preserved untouched — kept as the retro third option
🔒 Pi / Oh My Pi resume hardening (#224)
- Non-Pi Resume buttons no longer routed through Pi-specific helper
/api/launchonly forwardsresumeTargetfortool === 'pi'- Scope tightened to what was actually needed in #221
Install / upgrade
npm install -g codbash-app@7.5.1Or pick up via the in-app update prompt.
v7.5.0 — Five new agents, project launcher, subscriptions
What's new
🆕 5 new agents
- Qwen Code (#201, #202, #203)
- GitHub Copilot Chat (VS Code) (#171)
- GitHub Copilot CLI (#207)
- Pi (#221)
- Oh My Pi (#221)
✨ Features
- Project launcher with
▶ New/⟳ Last+ GitHub repo Clone & Add (#210) - Projects/History subtabs + per-launch agent picker (#211)
- Subscriptions management in Analytics → History — track paid plans + API deposits across all 7+ agents (#218)
- Background git fetch worker — auto-refresh on new chat / manual / startup (#213)
- Sidebar customization — collapsible Workspace/Agents/Tools groups + Settings → Sidebar tab (#216)
- Structured message rendering for Codex (#197) and Claude Code (#198)
🐛 Stability
- Codex Desktop dedup: imported Claude sessions no longer appear twice (#220)
- Codex cache invalidation watches
~/.codex/sessionsfor new files (#215) - Active-process mapping no longer misattributes via
fallback-latest(#215) - Worktrees collapse to main repo;
$HOME-as-git-root ignored (#212) - Windows: Resume preserves backslash paths via
Start-Process -WorkingDirectory(#214) - Repo-refresh manager-owned timers cancelled on shutdown + tests now actually run in CI (#217)
- Subscriptions: entries beyond 3 now appear without refresh + XSS hardening (#208)
- Cloud Sync: project paths remap across machines via git remote (#154)
Install / upgrade
```bash
npm install -g codbash-app@7.5.0
```
Or upgrade in place via the in-app update prompt.
Full changelog: 27 commits since v7.4.0 (2026-04-17) → v7.5.0 (2026-05-25).
v7.0.0 — codbash
codedash is now codbash
Install:
npm i -g codbash-app && codbash run
Both codbash and codedash commands work. Old binary prints migration notice.
Migrating:
npm uninstall -g codedash-app
npm i -g codbash-app
codbash run
v6.16.4
Fix: clear Node.js >= 18 version check at startup (instead of cryptic SyntaxError on old Node).
v6.16.3 — community PRs
Merged 6 community PRs:
- #156 @akolotov — detail panel star button sync
- #155 @akolotov — clipboard copy fallback for non-secure origins (remote access)
- #159 @akolotov — proper bind vs browser URL handling
- #157 @akolotov — separate session name from first prompt
- #160 @akolotov — configurable MCP/Skills badges visibility in settings
- #100 @aystream — Warp terminal via Launch Configurations API (replaces keystroke hack)
Closed:
- #128 — sort-by already implemented
- #148 — account suspended
- #161 — 20k line diff (entire codebase overwritten)
Pending rebase (5): #162 (Factory Droid), #131 (Qwen), #150 (WSL), #86, #85 (analytics tabs)
v6.16.2
Fix: dashboard leaderboard now also shows interactions count + GOD OF AUTOMATION badge (was only on public site).
v6.16.1 — GOD OF AUTOMATION
Two metrics:
- Prompts — real human inputs only (tool_result filtered)
- Interactions — all user messages including sub-agents, tool responses
Leaderboard:
- Purple "interactions" count shown when > prompts
- ⚡ GOD OF AUTOMATION badge for users with 5x+ automation ratio (80%+ automated)
Data verified across all 5 agents: Claude, Codex, Cursor, OpenCode, Kiro.
v6.16.0 — honest prompt counts
Breaking change in prompt counting:
Claude Code sub-agent tool_result messages are now excluded from prompt counts. These are auto-generated messages from the agent loop (tool responses, sub-agent results) — not real human prompts.
Before: a session with heavy Agent tool usage showed ~415 prompts
After: same session shows ~25 real human prompts
This affects leaderboard stats, analytics, and session card counts. Numbers will drop for users with heavy sub-agent usage — this is intentional and more accurate.
v6.15.14
v6.15.13
Fixes:
- #152: Settings dropdown now styled with dark theme (options bg + custom arrow)
- #146: OpenCode SQLite JSON path error on Windows (execSync → execFileSync)
- #143: Leaderboard mobile text overlap (deployed to leaderboard.neuraldeep.ru)
- #140: Closed (migrated to self-hosted server)
- #112: Closed (already implemented)