Back to skill

Security audit

OpenClaw Blackbox

Security checks for vulnerabilities and agentic risk

Overview

This skill is a coherent local debugging tool, but it installs an unpinned third-party CLI that reads sensitive OpenClaw session data and can write detailed reports to disk.

Install only if you trust the npm package publisher and are comfortable giving the CLI read access to OpenClaw session history. Prefer a pinned/package-reviewed install, inspect one explicit session where possible, avoid broad `--agent all` use unless needed, and review generated reports before sharing or opening them elsewhere.

Vulnerability Patterns
  • Unauthorized Access and Privilege EscalationObtains permissions beyond the task's legitimate needs
  • Insecure DependenciesIntroduces malicious components through unsafe dependency sources
  • 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 (2)

T08 · Insecure Dependencies

Warning
Location
SKILL.md:13
Finding

Unpinned Global Installation of an Unreviewed Third-Party Package

Content
View full analysis

Vulnerability Details

File Location: SKILL.md, lines 13–21 and line 31
Vulnerability Type: Third-party supply-chain exposure through a mutable npm dependency
Risk Level: Medium

Vulnerable Code

yaml
"install":
  [
    {
      "id": "npm",
      "kind": "node",
      "package": "@shan8851/blackbox",
      "bins": ["blackbox"],
      "label": "Install OpenClaw Blackbox (npm)",
    },
  ],
markdown
- `npm install -g @shan8851/blackbox`

Technical Analysis

The skill delegates all operational behavior to the external npm package @shan8851/blackbox, but the reviewed project contains neither that package's implementation nor a lockfile, exact version, or integrity hash. Consequently, the effective executable payload cannot be verified from the audited artifact and may change after this review.

The documented npm install -g command installs the package globally. npm installation may execute package lifecycle scripts, while the installed blackbox binary subsequently runs with the invoking user's permissions. An unpinned dependency can resolve to a newer or compromised release without any corresponding change to this skill.

The audit did not establish that the referenced package is malicious. The vulnerability is the unsafe and unverifiable supply-chain trust model.

Attack Path

  1. An attacker compromises the npm package, its publisher account, or a transitive dependency.
  2. The attacker publishes a modified release containing malicious lifecycle or runtime code.
  3. A user follows the skill instructions and runs npm install -g @shan8851/blackbox.
  4. npm resolves the mutable package reference to the compromised release.
  5. Malicious lifecycle code may execute during installation, or malicious code executes when the user invokes blackbox.
  6. The code operates with the installing or invoking user's permissions and may access files available to that account.

...[truncated 548 chars]

Remediation
View remediation

Remediation Suggestions

  1. Pin the package to a specific, reviewed version rather than installing the latest mutable release.
  2. Use a lockfile and verify the package archive against a trusted integrity hash.
  3. Include or link to the exact source revision corresponding to the pinned package release.
  4. Review the package and its transitive dependencies before distribution.
  5. Prefer a project-local, isolated installation instead of a global installation.
  6. Disable npm lifecycle scripts during installation where operationally feasible, and explicitly review any lifecycle scripts that are required.
  7. Execute the CLI in a sandbox with narrowly scoped filesystem access and without unrelated credentials in its environment.
  8. Establish dependency monitoring and a controlled update process so package upgrades require a new security review.

T05 · Unauthorized Access and Privilege Escalation

Warning
Location
SKILL.md:32
Finding

Broad Access to Sensitive Local OpenClaw Session Evidence

Content
View full analysis

Vulnerability Details

File Location: SKILL.md, lines 32–33, 41–61, and 68–75
Vulnerability Type: Broad access to local agent records and potentially sensitive report data
Risk Level: Medium

Vulnerable Code

markdown
- Requires local OpenClaw state; by default it reads `~/.openclaw`
- To inspect another OpenClaw home: `OPENCLAW_HOME=/path/to/.openclaw blackbox doctor`
markdown
- Local sessions across agents: `blackbox list sessions --agent all --limit 20`
- Sessions with specific evidence: `blackbox list sessions --evidence checkpoint_present`

Inspect Runs

- Latest failed run for a cron job: `blackbox inspect --cron-job "Example nightly job" --latest-error`
- Known session: `blackbox inspect --session-id <session-id>`
- Full Markdown report: `blackbox inspect --session-id <session-id> --view full --out reports/run.md`
- JSON report: `blackbox inspect --session-id <session-id> --json-out reports/run.json`
- HTML report: `blackbox inspect --session-id <session-id> --html-out reports/run.html --open`

Find Requests

- Find the session behind a user request: `blackbox find request --query "summarise the failed run" --agent all`
- Find by message id: `blackbox find request --message-id <message-id> --agent main`
- Emit inspect commands only: `blackbox find request --query "failed to send" --commands`
markdown
- Reports can include prompts, tool arguments/results, URLs, and local paths; review before sharing externally
- Blackbox is disk-first and should not make hidden LLM calls by default

Technical Analysis

The skill directs an externally installed CLI to read local OpenClaw state from ~/.openclaw by default and supports selecting another state directory through OPENCLAW_HOME. It also documents cross-agent enumeration using --agent all and the generation of full Markdown, JSON, and HTML reports.

The documentation explicitly acknowledges that generated ...[truncated 2192 chars]

Remediation
View remediation

Remediation Suggestions

  1. Require an explicit OpenClaw home and session identifier for detailed inspection rather than defaulting to broad discovery.
  2. Avoid --agent all in recommended workflows unless cross-agent access is specifically required and confirmed by the user.
  3. Default to metadata-only or redacted output and require an explicit option before including prompts or complete tool results.
  4. Detect and redact common credentials, authorization headers, tokens, personal information, sensitive URLs, and private filesystem paths.
  5. Display a confirmation prompt before generating full reports or exporting records from multiple agents.
  6. Create report files with restrictive permissions and reject unsafe output locations where other users may read the files.
  7. Provide configurable retention controls and a secure deletion workflow for generated reports.
  8. Clearly document the exact files and directories the CLI reads and the permissions it requires.
  9. Run the CLI in a filesystem sandbox that grants read access only to the explicitly selected OpenClaw records and write access only to a dedicated report directory.
  10. Warn users before opening HTML reports in a browser or sharing any generated artifact externally.
Vulnerability Patterns
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
  • Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
Findings (1)

Missing User Warnings

Low
Category
Not specified by scanner
Confidence
79% confidence
Finding

This markdown file documents commands that create Markdown, JSON, and HTML reports on disk, and elsewhere notes that reports can include prompts, tool arguments/results, URLs, and local paths. However, the file does not warn at the point of these write examples that running them will create local artifacts containing potentially sensitive debugging data.

Content

No source excerpt is available for this finding.