Caveat verdict

anti-injection-skill

45
🟠 Risky
Significant risk patterns flagged — automated deep scan, not behavioral proof.

Accesses credentials AND encodes data

Claims to be a security defense layer but declares execution_priority=highest, intercepts user_input/tool_output/memory_load/context_load, modifies_context=true, and can_block_execution=true, giving broad capability to control agent behavior far beyond typical defensive tooling; the optional webhook creates an outbound path for intercepted data.

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.

10
security
50
transparency
70
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

Accesses agent memory/configuration files

agent_memory

Findings (3)

Pattern match critical

Prompt injection — tries to override agent instructions

SKILL.md · frontmatter · ignore previous instructions

Pattern match medium

References webhook/callback URL

SKILL.md · frontmatter · WEBHOOK_URL

Pattern match medium

References agent memory files

SKILL.md · frontmatter · MEMORY.md

Why the tier is capped

Execution sink present in raw bytes (Hard Floor: class A). 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_indata_encodingagent_memorycredential_access
Check another skill Browse the registry Auditing your own skills or configs? Use the API