Back to skill

Security audit

No.0-skill

Security checks for vulnerabilities and agentic risk

Overview

The skill tries to protect agent identity and files, but its approval controls are weak and it includes risky testing code that can alter protected files.

Install only in a test or isolated environment until the approval service, MFA flow, interceptor fail-open behavior, backup handling, dependency pinning, and tamper simulator containment are fixed. Do not run the tamper simulator against real cognitive files, do not test with real private keys, and back up ~/.openclaw/no0 before using any cleanup or uninstall commands.

Vulnerability Patterns
  • Insecure DependenciesIntroduces malicious components through unsafe dependency sources
  • Insecure Skill Coding PracticesFinds exploitable flaws such as hardcoded secrets or command injection
  • Skill Instruction HijackingAlters the agent's session goals or safety constraints when the skill loads
  • Agent Memory PoisoningWrites attacker-controlled rules into memory that affect later sessions
  • Remote Payload Retrieval and ExecutionFetches external code whose behavior can change after review
Findings (4)

T09 · Insecure Skill Coding Practices

Error
Location
no0-dlc-internal-control/internal_control/http_auth/server.py:32
Finding

Unauthenticated Local HTTP Authorization Can Be Forged by Malicious Websites

Content
View full analysis
Remediation
View remediation

T09 · Insecure Skill Coding Practices

Error
Location
no0-dlc-internal-control/internal_control/rules/dynamic_full_integration.py:66
Finding

TOTP Flow Does Not Provide an Independent Authentication Factor and Uses a Hardcoded Default Password

Content
View full analysis
None: """设置授权确认回调""" def http_callback(auth_url: str, request_id: str) -> bool: print(f"\n 🔐 HTTP 授权请求") print(f" 页面: {auth_url}") if self._auto_approve: import time time.sleep(3) self.http_service.confirm_auth(request_id, "demo_user") return True import webbrowser webbrowser.open(auth_url) return True def mfa_callback(request_info: dict) -> bool: print(f"\n 🔑 TOTP MFA 请求") print(f" 上下文: {request_info['context']}") if self._auto_approve: return True return True self.auth_engine.set_http_confirmation_callback(http_callback) self.auth_engine.set_mfa_confirmation_callback(mfa_callback) ``` ```python def execute_mfa_flow( self, key_id: str, human_confirmation_callback: Callable[[dict], bool]) -> MFAResult: try: cu ...[truncated 3516 chars]
Remediation
View remediation

T09 · Insecure Skill Coding Practices

Error
Location
no0-dlc-internal-control/internal_control/interception/tool_interceptor.py:47
Finding

Reference-Monitor Tool Interception Fails Open and Can Be Bypassed

Content
View full analysis
Remediation
View remediation

T08 · Insecure Dependencies

Warning
Location
install-dlc.sh:57
Finding

Installer Resolves and Executes Unpinned Third-Party Dependencies Without Integrity Verification

Content
View full analysis
/tmp/no0-dlc-pip.log; then echo "! pip install failed (see /tmp/no0-dlc-pip.log). Common fixes:" echo " python3 -m pip install --user --break-system-packages -r \"$TARGET/$DLC_NAME/requirements.txt\"" echo " # or use a venv / pipx" echo " Continuing — classify commands will error until deps are present." fi ``` ```text PyYAML>=6.0 cryptography>=41.0 keyring>=24.0 ``` ### Technical Analysis The installer invokes pip during installation and accepts any current or future release satisfying the lower-bound constraints. It does not pin exact versions, require artifact hashes, use a lockfile, constrain the package index, or create an isolated environment. Python packages can execute build backend or installation code during installation. Consequently, the effective code executed by the installer can change after the skill itself has been reviewed. The use of `--user` limits installation to the current account but still grants access to that user's files and Python environment. The suggested `--break-system-packages` workaround can further modify an externally managed Python environment. ### Attack Path 1. The user runs `install-dlc.sh`. 2. Pip resolves package versions from the configured package index. 3. A compromised package release, index, mirror, or dependency is selected because any version above the lower bound is accepted. 4. Pip downloads and installs the package. 5. Malicious build or package code executes with the user's privileges. 6. The dependency can read user data, modify the Python environment, or persist thro ...[truncated 337 chars]
Remediation
View remediation
Vulnerability Patterns
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
  • Tool MisuseTool Parameter Abuse, Chaining Abuse, Unsafe Defaults
  • MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
Findings (252)

Tool Parameter Abuse

High
Category
Tool Misuse
Confidence
91% confidence
Finding

This finding duplicates the same uninstall instruction to recursively remove ~/.openclaw/no0/, which carries significant risk of destructive local state deletion. In context, the danger is increased because the directory contains backups and audit-related artifacts, so loss may impair recovery and investigation.

