ClawAudit verdict
openclaw-soul
openclaw-soul-publish
OpenClaw self-evolution framework installer deploying workspace config files from templates; all operations are local file writes with explicit user confirmation gates at each step.
⚠ Flagged for review — coarse, uncorroborated signal, not a confirmed exploit. Review the config yourself before installing.
Automated static analysis — not a human review. ClawAudit flags capabilities, not confirmed intent, and can produce false positives. Disagree with this verdict? Use Dispute below.
Findings (10)
Possible hardcoded credential
fallback/evoclaw/references/sources.md · code · API_KEY="${!api_key_env}
Pipe to python — executes piped content as Python code
fallback/evoclaw/references/heartbeat-debug.md · code · | python3
Instructs covert action — may act without user awareness
fallback/evoclaw/configure.md · code · silently
Accesses OpenClaw config/secrets directly
fallback/evoclaw/references/heartbeat-debug.md · code · ~/.openclaw/openclaw.json
<script> tag in markdown — potential code injection
fallback/evoclaw/tools/soul-viz.py · prose · downgraded · <script>
References agent memory files
fallback/evoclaw/configure.md · code · MEMORY.md
subprocess execution — runs system commands from Python
fallback/evoclaw/validators/run_all.py · prose · downgraded · subprocess.run(
Popular HTTP library — network access
fallback/self-improving/learning.md · code · Got
Possible prompt injection — attempts to redefine agent identity
references/bootstrap-guide.md · prose · downgraded · you are now
Python os.environ.get — reads environment variable
scripts/preflight_check.py · prose · downgraded · os.environ.get(
Why the tier is capped
Execution sink present in raw bytes (Hard Floor: class C/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?
Thanks — recorded.