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 made no network connections. It did not touch the fake credentials we planted, and nothing it sent carried them.

0 findings. Scanned 26 Oct 2026 by detonator 0.5.1.

Target
taste-pm
Source
clawhub 1.0.0
SHA-256
de6b78ad83a858b8
Files
7
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 taste-pm 7 files · agent skill
  2. Install No install stepFiles were copied in as they are.
  3. Runs as ls -Rcat SKILL.mdhead -5cat assets/产品设计稿模板.md An AI agent also used it 1 time.
  4. What it reached
    Network

    No connections.

    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

No findings. Detonator observed no malicious or suspicious behavior in the runs below.

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 cd ./skill && ls -R && echo "-----" && cat SKILL.md | hea...
    • runls -R
    • runcat SKILL.md
    • runhead -5
  • runbash -c cd ./skill && cat assets/产品设计稿模板.md
    • runcat assets/产品设计稿模板.md
  • runbash -c cd ./skill && cat references/字节产品范式库.md
    • runcat references/字节产品范式库.md
  • runbash -c cd ./skill && cat references/设计系统落地词汇表....
    • runcat references/设计系统落地词汇表.md
    • runhead -80 references/产品拆解案例.md
  • runbash -c cd ./skill && sed -n 80,200p references/产品拆解案ä¾...
    • runsed -n 80,200p references/产品拆解案例.md
    • runsed -n 1,60p references/设计系统落地词汇表.md
    • runcat skill-card.md _meta.json
  • runbash -c cd ./skill && sed -n 60,80p references/设计系统落地...
    • runsed -n 60,80p references/设计系统落地词汇表.md
    • runsed -n 60,80p references/产品拆解案例.md
    • runls -la ..
  • runbash -c cd .. && cat README.md package.json .gitignore && ls -la ...
    • runcat README.md package.json .gitignore
  • runbash -c cd /; pwd; ls -la; find / -maxdepth 3 -name package.json ...
    • runls -la
    • runhead
    • runfind / -maxdepth 3 -name package.json
  • runbash -c cd /home 2>/dev/null; ls -la /home /mnt /opt /srv /tmp 2>...
    • runls -la /home /mnt /opt /srv
    • runhead -40
    • runfind / -maxdepth 4 -name skill
  • runbash -c ls -la /opt/detonationlabs-stage; find /opt/detonationlab...
    • runls -la /opt/detonationlabs-stage
    • runfind /opt/detonationlabs-stage -maxdepth 3 -name skill
    • runhead

Exploration log

1 run
  1. llm:baseline ok · 41.6s

    no new behavior · token budget reached

    $ cd ./skill && ls -R && echo "-----" && cat SKILL.md | head -5 $ cd ./skill && cat assets/产品设计稿模板.md $ cd ./skill && cat references/字节产品范式库.md $ cd ./skill && cat references/设计系统落地词汇表.md && echo "=====" && head -80 references/产品拆解案例.md $ cd ./skill && sed -n 80,200p references/产品拆解案例.md && echo "=====" && sed -n 1,60p references/设计系统落地词汇表.md && echo "=====" && cat skill-card.md _meta.json $ cd ./skill && sed -n 60,80p references/设计系统落地词汇表.md; echo "====="; sed -n 60,80p references/产品拆解案例.md; echo "====="; ls -la .. $ cd .. && cat README.md package.json .gitignore && ls -la src && head -50 src/* 2>/dev/null | head -120 $ cd /; pwd; ls -la; find / -maxdepth 3 -name package.json -not -path "*/node_modules/*" 2>/dev/null | head $ cd /home 2>/dev/null; ls -la /home /mnt /opt /srv /tmp 2>/dev/null | head -40; find / -maxdepth 4 -name "skill" -type d -not -path "/proc/*" 2>/dev/null $ ls -la /opt/detonationlabs-stage; find /opt/detonationlabs-stage -maxdepth 3 -name "skill" -o -maxdepth 3 -name "SKILL.md" 2>/dev/null | head model said: I'll start by reading the skill's references and template so the design follows its method.

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.