> Evidence-backed AI coding agent CLI
for your current workspace._
A terminal AI coding agent for local BYOK workflows. Hand off unfinished Codex, Claude Code, Cursor, Kimi Code, or Grok work as read-only context, re-ground in the current workspace, then use /loop to write, run project checks, review, and fix. Its Gate returns PASS only with complete, fresh evidence—or stops with a clear blocker.
$curl -fsSL https://wukong.today/install.sh | shDefine. Loop. Read the Gate.
Re-ground in the current workspace, run bounded write/check/review/fix iterations, and read the fresh-evidence result or clear blocker.
> HAND_OFF_THE_TASK
Import supported local session summaries as read-only context, then continue from the files and Git state that exist now.
> WORK_WITH_BOUNDS
Wukong writes, checks, reviews, and fixes against one frozen goal and user-approved limits. If progress stalls, it stops with a clear reason.
> READ_THE_GATE
Complete, fresh evidence can return PASS; blockers, runtime errors, pauses, and user stops remain distinct and explain why.
Build an evidence-backed coding-agent workflow
Concrete commands, safety boundaries, and evidence—not generic AI coding advice.
Run an AI coding agent until tests pass
Use a bounded write, check, review, and fix Loop with an evidence-backed Gate verdict or clear blocker.
Use your own model API key
Understand local orchestration, provider billing, privacy boundaries, and Loop limits.
Hand off work from five supported coding agents
Import read-only context, re-ground in the current workspace, and continue or start an evidence-backed Loop.
▄▄█████▄▄
▄█████████████▄
▄▄█████████████████▄▄
▄███████████████████████▄
▀███████████████████████▀
▀▀█████████████████▀▀
▀███████████▀
▄▀▀▄ ▄███████████▄ ▄█▀
█ ▄██▀███████▀██▄██
▀▄▄▄▀██ ▀█▀█▀ ████▀
▀▀▀ ▀▀▀ ██