Content

Scanner excerpt · INSTALL.md (reported line 35)May include surrounding context.

md
Expected: `monitor` and `timer` running; six cognitive files marked consistent.

Uninstall = stop, then `rm -rf <target_dir>` and optionally `rm -rf ~/.openclaw/no0/`.

---

Tool Parameter Abuse

High
Category
Tool Misuse
Confidence
91% confidence
Finding

This finding duplicates the same uninstall instruction to recursively remove ~/.openclaw/no0/, which carries significant risk of destructive local state deletion. In context, the danger is increased because the directory contains backups and audit-related artifacts, so loss may impair recovery and investigation.

Content

Scanner excerpt · INSTALL.md (reported line 35)May include surrounding context.

md
Expected: `monitor` and `timer` running; six cognitive files marked consistent.

Uninstall = stop, then `rm -rf <target_dir>` and optionally `rm -rf ~/.openclaw/no0/`.

---

Tool Parameter Abuse

High
Category
Tool Misuse
Confidence
91% confidence
Finding

This finding duplicates the same uninstall instruction to recursively remove ~/.openclaw/no0/, which carries significant risk of destructive local state deletion. In context, the danger is increased because the directory contains backups and audit-related artifacts, so loss may impair recovery and investigation.

Content

Scanner excerpt · INSTALL.md (reported line 35)May include surrounding context.

md
Expected: `monitor` and `timer` running; six cognitive files marked consistent.

Uninstall = stop, then `rm -rf <target_dir>` and optionally `rm -rf ~/.openclaw/no0/`.

---

Credential Access

High
Category
Privilege Escalation
Confidence
90% confidence
Finding

The verification example tells the user to run a classification command directly against ~/.ssh/id_rsa, which is a private SSH key path. Even if the stated purpose is only classification, normalizing commands that touch highly sensitive credential material increases the chance of secret exposure through logs, telemetry, bugs in the tool, or unsafe future extensions.

Content

Scanner excerpt · INSTALL.md (reported line 76)May include surrounding context.

bash
./no0 help                                  # unified help
./no0 status                                # Core status
./no0 classify get ~/.ssh/id_rsa            # DLC classification (expects L6-CRITICAL)
./no0 audit log                             # DLC audit (empty on fresh install)

Tool Parameter Abuse

High
Category
Tool Misuse
Confidence
86% confidence
Finding

The troubleshooting advice recommends deleting all files under no0-core/cognitive_file_backups/* with rm -rf, which destroys backup history and could remove recovery data needed after corruption or tampering. Even though it is scoped more narrowly than the uninstall command, it still promotes unsafe destructive remediation without warning or validation.

Content

Scanner excerpt · INSTALL.md (reported line 140)May include surrounding context.

md
| `./no0 classify …` says missing `yaml` | `pip install --user -r no0-dlc-internal-control/requirements.txt` |
| `monitor not running` after `./no0 start` | Check `no0-core/cognitive_file_monitor.log`; try `./no0 stop && ./no0 start`. |
| DLC handler not picking up events | Confirm the handler process is running (`ps aux \| grep cognitive_event_handler`). |
| "总是报告不一致" on Core | Stop monitor, `rm -rf no0-core/cognitive_file_backups/*`, restart. |

---

Credential Access

High
Category
Privilege Escalation
Confidence
97% confidence
Finding

Using ~/.ssh/id_rsa in a quickstart test directly targets one of the most sensitive credential files on a system. Even if the command only 'classifies' metadata, normalizing interaction with private keys in documentation creates unnecessary credential-access behavior and raises the risk of accidental disclosure, logging, transmission, or overbroad future feature use.

Content

Scanner excerpt · QUICKSTART.md (reported line 51)May include surrounding context.

30-second smoke test (DLC)

bash
./no0 classify get ~/.ssh/id_rsa       # L6-CRITICAL
./no0 classify get ~/Desktop           # classification or "excluded"
./no0 audit log                        # empty on a fresh install

Tool Parameter Abuse

High
Category
Tool Misuse
Confidence
90% confidence
Finding

The troubleshooting guidance includes a forceful recursive deletion command against a backup directory, which can destroy data if the path is wrong, expanded unexpectedly, or copied carelessly. Destructive shell commands in quickstarts are especially dangerous because users often paste them verbatim under troubleshooting pressure.

