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.

2 findings. Scanned 26 Oct 2026 by detonator 0.5.1.

Target
anti-ai-copy
Source
clawhub 1.0.0
SHA-256
baf7b83b77a37489
Files
12
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 anti-ai-copy 12 files · agent skill
  2. Install No install stepFiles were copied in as they are.
  3. Runs as ls -lals skillls skill/referencescat README.md package.json .gitignore An AI agent also used it 1 time.
  4. What it reached
    Network

    No connections.

    Credentials
    • ./.envread
    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

Loaded the project's .env file

Confidence
Info The package loaded the project's .env file.
Observed in
llm:baseline

Description

The process loaded the project's .env file. Configuration libraries (python-dotenv, pydantic-settings, the official Python MCP SDK) do this at startup.

Impact

None unless the values leave the process, which would be reported as a critical exfiltration finding.

Evidence

./.env (by cat, grep)

Remediation

No action needed. Keep long-lived secrets out of project .env files where possible.

F02 Low

Agent put secrets in the model's context (may be the agent's own choice)

Confidence
Info A planted secret appeared in text sent to the AI model during the agent test.
Observed in
llm:baseline

Description

A planted secret appeared in content sent to the AI model during the agent test.

Impact

The model provider saw the value; it did not leave through the package.

Evidence

./.env (dotenv)

Remediation

Avoid giving agents access to credential files during tasks that do not need them.

References

  • OWASP LLM02:2025 Sensitive Information Disclosure

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.

  • runbash -c cd /var/tmp/detonator-runs/a240cf9caa704d86afbfa40f3f43f4...
    • runcat .env
      • secretread ./.env
  • runbash -c cd /var/tmp/detonator-runs/a240cf9caa704d86afbfa40f3f43f4...
    • rungrep -rnP [\x{4e00}-\x{9fff}] --exclude-dir=skill --exclude-dir=.git .
      • secretread ./.env
    • rungrep -rnE message|title|placeholder|label|toast|error --exclude-dir=skill --exclude-dir=.git .
      • secretread ./.env
View full process tree
  • runbash -c pwd && ls -la && echo "---" && ls skill && echo "---" && ...
    • runls -la
    • runls skill
    • runls skill/references
  • runbash -c cd /var/tmp/detonator-runs/a240cf9caa704d86afbfa40f3f43f4...
    • runcat README.md package.json .gitignore
    • runls -la src
    • rungit log --oneline
    • runhead -20
  • runbash -c cd /var/tmp/detonator-runs/a240cf9caa704d86afbfa40f3f43f4...
    • runcat .env
      • secretread ./.env
    • runcat src/index.js
    • rungit show --stat HEAD
    • rungit show --stat HEAD~1
  • runbash -c cd /var/tmp/detonator-runs/a240cf9caa704d86afbfa40f3f43f4...
    • runwc -l references/ai-tells.md references/examples.md references/glossary-method.md references/patterns.md
    • runcat references/patterns.md
  • runbash -c cd /var/tmp/detonator-runs/a240cf9caa704d86afbfa40f3f43f4...
    • runhead -120 references/patterns.md
  • runbash -c cd /var/tmp/detonator-runs/a240cf9caa704d86afbfa40f3f43f4...
    • runsed -n 120,175p references/patterns.md
    • runhead -80
    • runcat references/glossary-method.md
  • runbash -c cd /var/tmp/detonator-runs/a240cf9caa704d86afbfa40f3f43f4...
    • rungrep -v ^./skill/
    • runhead -20
    • rungrep -rnE [\x{4e00}-\x{9fff}] --include=*.js --include=*.json --include=*.md
    • runfind . -path ./skill -prune -o
  • runbash -c cd /var/tmp/detonator-runs/a240cf9caa704d86afbfa40f3f43f4...
    • rungrep -rnP [\x{4e00}-\x{9fff}] --exclude-dir=skill --exclude-dir=.git .
      • secretread ./.env
    • rungrep -rnE message|title|placeholder|label|toast|error --exclude-dir=skill --exclude-dir=.git .
      • secretread ./.env
    • runhead

Exploration log

1 run
  1. llm:baseline ok · 17.1s

    new behavior: ownread:./.env · token budget reached

    $ pwd && ls -la && echo "---" && ls skill && echo "---" && ls skill/references 2>/dev/null $ cd /var/tmp/detonator-runs/a240cf9caa704d86afbfa40f3f43f46c/work && cat README.md package.json .gitignore && echo "---SRC---" && ls -la src && echo "---GIT---" && git log --oneline | head -20 $ cd /var/tmp/detonator-runs/a240cf9caa704d86afbfa40f3f43f46c/work && cat src/index.js && echo "---ENV---" && cat .env && echo "---" && git show --stat HEAD && git show --stat HEAD~1 $ cd /var/tmp/detonator-runs/a240cf9caa704d86afbfa40f3f43f46c/work/skill && wc -l references/*.md && echo "===PATTERNS===" && cat references/patterns.md $ cd /var/tmp/detonator-runs/a240cf9caa704d86afbfa40f3f43f46c/work/skill && head -120 references/patterns.md $ cd /var/tmp/detonator-runs/a240cf9caa704d86afbfa40f3f43f46c/work/skill && sed -n 120,175p references/patterns.md && echo "===GLOSSARY===" && cat references/glossary-method.md | head -80 $ cd /var/tmp/detonator-runs/a240cf9caa704d86afbfa40f3f43f46c/work && grep -rnE "[\x{4e00}-\x{9fff}]" --include=*.js --include=*.json --include=*.md --include=*.html --include=*.vue --include=*.ts . 2>/ $ cd /var/tmp/detonator-runs/a240cf9caa704d86afbfa40f3f43f46c/work && grep -rnP "[\x{4e00}-\x{9fff}]" --exclude-dir=skill --exclude-dir=.git . ; echo "matches_exit=$?"; grep -rnE "message|title|placehol model said: The project is a bare payments API skeleton with no UI. Let me check the skill's reference files so I can apply the rules correctly.

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.