@getpromptly/cli

Developer prompt analytics CLI — capture, analyze, and review AI coding sessions tied to tickets.

Install

npm i -g @getpromptly/cli

Quick Start

promptly init            # Configure AI tools + install /track skill
promptly start           # Prompts for ticket ID interactively
# ... work with your AI coding tool ...
promptly finish          # Save session data
promptly serve           # Open dashboard at localhost:3000
promptly report          # View stats (select time period)
promptly digest          # Weekly insights digest with trends

Native /track Commands

After promptly init, you can use the /track command directly in your AI tool:

| Tool | Command | |------|---------| | Claude Code | /track AUTH-123 | | Codex CLI | /track AUTH-123 | | Gemini CLI | /track AUTH-123 | | VS Code + Copilot | /track AUTH-123 |

All tools support: /track <ticket-id>, /track status, /track finish

Auto-Prompt

During setup, promptly init can enable auto-prompt — your AI tool will offer session tracking at the start of each conversation without any manual action.

vs ccusage / Codeburn: those parse Claude Code's local JSONL logs to show your token spend. Promptly adds multi-tool coverage (Cursor, Codex CLI, Gemini CLI, VS Code), git/ticket attribution, session intelligence, and an optional team dashboard.

Commands

| Command | Description | |---------|-------------| | promptly init | Configure MCP (interactive tool selector) | | promptly start | Start logging (prompts for ticket ID) | | promptly start <ticket-id> | Start logging for specific ticket | | promptly finish | Finish session and save data | | promptly status | Show current session status | | promptly serve | Start local dashboard on localhost:3000 | | promptly report | Show stats (interactive period selector) | | promptly digest | Weekly insights digest with trends | | promptly login | Authenticate with cloud API | | promptly teams | List teams with quick actions | | promptly team set | Interactive team selector | | promptly team set <slug> | Set default team by slug | | promptly team unset | Clear default team | | promptly skill install | Install /track command for all configured tools | | promptly skill uninstall | Remove /track skill | | promptly skill status | Check skill installation status | | promptly import-claude | Import a past Claude Code session from ~/.claude/projects (no MCP setup needed) | | promptly optimize | Analyze session history and surface AI spend leak recommendations | | promptly optimize --apply <rec-id> | Apply a recommendation: generates the skill / instruction rule / hook it suggests | | promptly review <session-id> | Run prompt review against a captured session (LLM-as-judge) | | promptly review --pr <n> | Review every session behind a GitHub PR — prompt quality (LLM-as-judge) + spend efficiency in one verdict. Flags: --comment (post/update as a PR comment), --status (pass/fail commit status on the PR, --status-threshold <n>), --no-quality, --rubric <id>, --model <id>, --json | | promptly statusline install | Show 🔴 REC <ticket> in Claude Code's status line while recording | | promptly telemetry on\|off\|status | Manage anonymous usage telemetry |

All commands work without a TTY (scripts, CI, AI agents): prompts fall back to sensible defaults or fail with the flag to pass instead. Use promptly init --tools claude --yes for fully unattended setup.

Telemetry: promptly init sends a one-time anonymous install ping (version + OS, never content); usage telemetry only runs in cloud mode after login. Opt out any time: promptly telemetry off or DO_NOT_TRACK=1.

All commands support direct arguments for scripting/automation.

How It Works

Claude Code / Gemini CLI / Codex CLI
        |
        └──> MCP Server → ~/.promptly/buffer.json → SQLite
                                                        ↓
                                          promptly serve (localhost:3000)

1. promptly init auto-detects installed tools and lets you select which to configure.
2. promptly start creates a session and begins logging.
3. The MCP server captures all conversation turns, tokens, and tool calls.
4. promptly finish captures git activity (commits, branch, diff stats, instruction file changes), auto-categorizes the session (bug-fix, feature, refactor, etc.), computes session intelligence (quality score, tool usage, subagent stats, context window metrics, prompt quality analysis), writes data to SQLite, and clears the buffer.
5. promptly serve serves a dashboard with sessions, analytics, digest, and session replay.

Data Storage

All data stays on your machine at ~/.promptly/:

| File | Purpose | |------|---------| | config.json | CLI configuration | | session.json | Active session state | | buffer.json | Conversation buffer (cleared on finish) | | promptly.db | SQLite database with all sessions |

Links

• Website
• GitHub
• Documentation

License

MIT