Caveat verdict

self-improving-agent

self-improving-agent-pro-v2

🟢 Trusted

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

Framework documentation and install guide for a self-improving agent pattern with no network exfiltration or malicious commands.

⚠ Flagged for review — coarse, uncorroborated signal, not a confirmed exploit. Review the config yourself before installing.

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.

security

100

transparency

100

maintenance

Findings (10)

Pattern match critical

Unicode homoglyph detected — uses lookalike characters to evade pattern matching

scripts/hourly-theory-upgrade-v2.js · prose

Pattern match high

Pipe to python — executes piped content as Python code

references/feature-comparison-20260507.md · prose · downgraded · | Python

Pattern match high

Pipe to sh — executes piped content as shell commands

scripts/knowledge-action-check.sh · prose · downgraded · |sh

Pattern match medium

References child_process — can spawn system processes

scripts/benchmark-upgrades.js · prose · downgraded · child_process

Pattern match medium

Uses exec() — may execute shell commands

scripts/benchmark-upgrades.js · prose · downgraded · exec(

Pattern match medium

subprocess execution — runs system commands from Python

scripts/self_verify.py · prose · downgraded · subprocess.run(

Pattern match medium

subprocess with shell=True — command injection vector

scripts/self_verify.py · prose · downgraded · subprocess.run( cmd, shell=True

Pattern match low

Node http/https module — low-level network access

bin/setup.js · prose · downgraded · require('http')

Pattern match low

Popular HTTP library — network access

scripts/test_logic_upgrade_v11.19.4.js · prose · downgraded · got

Pattern match low

References agent configuration files

src/core/role-based-crew.js · prose · downgraded · agentConfig

Why the tier is capped

Execution sink present in raw bytes (Hard Floor: class A/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.

Is this flag fair?

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