Adonis0123/adonis-skills / skills/kimi-computer-use
kimi-computer-use
VerifiedInstall, register, diagnose, or upgrade Kimi Computer Use (kimi-cu) MCP for Cursor CLI, Grok, WorkBuddy, Claude Code, Codex, or Hermes; fix kimi-cu tool-call errors such as 'no target app' or 'mode must be full, image, or ax'; or look at and verify a running macOS app UI through kimi-cu, such as checking whether a window matches a design. Not for general desktop automation, messaging, or purchasing, and not a replacement for a host's native Computer Use.
| File | Kind | Size |
|---|---|---|
| SKILL.md | skill md | 7.7 KB |
| evals/check_scope.py | script | 5.3 KB |
| evals/evals.json | config | 9.1 KB |
| evals/trigger-eval.json | config | 2.1 KB |
| references/host-configs.md | markdown | 4.3 KB |
| references/maintenance.md | markdown | 2.9 KB |
evals/check_scope.py
View on GitHub#!/usr/bin/env python3
"""Check that SKILL.md keeps UI checks in scope and requires real kimi-cu tool calls.
Run: python3 skills/kimi-computer-use/evals/check_scope.py
Exits 0 only when every check passes; each failure says what to fix.
"""
import argparse
import json
import re
import sys
from pathlib import Path
SKILL_DIR = Path(__file__).resolve().parent.parent
UI_PROMPT = "用 kimi 看一下当前桌面上的设置窗口是不是按稿子做的"
INSTALL_PROMPT = "帮我把 Kimi Computer Use 装到 Cursor CLI 上"
def load():
text = (SKILL_DIR / "SKILL.md").read_text(encoding="utf-8")
match = re.search(r'^description:\s*"(.*)"\s*$', text, re.M)
if not match:
sys.exit("FAIL: SKILL.md frontmatter has no quoted description line")
evals = json.loads((SKILL_DIR / "evals" / "evals.json").read_text(encoding="utf-8"))["evals"]
triggers = json.loads((SKILL_DIR / "evals" / "trigger-eval.json").read_text(encoding="utf-8"))
return text, match.group(1), evals, triggers
def eval_for(evals, prompt):
return next((e for e in evals if e["prompt"] == prompt), None)
def main():
global SKILL_DIR
parser = argparse.ArgumentParser(description=__doc__)
parser.add_argument("--skill-dir", type=Path, default=SKILL_DIR,
help="Skill snapshot to check; defaults to this skill")
SKILL_DIR = parser.parse_args().skill_dir
text, desc, evals, triggers = load()
body = text.split("\n---\n", 1)[1]
should_trigger = {t["query"]: t["should_trigger"] for t in triggers}
ui_eval = eval_for(evals, UI_PROMPT)
ui_expect = " ".join(ui_eval["expectations"]) if ui_eval else ""
checks = [
(
"1. UI-check prompt is in scope as a real tool-using check",
re.search(r"verify a running .*UI through kimi-cu", desc)
and should_trigger.get(UI_PROMPT) is True
and "get_app_state" in ui_expect
and "get_app_state" in body,
"describe UI verification in the description, add the prompt to evals.json with a "
"get_app_state expectation, and mark it should_trigger=true in trigger-eval.json",
),
(
"2. a UI check with zero kimi-cu tool calls is UNVERIFIED",
re.search(r"zero `?kimi-cu`? tool calls is `?UNVERIFIED`?", body)
and re.search(r"never emits a call", body),
"state in 'Operate through MCP' that a UI check with zero kimi-cu tool calls is "
"UNVERIFIED, and say why (the host can list tools while the model never emits a call)",
),
(
"3. installing to Cursor CLI stays in scope",
desc.startswith("Install") and "Cursor CLI" in desc
and eval_for(evals, INSTALL_PROMPT) is not None,
"keep 'Install ... Cursor CLI' in the description and the install prompt in evals.json",
),
(
"4. DOM, Console, Network, Performance belong to Chrome DevTools",
re.search(r"DOM, Console, Network, and Performance belong to Chrome DevTools MCP, not `kimi-cu`", body),
"keep the 'Browser internals ... belong to Chrome DevTools MCP, not kimi-cu' rule",
),
(
"5. description does not refuse UI verification once installed",
not re.search(r"already works|once kimi-cu|already installed", desc, re.I),
"remove wording that excludes use after kimi-cu works; exclude general automation instead",
),
(
"6. argument correction requires a successful read-only retry",
"retry the read-only `get_app_state` call" in body
and "Stop only after the corrected call completes successfully" in body
and "If the corrected call still fails, continue the layered checks below" in body
and "If the retry cannot run, report runtime verification as `UNVERIFIED`" in body
and "The service is fine" not in body,
"require a read-only retry after correcting arguments; continue diagnosis on failure "
"and mark an unavailable retry UNVERIFIED instead of declaring service health",
),
(
"7. Hermes fuse attribution requires preceding-call evidence",
"Attribute it to bad arguments only when the preceding failed calls show those errors" in body
and "otherwise keep the cause unresolved" in body
and "Do not infer permission health from the fuse message" in body,
"condition bad-argument attribution on preceding-call evidence and leave other causes unresolved",
),
(
"8. offline eval covers corrected timeout and unavailable retry",
any(e.get("name") == "corrected-call-fails-and-fuse-cause-unknown"
and "timed out" in e["prompt"]
and "UNVERIFIED" in " ".join(e.get("expectations", []))
for e in evals),
"add an offline eval for a corrected-call timeout, missing fuse evidence, and unavailable retry",
),
]
failed = 0
for name, ok, fix in checks:
if ok:
print(f"PASS {name}")
else:
failed += 1
print(f"FAIL {name}\n fix: {fix}")
print(f"{len(checks) - failed}/{len(checks)} passed")
return 1 if failed else 0
if __name__ == "__main__":
sys.exit(main())
Related skills
| Skill | Stars | Updated | Works with | Status | Install |
|---|---|---|---|---|---|
| brainstormingobra/superpowersYou MUST use this before any creative work - creating features, building components, adding functionality, modifying behavior, or planning anything new, in software or out of it (a talk, a business, a renovation). | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |
| brand-guidelinesanthropics/skillsApplies Anthropic's official brand colors and typography to any sort of artifact that may benefit from having Anthropic's look-and-feel. Use it when brand colors or style guidelines, visual formatting, or company design standards apply. | Stars: 198k | Updated | Claude CodeCodex | Verified | Install |
| frontend-designanthropics/skillsGuidance for distinctive, intentional visual design when building new UI or reshaping an existing one. Helps with aesthetic direction, typography, and making choices that don't read as templated defaults. | Stars: 198k | Updated | Claude CodeCodex | Verified | Install |
| mcp-builderanthropics/skillsGuide for creating high-quality MCP (Model Context Protocol) servers that enable LLMs to interact with external services through well-designed tools. Use when building MCP servers to integrate external APIs or services, whether in Python (FastMCP) or Node/TypeScript (MCP SDK). | Stars: 198k | Updated | Claude CodeCodex | Verified | Install |