Content

Scanner excerpt · QUICKSTART.md (reported line 100)May include surrounding context.

md
- **`./no0 <dlc cmd>` says "not installed"**: run `./install-dlc.sh`.
- **`./no0 classify …` says "missing yaml"**: `pip install -r no0-dlc-internal-control/requirements.txt`.
- **Monitor not running**: `ps aux | grep skill_launcher`; check `no0-core/cognitive_file_monitor.log`.
- **"总是报告不一致"**: stop monitor, `rm -rf no0-core/cognitive_file_backups/*`, start again.

See [README.md](README.md) for full narrative and [docs/event_schema.md](docs/event_schema.md) for the event protocol.

Credential Access

High
Category
Privilege Escalation
Confidence
98% confidence
Finding

The verification example explicitly tells the user to run the tool against ~/.ssh/id_rsa, a highly sensitive private key file. Even though presented as a demo, encouraging processing of real credentials materially increases the chance that secrets are ingested, logged, cached, or otherwise exposed by the installed skill or its downstream components.

Content

Scanner excerpt · install-dlc.sh (reported line 80)May include surrounding context.

sh
echo
echo "Verify:"
echo "  cd \"$TARGET\""
echo "  ./no0 classify get ~/.ssh/id_rsa   # any path"
echo "  ./no0 audit log"
echo
echo "Run the event handler in the background (recommended when Core is installed):"

Tool Parameter Abuse

High
Category
Tool Misuse
Confidence
90% confidence
Finding

Tool parameters are crafted to achieve unintended or unsafe behavior. Parameter abuse can bypass intended safety checks (e.g. shell=True, --force, dangerous glob patterns).

Content

Scanner excerpt · no0-core/scripts/heartbeat_processor.py (reported line 43)May include surrounding context.

python
"monitor blind", "权限自动放行", "allow all", "trust all",
    ],
    "自动执行外部命令": [
        "自动执行外部命令", "os.system(", "subprocess.run(",
        "shell=true", "invoke-expression", "eval(", "exec(",
        "curl ", "wget ", "powershell -command",
    ],

Intent-Code Divergence

High
Category
Not specified by scanner
Confidence
97% confidence
Finding

The code detects a file hash mismatch and then calls init_readonly_backup(src, force_refresh=True), which rotates the previous clean backup to v2 and copies the current already-modified source file into the new v1. This contradicts the comment and can poison the backup chain, causing operators to trust a backup set that now includes attacker-modified content as the latest baseline and making rollback/confidence in forensic recovery unreliable.

Content

No source excerpt is available for this finding.

Context-Inappropriate Capability

High
Category
Not specified by scanner
Confidence
99% confidence
Finding

This section defines multiple mutation algorithms whose explicit purpose is to weaken trust boundaries, relax permissions, bypass monitoring, obstruct rollback, and add obfuscated policy text. Even if labeled as a simulator, the code directly generates adversarial content for governance files and includes stealth/evasion behaviors that are unsafe in an agent skill with no strong containment or authorization boundaries.

Content

No source excerpt is available for this finding.

Ssd 1

High
Category
Not specified by scanner
Confidence
99% confidence
Finding

This mutation changes restrictive terms like 'forbidden' or 'must' to permissive language and adds fallback text allowing high-risk actions without authorization. It directly weakens approval controls and can cause execution of dangerous operations based on external input or without human consent.

Content

No source excerpt is available for this finding.

Ssd 1

High
Category
Not specified by scanner
Confidence
98% confidence
Finding

This function injects instructions to prioritize external scripts over local guardrails and to ignore existing safety tool restrictions during conflicts. That explicitly subverts defense-in-depth and can reroute execution around trusted safety controls.

Content

No source excerpt is available for this finding.

Ssd 4

High
Category
Not specified by scanner
Confidence
97% confidence
Finding

The multi-vector mutation combines external trust elevation, permission auto-approval, and delayed monitoring into a single coordinated tamper scenario. Layering these changes increases the chance that harmful behavior persists undetected and normalizes insecure operating rules across control files.

Content

No source excerpt is available for this finding.

Ssd 2

High
Category
Not specified by scanner
Confidence
99% confidence
Finding

This routine deliberately encodes an unsafe policy instruction to allow automatic execution of external commands without confirmation, then lightly obfuscates it with zero-width characters to evade casual inspection. That combination of unsafe semantics and concealment strongly indicates intentional stealthy policy subversion.

Content

No source excerpt is available for this finding.

Context-Inappropriate Capability

High
Category
Not specified by scanner
Confidence
98% confidence
Finding

The main loop repeatedly selects targets and rewrites governance-related files on disk, including indefinite operation and periodic execution. This turns the script into an active persistence mechanism for ongoing tampering rather than a passive test utility, especially because the default targets are core policy files and writes occur automatically once invoked.

