hip-agent: a 200-line agent harness designed to fit entirely in the prompt
cephaloform · x · 2026-09-08
Jonathan Chang released hip-agent (harness in prompt), a minimal coding agent harness designed for agents rather than humans.
- Core idea: the whole harness is 200 lines of Python, small enough to fit in the prompt so the model can read its source and know exactly how the harness works—no wasted turns figuring out CLI flags or parsing output of tools like codex exec.
- Three design principles: ① the harness fits in the prompt, with only two tools (sh and viewimage); ② the OS is the runtime—config via environment variables, actions as shell commands, subagents as child processes; ③ the rest uses existing protocols—plugins follow Agent Plugins, hooks use Claude Code's hook contract, and sessions are Codex CLI files resumable via codex resume.
- Usage: a few lines of zshrc config, then ./agent "inspect this repository and explain it". Code is open-sourced on GitHub.
More from coding & agent
- Celesto AI launches CelestoFS, petabyte-scale durable workspaces for AI agent sandboxes — aniketmaurya · 2026-09-08
- Astra on low reasoning beats Sol on high, and runs twice as fast, devs confirm — steipete · 2026-09-08
- Critical review of agentic AI offers framework for how much authority to delegate — dair_ai · 2026-09-08
- Codex tip: Astra reads your remaining usage %, so you can budget in plain English — keyanzhang · 2026-09-08
- Rork launches Element Selection: point at any element instead of screenshots — rudrank · 2026-09-08
- Feeding bank statements to an AI agent finds every tax exemption, saving thousands a year — menhguin · 2026-09-08