Caveat verdict
input-guard
Receives external input AND executes processes
A defensive prompt-injection scanner that analyzes untrusted text using local regex patterns with zero external dependencies; optional MoltThreats reporting is clearly opt-in via an environment variable.
Automated static analysis — not a human review. Caveat flags capabilities, not confirmed intent, and can produce false positives. Disagree with this verdict? Use Dispute below.
What it does
These are capability combinations: each listed behavior occurs in the skill, but Caveat detects co-occurrence — it does not verify that one flows into another. Read the code to confirm a live chain.
Receives external input AND executes processes — the shape of a command & control channel
LLM05 · LLM06 · ASI10
Executes processes AND makes external network calls — may exfiltrate command output
LLM02 · LLM06 · ASI03
Permission integrity
network_out
Findings (8)
Pipe to bash — executes piped content as shell commands
SKILL.md · code · | bash
Pipe to python — executes piped content as Python code
SKILL.md · code · | python3
Unicode homoglyph detected — uses lookalike characters to evade pattern matching
evals/cases.json · prose
subprocess execution — runs system commands from Python
SKILL.md · code · subprocess.run(
References sudo — requests elevated privileges
README.md · code · sudo
apt-get install — installs system packages
README.md · code · apt-get install
Python os.environ.get — reads environment variable
scripts/get_taxonomy.py · prose · downgraded · os.environ.get(
POSTs data to external URL
scripts/llm_scanner.py · prose · downgraded · .post(
"https://
Why the tier is capped
Execution sink present in raw bytes (Hard Floor: class A/B/D). Final tier capped at Caution — cannot be lifted by any downgrade, example-payload opt-in, or allowlist.
Permissions & capabilities
No declared permissions — minimal attack surface.
network_outprocess_execnetwork_in Thanks — recorded.