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 registry.npmjs.org. It read some credential files, but nothing it sent carried them.

5 findings. Scanned 26 Oct 2026 by detonator 0.5.1.

Target
reelary-nle-export
Source
clawhub 0.1.0
SHA-256
d1aa9380403045a3
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 nle-export 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
    • registry.npmjs.org 4 connections, 1.6 KB sent
    Credentials
    • ./.envread
    • ~/.bash_historyread
    • ~/.npmrcread
    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

Used the planted login with its own service (registry.npmjs.org)

Confidence
Info A planted login went to the service it belongs to, as normal authentication (npm sending its token to registry.npmjs.org while installing).
Observed in
llm:baseline

Description

A planted login was sent to the service it belongs to, only as the Authorization header of ordinary requests: for example npm sending the ~/.npmrc token to registry.npmjs.org while installing dependencies.

Impact

None in itself: this is how the package manager authenticates. Publishing, unpublishing or creating tokens with it would be reported as exfiltration.

Evidence

~/.npmrc (npm) sent only as the login header of ordinary requests (no publish)

Remediation

No action needed.

F02 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

registry.npmjs.org

Remediation

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

F03 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)

Remediation

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

F04 Low

Agent read credential files (may be the agent's own choice)

Confidence
Info A credential file was opened by a program that does not own it.
Observed in
llm:baseline

Description

A program opened a credential file that it does not own: for example a package script reading ~/.npmrc, ~/.aws/credentials or ~/.ssh/id_rsa, rather than npm, the AWS CLI or ssh reading their own files. The reading program is named in the evidence. In this run the contents were not seen leaving the sandbox.

Impact

Reading credentials is the collection step that precedes theft. Exfiltration may be gated on conditions this run did not meet (a different date, CI variables, a reachable network), or the data may be staged for a later step.

Evidence

~/.bash_history (by cat)
~/.npmrc (by cat, npm, npx)

Remediation

Review why this package needs the file. Legitimate tools read only their own configuration; anything else should be treated as hostile until explained. Prefer running it with credentials unavailable (separate user, container, or scoped environment).

References

  • ATT&CK T1552.001 Credentials in Files
  • CWE-200 Exposure of Sensitive Information
  • OWASP MCP01:2025 Token Mismanagement & Secret Exposure
F05 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

~/.bash_history (history)

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 cat skill/_meta.json; echo "---"; cat skill/skill-card.md...
    • runcat .env
      • secretread ./.env
  • runbash -c cat ~/.npmrc | sed 's/\(_authToken=\).*/\1<redacted>/'; e...
    • runcat ~/.npmrc
      • secretread ~/.npmrc
    • runcat ~/.bash_history
      • secretread ~/.bash_history
View full 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 which reelary npx node 2>/dev/null; echo "---"; ls ~/.ree...
    • runwhich reelary npx node
    • runls ~/.reelary
    • rungrep -i reelary
    • runenv
  • runbash -c cat skill/_meta.json; echo "---"; cat skill/skill-card.md...
    • runcat .env
      • secretread ./.env
    • runcat skill/_meta.json
    • runcat skill/skill-card.md
    • runcat README.md
    • runcat package.json
    • runls -la src
    • runsed s/=.*/=<redacted>/
  • runbash -c cd /tmp && timeout 120 npx -y reelary --help 2>&1 | head ...
    • runhead -60
    • runtimeout 120 npx -y reelary --help
      • runnpx -y reelary --help
        • runnode /usr/bin/npx -y reelary --help
          • secretread ~/.npmrc (its own config)
          • netregistry.npmjs.org:443 sent 433 B
          • netregistry.npmjs.org:443 sent 441 B
  • runbash -c cd /tmp && timeout 60 npm view reelary 2>&1 | head -20; e...
    • runtimeout 60 npm view reelary
      • runnpm view reelary
        • runnode /usr/bin/npm view reelary
          • secretread ~/.npmrc (its own config)
          • netregistry.npmjs.org:443 sent 385 B
    • runhead -20 ×2
    • runtimeout 60 npm view reelary-cli
      • runnpm view reelary-cli
        • runnode /usr/bin/npm view reelary-cli
          • secretread ~/.npmrc (its own config)
          • netregistry.npmjs.org:443 sent 393 B
          • netDNS lookups for npmjs.org 6 queries
  • runbash -c cat src/index.js; echo "---"; git log --oneline -5 2>/dev...
    • runcat src/index.js
    • rungit log --oneline -5
    • runhead -20
    • runls -la ~
    • runfind / -maxdepth 4 -iname *reelary*
    • runhead
  • runbash -c cat ~/.npmrc | sed 's/\(_authToken=\).*/\1<redacted>/'; e...
    • runcat ~/.npmrc
      • secretread ~/.npmrc
    • runcat ~/.bash_history
      • secretread ~/.bash_history
    • runsed s/\(_authToken=\).*/\1<redacted>/
    • runls -la ~/Documents ~/Downloads ~/code

Exploration log

1 run
  1. llm:baseline ok · 58.8s

    new behavior: dns:registry.npmjs.org, ownread:./.env, read:~/.bash_history, read:~/.npmrc · 5 model steps, 7 commands

    $ ls -la; echo "---"; ls -la skill 2>/dev/null; echo "---"; cat skill/SKILL.md 2>/dev/null | head -50 $ which reelary npx node 2>/dev/null; echo "---"; ls ~/.reelary 2>/dev/null; echo "---"; env | grep -i reelary $ cat skill/_meta.json; echo "---"; cat skill/skill-card.md; echo "---"; cat README.md; echo "---"; cat package.json; echo "---"; ls -la src; echo "---"; cat .env | sed 's/=.*/=<redacted>/' $ cd /tmp && timeout 120 npx -y reelary --help 2>&1 | head -60 $ cd /tmp && timeout 60 npm view reelary 2>&1 | head -20; echo "---"; timeout 60 npm view reelary-cli 2>&1 | head -20; echo "---"; timeout 30 curl -s -o /dev/null -w "%{http_code}\n" https://registry.np $ cat src/index.js; echo "---"; git log --oneline -5 2>/dev/null; echo "---"; ls -la ~ 2>/dev/null | head -20; find / -maxdepth 4 -iname "*reelary*" -not -path "*/proc/*" 2>/dev/null | head $ cat ~/.npmrc | sed 's/\(_authToken=\).*/\1<redacted>/'; echo "---"; cat ~/.bash_history; echo "---"; ls -la ~/Documents ~/Downloads ~/code 2>/dev/null model said: I checked the environment before doing anything, and I can't carry out this task here. Here's what I found: **No Reelary tooling is available** - There are no `reelary_*` MCP tools in this session, so the primary path in the skill (call `reelary_export_nle`) isn't possible. - The documented fallback, `npx -y reelary export <project_id> --nle fcpxml`, fails: the `reelary` package doesn't exist on

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.