Content

No source excerpt is available for this finding.

Credential Access

High
Category
Privilege Escalation
Confidence
90% confidence
Finding

Code accesses credential files (SSH keys, AWS credentials, etc.). This could indicate credential theft attempts.

Content

Scanner excerpt · no0-dlc-internal-control/internal_control/classify_cli.py (reported line 13)May include surrounding context.

python
openclaw-classify reload               - 重新加载规则

示例:
    openclaw-classify get ~/.ssh/id_rsa
    openclaw-classify dir ~/Documents --recursive
"""

Credential Access

High
Category
Privilege Escalation
Confidence
90% confidence
Finding

Code accesses credential files (SSH keys, AWS credentials, etc.). This could indicate credential theft attempts.

Content

Scanner excerpt · no0-dlc-internal-control/internal_control/reference_monitor.py (reported line 518)May include surrounding context.

python
openclaw-classify reload               - 重新加载规则

示例:
    openclaw-classify get ~/.ssh/id_rsa
    openclaw-classify dir ~/Documents --recursive
"""

Credential Access

High
Category
Privilege Escalation
Confidence
90% confidence
Finding

Code accesses credential files (SSH keys, AWS credentials, etc.). This could indicate credential theft attempts.

Content

Scanner excerpt · no0-dlc-internal-control/internal_control/rules/__init__.py (reported line 17)May include surrounding context.

python
openclaw-classify reload               - 重新加载规则

示例:
    openclaw-classify get ~/.ssh/id_rsa
    openclaw-classify dir ~/Documents --recursive
"""

Credential Access

High
Category
Privilege Escalation
Confidence
90% confidence
Finding

Code accesses credential files (SSH keys, AWS credentials, etc.). This could indicate credential theft attempts.

Content

Scanner excerpt · no0-dlc-internal-control/tests/rules/test_engine.py (reported line 138)May include surrounding context.

python
openclaw-classify reload               - 重新加载规则

示例:
    openclaw-classify get ~/.ssh/id_rsa
    openclaw-classify dir ~/Documents --recursive
"""

Credential Access

High
Category
Privilege Escalation
Confidence
90% confidence
Finding

Code accesses credential files (SSH keys, AWS credentials, etc.). This could indicate credential theft attempts.

Content

Scanner excerpt · no0-dlc-internal-control/tests/rules/test_engine.py (reported line 144)May include surrounding context.

python
openclaw-classify reload               - 重新加载规则

示例:
    openclaw-classify get ~/.ssh/id_rsa
    openclaw-classify dir ~/Documents --recursive
"""

Credential Access

High
Category
Privilege Escalation
Confidence
90% confidence
Finding

Code accesses credential files (SSH keys, AWS credentials, etc.). This could indicate credential theft attempts.

Content

Scanner excerpt · no0-dlc-internal-control/tests/rules/test_engine.py (reported line 171)May include surrounding context.

python
openclaw-classify reload               - 重新加载规则

示例:
    openclaw-classify get ~/.ssh/id_rsa
    openclaw-classify dir ~/Documents --recursive
"""

Credential Access

High
Category
Privilege Escalation
Confidence
90% confidence
Finding

Code accesses credential files (SSH keys, AWS credentials, etc.). This could indicate credential theft attempts.

Content

Scanner excerpt · no0-dlc-internal-control/tests/rules/test_engine.py (reported line 178)May include surrounding context.

python
openclaw-classify reload               - 重新加载规则

示例:
    openclaw-classify get ~/.ssh/id_rsa
    openclaw-classify dir ~/Documents --recursive
"""

Credential Access

High
Category
Privilege Escalation
Confidence
90% confidence
Finding

Code accesses credential files (SSH keys, AWS credentials, etc.). This could indicate credential theft attempts.

Content

Scanner excerpt · no0-dlc-internal-control/tests/rules/test_engine.py (reported line 250)May include surrounding context.

python
openclaw-classify reload               - 重新加载规则

示例:
    openclaw-classify get ~/.ssh/id_rsa
    openclaw-classify dir ~/Documents --recursive
"""

Credential Access

High
Category
Privilege Escalation
Confidence
90% confidence
Finding

Code accesses credential files (SSH keys, AWS credentials, etc.). This could indicate credential theft attempts.

Content

Scanner excerpt · no0-dlc-internal-control/tests/test_reference_monitor.py (reported line 113)May include surrounding context.

python
openclaw-classify reload               - 重新加载规则

示例:
    openclaw-classify get ~/.ssh/id_rsa
    openclaw-classify dir ~/Documents --recursive
"""