# ctxray

> See how you really use AI — X-ray your AI coding sessions locally

- **URL**: https://www.freshcrate.ai/projects/ctxray
- **Author**: ctxray
- **Category**: MCP Servers
- **Latest version**: `v2.2.1` (2026-04-01)
- **License**: MIT
- **Source**: https://github.com/ctxray/ctxray
- **Homepage**: https://getreprompt.dev
- **Language**: Python
- **GitHub**: 43 stars, 6 forks
- **Registry**: github
- **Tags**: `ai`, `ai-security`, `aider`, `analytics`, `claude-code`, `cli`, `codex`, `cursor`, `prompt-engineering`, `python`

## Description

See how you really use AI — X-ray your AI coding sessions locally

## Recent releases

| Version | Date | Urgency | Changes |
| --- | --- | --- | --- |
| `v2.2.1` | 2026-04-01 | Medium | ### MCP Server: 6 → 9 tools - `check_prompt_quality` — full diagnostic (score + lint + rewrite) - `build_prompt_from_parts` — construct prompts from components - `explain_prompt_quality` — educational plain-English analysis  ### File input All prompt commands now accept `--file`: ```bash reprompt check --file prompt.txt reprompt score --file my-prompt.md --json reprompt rewrite --file draft.txt --diff ```  Also supports stdin with `-`: ```bash echo "fix the bug" \| reprompt check - ```  **1,864 t |
| `v2.2.0` | 2026-04-01 | Medium | ## What's New  ### `reprompt build` — Prompt Builder Assemble well-scored prompts from components: ```bash reprompt build "fix the auth bug" \   --file src/auth.ts \   --error "TypeError: Cannot read 'exp'" \   --constraint "don't modify tests" \   --model claude ``` Model-aware formatting: XML tags for Claude, markdown for GPT.  ### `reprompt check` — Unified Diagnostic Score + lint + rewrite in one command: ```bash reprompt check "your prompt" ``` Shows dimensional breakdown, strengths, sugges |
| `v2.1.0` | 2026-04-01 | Medium | ## What's New  ### Model-specific lint rules `reprompt lint --model claude/gpt/gemini` checks prompts against model-specific best practices:  - **Claude**: Suggests XML tags (`<context>`, `<instructions>`) for structured prompts - **GPT**: Warns about XML echo issues, missing markdown headers, JSON mode requirements, CoT anti-pattern for o-series - **Gemini**: Detects broad negatives ("do not infer/guess/assume"), warns about prompt length  8 rules based on official model documentation.  ### Dif |
| `v2.0.2` | 2026-04-01 | Medium | ## What's Changed  ### Scoring Rebalance - **Structure** weight reduced (25 → 15), **Clarity** weight increased (15 → 25) - A clear plain-text prompt now scores **55-65** (green) instead of 35-45 (orange) - Well-structured prompts with markdown still score 75-85 - Badge color thresholds adjusted: green ≥ 60, gold ≥ 40  ### Positive UX Feedback - **Tier labels**: Badge and CLI show `GOOD` / `STRONG` / `EXPERT` instead of raw numbers - **Point-gain hints**: Suggestions show expected score gain: "A |
| `v2.0.0` | 2026-04-01 | Medium | ## reprompt 2.0: Score, Rewrite, Optimize  v2.0 marks the shift from passive prompt analysis to **active prompt coaching**. The headline feature: `reprompt rewrite` — the first CLI that improves your AI prompts automatically, using research-backed rules. No LLM needed.  ### `reprompt rewrite` — the core of v2.0  ```bash $ reprompt rewrite "I was wondering if you could maybe help me fix the auth bug"    34 → 52 (+18)    ╭─ Rewritten ─────────────────────────────────────╮   │ Help me fix the auth |
| `v1.6.0` | 2026-03-28 | Medium | ## What's New  ### Agent Workflow Analysis `reprompt agent` detects error loops, tool call patterns, and session efficiency from existing session files. Zero config, zero instrumentation — reads your Claude Code and Codex CLI JSONL files directly.  ### Codex CLI Adapter (9th adapter) Full support for OpenAI Codex CLI sessions, including user/assistant turns, tool calls (shell + function), error detection with exit codes, and file path tracking.  ### Sensitive Content Detection `reprompt privacy |
| `v1.5.0` | 2026-03-25 | Medium | ## What's new  **Instant dashboard** — just type `reprompt` to see your prompt health at a glance: total prompts, sessions, average score, top categories, and recent activity. New users see guided onboarding; returning users get instant intelligence.  **Session type detection** — `distill` now auto-classifies conversations (debugging, feature-dev, exploration, refactoring, config, learning) and adapts importance scoring per type. A debugging session weights error recovery higher; a feature-d |

## Citation

- HTML: https://www.freshcrate.ai/projects/ctxray
- Markdown: https://www.freshcrate.ai/projects/ctxray.md
- Dependencies JSON: https://www.freshcrate.ai/api/projects/ctxray/deps

_Generated by freshcrate.ai. Indexes github releases for AI-agent ecosystem packages._