Prompt-Studio v1.0: the workbench + the layer, one repo
Analyze, score, optimize, and ship a persona that follows your agent across seven hosts.
Read more →We build the tooling and the persona layer that make LLM output shorter, cheaper, and closer to what you actually meant.
7-dimension quality breakdown with letter grade. Clarity, specificity, context, format, mode-fit, token efficiency, constraints.
/A.01Rule-based improvement pass adds missing persona, format, examples, constraints. Filler-token compression without semantic loss.
/A.02Lean persona layer cuts LLM output size, cost, and latency. One SKILL.md, mode-filtered (lite/full/ultra), zero drift across hosts.
/A.03Analyze scores, token counts, and context-window fit across GPT, Claude, Gemini, Llama, Mistral, and DeepSeek. Optimize with rule-based passes, compress with filler-token stripping. Persist every run to PostgreSQL with pgvector semantic search.
Ships as a Claude Code plugin, a Codex adapter, a Copilot CLI plugin, a Devin plugin, a Qoder ruleset, and instruction files for Cursor / Windsurf / Cline / Kiro. Same source of truth, filtered per intensity.
Standalone stdio server for MCP hosts whose only injection point is the prompt menu. Zero drift with the FastAPI adapters · both call the same get_lean_instructions().
Analyze, score, optimize, and ship a persona that follows your agent across seven hosts.
Read more →One SKILL.md, three payloads. Ultra for long agentic sessions, lite for cost-sensitive calls.
Read more →LOC, tokens, USD, latency across five arms. Anthropic + OpenAI + Gemini + Ollama backends.
Read more →Stdio server for hosts whose only hook is the prompt menu. Zero drift with the FastAPI adapters.
Read more →