code-yeongyu/senpi

pi had nothing (nothing), so I made something (something) — sorry mariozechner-senpai, I went ahead and lovingly soiled your pure pi for you. opinionated fork of badlogic/pi-mono with extension-first additions. ganbare ganbare senpi 頑張れ頑張れ先輩

What it solves

senpi provides a more feature-rich, opinionated runtime for AI coding agents. It addresses the need for a "sane" version of the pi-mono framework by bundling a curated set of built-in extensions and core tweaks that improve agent reliability, context management, and tool execution, specifically tuned for the needs of coding assistants like Dori.

How it works

It is a TypeScript monorepo and a fork of pi-mono. It maintains a minimal core while implementing most of its added functionality as extensions. Key architectural improvements include:

  • Dynamic System Prompts: Replaces static prompts with an adaptive builder that includes forced intent gates, exploration discipline, and per-model tuning.
  • Compaction Pipeline: Uses adaptive thresholds and restoration trackers to manage long-term context and prevent token overflow.
  • Tool-Call Middleware: Implements various text-tool protocols (XML, Hermes, YAML+XML) to support models that lack native function calling.
  • Extension System: Allows users to install additional capabilities (like LSP or AST-grep) via a CLI command (senpi install).

Who it’s for

Developers building AI coding agents, users of the pi-mono ecosystem, and those migrating from the OMO (oh-my-openagent) framework who want a lighter, single-binary CLI experience.

Highlights

  • Built-in Tooling: Includes native web search, bash timeout enforcement, and a permission system for tool calls.
  • gpt-apply-patch: Swaps standard edit tools for Codex-style freeform patching when using GPT models.
  • Context Management: Features an adaptive compaction pipeline to handle session recovery and tool-result truncation.
  • Broad Model Support: Provides specific prompt presets for various model families (GPT, Claude, Kimi).
  • Extensible Architecture: Supports a wide range of pluggable extensions for LSP integration, AST-aware search, and computer-use bindings.

Related

  • Project
    ogulcancelik/pi-extensions

    A set of npm‑based plug‑ins that extend the terminal‑based coding assistant *Pi* with UI footers, session search, auto‑permissions, sub‑agents, worktree handling and other utilities.

  • Project
    mitsuhiko/agent-stuff

    A collection of reusable skills, extensions, and prompt commands for the Pi Coding Agent to enhance its capabilities and workflow automation.

  • Project
    lessweb/deepcode-cli

    A terminal-based AI coding assistant specifically optimized for DeepSeek models, featuring agent skills, MCP integration, and granular permission controls.

  • Project
    HKUDS/CLI-Anything

    CLI‑Anything is a Python framework that automatically generates agent‑friendly Click CLIs for any software, outputting JSON for easy parsing. It includes a 7‑phase generation pipeline (analyze → design → implement → test → document → publish) and a package manager called CLI‑Hub to browse, install, and run a growing catalog of pre‑built CLIs (e.g., GIMP, Blender, Zoom). The project targets AI agents (Claude Code, Cursor, OpenClaw, etc.) that need deterministic, composable command interfaces, and it ships with a comprehensive test suite and security hardenings.

  • Project
    huisezhiyin/sdd-riper

    SDD‑RIPER Light is a lightweight, repo‑native framework that adds a control plane around LLM‑based coding agents. It defines a clear loop—restating goals, creating a minimal spec, awaiting human approval, executing, validating with evidence, and syncing results back—so agents can modify code safely, auditable and recoverable. Four modular “skills” (light, strict, codemap, new‑chat‑ready) cover daily work, high‑risk tasks, unfamiliar codebases, and hand‑offs.