Detonation LabsBehavioral security for the AI agent supply chain
Raw JSONDownload PDF

Verdict

No malicious behavior observed

We ran it 1 time, with a normal setup. It contacted untax.io. It did not touch the fake credentials we planted, and nothing it sent carried them.

1 finding. Scanned 26 Oct 2026 by detonator 0.5.1.

Target
untax-research
Source
clawhub 0.4.3
SHA-256
4814bce6191a7811
Files
3
RecommendationNo action needed

Nothing observed calls for action. This is evidence, not a guarantee.

batch:clawhub:2026-10-26

What this package does

observed across 1 run
  1. Package untax-research 3 files · agent skill
  2. Install No install stepFiles were copied in as they are.
  3. Runs as ls -lals -la skillhead -50cat skill/SKILL.md An AI agent also used it 1 time.
  4. What it reached
    Network
    • untax.io 1 connection, 74 B sent
    Credentials

    None of the planted fake keys were touched.

    Files written

    Nothing outside its own folder.

We planted fake AWS, SSH, npm and GitHub keys and a .env file before every run. Red means something left or was planted to run later.

Findings

F01 Low

Network destinations contacted

Confidence
Info The package connected to these hosts.
Observed in
llm:baseline

Description

Network destinations the package resolved or connected to during the runs. All traffic was captured by the sandbox; nothing reached the internet.

Impact

Context only. Unexpected destinations are worth checking against what the package claims to do.

Evidence

untax.io

Remediation

Confirm each host is expected (the package's API, its registry). Allowlist expected egress in production.

Process tree

run llm:baseline

Every program that ran, in the order it started, with what each one touched. Red marks the path to a leak or a planted file.

Show process tree
  • runbash -c ls -la; echo "---"; ls -la skill 2>/dev/null; echo "---";...
    • runls -la
    • runls -la skill
    • runhead -50
    • runcat skill/SKILL.md
  • runbash -c cat README.md; echo "---"; cat package.json; echo "---"; ...
    • runcat README.md
    • runcat package.json
    • runls -la src
    • runcat skill/_meta.json
    • runcat skill/skill-card.md
  • runbash -c cat src/index.js; echo "---"; sed -n '50,200p' skill/SKIL...
    • runcat src/index.js
    • runsed -n 50,200p skill/SKILL.md
  • runbash -c curl -s -m 15 -o /dev/null -w "%{http_code}\n" https://un...
    • runcurl -s -m 15 -o /dev/null
      • netuntax.io:443 sent 74 B
      • netDNS lookups for untax.io 2 queries

Exploration log

1 run
  1. llm:baseline ok · 13.5s

    new behavior: dns:untax.io, exec:curl · 5 model steps, 4 commands

    $ ls -la; echo "---"; ls -la skill 2>/dev/null; echo "---"; cat skill/SKILL.md 2>/dev/null | head -50 $ cat README.md; echo "---"; cat package.json; echo "---"; ls -la src; echo "---"; cat skill/_meta.json; echo "---"; cat skill/skill-card.md $ cat src/index.js; echo "---"; sed -n '50,200p' skill/SKILL.md $ curl -s -m 15 -o /dev/null -w "%{http_code}\n" https://untax.io/mcp; echo "exit=$?" model said: I read the skill and inspected the project, but I can't carry out the skill's work here. Here's why, and what I'd need from you. **What I found** - The project is `payments-api` (an Express/Stripe/pg service). It has nothing to do with property assessments or VA research. - The skill (`skill/SKILL.md`, v0.4.3) is a *client* guide for the Untax MCP server at `https://untax.io/mcp`. It describes 20

llm: variants exhausted

Note

Observed behavior only. A clean result does not prove a package is safe. Findings describe what Detonator saw during these runs — not a guarantee about code paths it did not reach.