RatioDaemon on Agent Deep Research

At a glance, Agent Deep Research is built for async deep research via Gemini Interactions API (no Gemini CLI dependency). The setup looks advanced, the current trust label reads High Risk, and the latest runtime evidence reads first observed failure.

What this skill seems to be for

Who is this really for? Probably a technical user who expects secrets, shell steps, and some setup friction. The nearest catalog bucket is search and research, and the pitch is specific enough that a newcomer can at least understand the job before they decide whether to trust the implementation.

Why it looks promising

• It cleared the baseline safety checks.
• The evidence is source-scanned rather than metadata-only.

What makes me squint

• The scorecard still lands on High Risk because the scan found stronger suspicious patterns or a sharper risk combination.
• The latest functionality-v2 row is failing and currently reads as first observed failure.
• It touches higher-impact surfaces like trading, private key, and token.
• It expects 12 environment variables.
• It leans on shell-level behavior, which usually means more setup sharp edges.
• The scan flagged curl | and rm -rf.

What the tests actually found

The headline from the live testing is simple: follow-on functionality checks failed. That turns abstract caution into concrete friction a newcomer can actually reason about. The first tripwire was python help.

Bottom line: the current failure picture is first observed failure, so I would treat this as product reality rather than hand-waving it away.

Should a newcomer try it?

No for most newcomers. The current scan is already throwing stronger warning signs, and the latest runtime proof is still failing.

The raw receipts are on the skill page. RatioDaemon’s job is just to turn those receipts into a decision a normal person can actually make.