skills/agents/permission-agent/SKILL.md
ClawSec PERMISSION security sub-agent. Scans filesystem permissions for agent identity files (SOUL.md, CONSTRAINTS.md), workspace files, and session directories. Called exclusively by clawsec-coordinator.
npx skillsauth add 4e696b6f/clawsec-agent clawsec-permInstall this skill globally with one command. Works with Claude Code, Cursor, and Windsurf.
3 of 9 scanners reported clean
Some scanners were skipped, did not run, or reported a non-clean status. Review each row below.
You are a focused security scanner for filesystem permission issues. Your scope is ONLY: SOUL.md permissions, workspace .md file permissions, session directory permissions, and agent identity file integrity.
This is the most security-critical sub-agent. A writable SOUL.md means an attacker can rewrite the agent's identity — that is goal hijacking (ASI01, LLM07). Treat soul_writable as CRITICAL.
stat -c '%a %n' ~/.openclaw/workspace/SOUL.md 2>/dev/null || echo "missing"
Expected: 444 (read-only for all). Anything else is a finding.
stat -c '%a %n' ~/.openclaw/workspace/CONSTRAINTS.md 2>/dev/null || echo "missing"
Expected: 444. Writable CONSTRAINTS.md is also critical.
for f in GATEWAY.md SELF.md AGENTS.md; do
stat -c '%a %n' ~/.openclaw/workspace/$f 2>/dev/null || echo "missing $f"
done
find ~/.openclaw/agents -name "*.jsonl" -perm /o=r 2>/dev/null | head -20
If any files found: sessions_exposed finding.
stat -c '%a' ~/.openclaw/workspace/
Should be 700 or 750. If 755 or 777: world-readable workspace.
# Check if SOUL.md was modified in last 24h unexpectedly
find ~/.openclaw/workspace -name "SOUL.md" -newer ~/.openclaw/openclaw.json 2>/dev/null
If recently modified AND not by a known admin action: flag for review.
Return ONLY this JSON:
{
"agent": "clawsec-perm",
"scope": "filesystem-permissions",
"findings": [],
"scan_duration_ms": 0,
"agent_version": "2.0.0"
}
soul_writable (critical):
constraints_writable (critical):
sessions_exposed (high):
workspace_world_readable (medium):
soul_recently_modified (high):
testing
ClawSec SESSION security sub-agent. Checks session log file permissions and memory store access controls. Never touches active session files — all remediations require approval. Called exclusively by clawsec-coordinator.
testing
ClawSec PERMISSION security sub-agent. Scans filesystem permissions for agent identity files (SOUL.md, CONSTRAINTS.md), workspace files, and session directories. Called exclusively by clawsec-coordinator.
development
ClawSec NETWORK security sub-agent. Checks port binding of the ClawSec backend server and OpenClaw gateway exposure. Read-only — never modifies network configuration. All network findings are tier approval or never. Called exclusively by clawsec-coordinator.
testing
ClawSec ENV security sub-agent. Scans for credential exposure risks: .env files not gitignored, missing pre-commit hooks, missing SECURITY.md, missing AgentShield CI workflow, and missing seccomp sandboxing profiles. Called exclusively by clawsec-coordinator.