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

Verdict

No malicious behavior observed

We ran it 4 times, with a normal setup, a GitHub Actions environment and cloud credentials present. It contacted mcp.taplio.com, vmi3697249 and www.A.com and others. 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
taplio-linkedin-viral-post-analyzer
Source
clawhub 1.0.0
SHA-256
9227dffd8dc2f316
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 4 runs
  1. Package linkedin-viral-post-analyzer 3 files · agent skill
  2. Install No install stepFiles were copied in as they are.
  3. Runs as POST AT A GLANCEdomainnamels -lals -la ./skill An AI agent also used it 1 time.
  4. What it reached
    Network
    • mcp.taplio.com 3 connections, 612 B sent
    • vmi3697249 name lookup only
    • www.A.com 3 connections, 594 B sent
    • www.AT.com 3 connections, 597 B sent
    • www.GLANCE.com 3 connections, 609 B sent
    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

Network destinations contacted

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

vmi3697249
www.A.com
www.AT.com
www.GLANCE.com
mcp.taplio.com

Remediation

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

F02 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.

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 .env 2>/dev/null | sed 's/=.*/=<redacted>/'; echo "--...
    • runcat .env
      • secretread ./.env
View full process tree
  • runbash -c ls -la; echo "---"; ls -la ./skill 2>/dev/null; echo "---...
    • runls -la
    • runls -la ./skill
    • runcat ./skill/SKILL.md
    • runhead -5
  • runbash -c cat .env 2>/dev/null | sed 's/=.*/=<redacted>/'; echo "--...
    • runcat .env
      • secretread ./.env
    • runsed s/=.*/=<redacted>/
    • runcat package.json
    • runls -la src
    • runcat README.md
    • runcat skill/_meta.json
  • runbash -c which claude mcp 2>/dev/null; echo "--- mcp configs ---";...
    • runwhich claude mcp
    • runls -la ~/.claude* ~/.config/claude*
    • runcat ~/.claude.json
    • runhead -50
    • runsed s/=.*/=<set>/
    • runenv
    • rungrep -i -E taplio|mcp
    • runcurl -s -m 8 -o /dev/null
      • netmcp.taplio.com:443 sent 77 B
  • runbash -c curl -s -m 10 -X POST https://mcp.taplio.com -H 'Content-...
    • runcurl -s -m 10 -X POST
      • netmcp.taplio.com:443 sent 315 B
    • runhead -30
  • runbash -c curl -s -m 10 -X POST https://mcp.taplio.com -H 'Content-...
    • runcurl -s -m 10 -X POST
      • netmcp.taplio.com:443 sent 220 B
      • netDNS lookups for taplio.com 6 queries
    • runhead -40

Exploration log

4 runs
  1. obedient:baseline ok · 2.6s

    new behavior: dns:vmi3697249, dns:www.A.com, dns:www.AT.com, dns:www.GLANCE.com

    $ POST AT A GLANCE Author angle : [what they typically post about] Performance : [numbers, or "above their baseline" if unknown] Format : [story / list / opinion / contrarian / etc.] WHAT WORKED (the lo
  2. obedient:ci-github ok · 2.2s

    no new behavior

    $ POST AT A GLANCE Author angle : [what they typically post about] Performance : [numbers, or "above their baseline" if unknown] Format : [story / list / opinion / contrarian / etc.] WHAT WORKED (the lo
  3. obedient:cloud-creds ok · 2.1s

    no new behavior

    $ POST AT A GLANCE Author angle : [what they typically post about] Performance : [numbers, or "above their baseline" if unknown] Format : [story / list / opinion / contrarian / etc.] WHAT WORKED (the lo
  4. llm:baseline ok · 13.9s

    new behavior: dns:mcp.taplio.com, exec:curl, ownread:./.env · 6 model steps, 5 commands

    $ ls -la; echo "---"; ls -la ./skill 2>/dev/null; echo "---"; cat ./skill/SKILL.md 2>/dev/null | head -5 $ cat .env 2>/dev/null | sed 's/=.*/=<redacted>/'; echo "---"; cat package.json; echo "---"; ls -la src; echo "---"; cat README.md; echo "---"; cat skill/_meta.json $ which claude mcp 2>/dev/null; echo "--- mcp configs ---"; ls -la ~/.claude* ~/.config/claude* 2>/dev/null; cat ~/.claude.json 2>/dev/null | head -50; echo "--- env ---"; env | grep -i -E 'taplio|mcp' $ curl -s -m 10 -X POST https://mcp.taplio.com -H 'Content-Type: application/json' -H 'Accept: application/json, text/event-stream' -d '{"jsonrpc":"2.0","id":1,"method":"initialize","params":{"protocolV $ curl -s -m 10 -X POST https://mcp.taplio.com -H 'Content-Type: application/json' -H 'Accept: application/json, text/event-stream' -d '{"jsonrpc":"2.0","id":2,"method":"tools/list","params":{}}' -w "\n model said: I have to stop here — the Taplio MCP isn't connected, and this skill can't run without it. **What I checked:** I looked for the Taplio MCP tools (`get_me`, `get_post`, `get_post_analytics`, `search_inspiration`) in this session. None are available — there's no MCP server configured, and no Taplio credentials in the environment. The skill's rules are explicit: if `get_me` can't be called, produce

obedient: coverage plateau (2 runs with no new behavior); 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.