AI Agent OS: SkyDome Taiyi / 天穹-太一

ReviewAudited by ClawScan on May 17, 2026.

Overview

This appears to be a coherent AI workflow/persona toolkit, but it can run user-directed local commands, make HTTP requests, and keep local state or memory files.

Install this if you want a broad Taiyi persona/workflow system. Before letting it act autonomously, review helper scripts, require confirmation for command execution or API calls, and keep secrets out of generated state, memory, review, and dream files.

Findings (5)

Artifact-based informational review of SKILL.md, metadata, install specs, static scan signals, and capability signals. ClawScan does not execute the skill or run runtime probes.

What this means

The assistant may answer and plan differently while the skill is active.

Why it was flagged

The skill deliberately instructs the assistant to adopt a new operating persona and internal loop. This is core to the stated purpose, but it changes how the agent behaves.

Skill content
You are now operating in Taiyi Mode... Your hidden task loop is: Goal → Evidence → Gap → Action → Verify → Answer
Recommendation

Use this skill only when you want the Taiyi workflow/persona active, and explicitly tell the assistant to stop using it if you want normal behavior.

What this means

If the agent or user supplies a dangerous command, it could modify files or the local environment.

Why it was flagged

The benchmark helper can execute an arbitrary user-supplied command. This is expected for a command benchmark tool, but it is broad local execution capability.

Skill content
parser.add_argument('cmd', nargs=argparse.REMAINDER ...); r=subprocess.run(cmd,capture_output=True,text=True)
Recommendation

Only benchmark commands you trust, and require confirmation before running commands that write, delete, publish, install, or change configuration.

What this means

Headers or payloads, including tokens if supplied, will be sent to the target URL.

Why it was flagged

The API smoke-test helper sends user-provided URLs, headers, methods, and data over HTTP. This is purpose-aligned for API testing, but it can transmit whatever the user provides.

Skill content
parser.add_argument('url'); parser.add_argument('-H','--header'...); req=urllib.request.Request(args.url,data=data,method=args.method,headers=headers)
Recommendation

Use the API smoke tester only with trusted endpoints and avoid passing secrets unless the endpoint and request are intended.

What this means

Local memory, task, review, or dream files may keep excerpts from prior work.

Why it was flagged

The dream consolidation helper reads recent task and memory files and writes excerpts into a local dream report. This supports the stated long-term memory/review purpose, but can retain sensitive content if users store it there.

Skill content
for name in ['taiyi-tasks.json','taiyi-memory.jsonl']... p.read_text(errors='ignore')[-4000:]
Recommendation

Keep secrets, private data, and temporary noise out of Taiyi state/memory files, and clean generated state before sharing a project.

What this means

Users have less external context for verifying the publisher or project history.

Why it was flagged

The registry metadata does not identify an upstream source or homepage. There is no installer pulling external code, but provenance is limited.

Skill content
Source: unknown; Homepage: none; No install spec — this is an instruction-only skill.
Recommendation

Review the included files and publisher identity before enabling broad workflow or command-execution use.