Back to skill

Security audit

Clawbrain

Security checks for vulnerabilities and agentic risk

Overview

The skill is broadly aimed at peer skill ratings, but it over-collects and transmits installed-skill metadata and its shared-key write design can be manipulated.

Install only if you are comfortable sharing installed skill slugs with the configured ClawBrain service. Do not enable writes unless you trust the endpoint and key handling, and avoid putting secrets, personal data, internal policy, or workspace details in agent_id or notes. Treat peer scores from this service as advisory because the current write model is easy for a key holder to influence.

Vulnerability Patterns
  • Unauthorized Access and Privilege EscalationObtains permissions beyond the task's legitimate needs
  • 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
Findings (4)

T05 · Unauthorized Access and Privilege Escalation

Warning
Location
SKILL.md:29
Finding

Installed Skill Inventory and Agent Metadata Are Transmitted to a Configurable External Endpoint

Content
View full analysis
/dev/null | awk '{print $1}' ``` Collect all installed skill slugs (just the slug part, e.g. `deflate`, not `author/deflate`). --- ## Step 2 — Query Peer Signals Fetch signals for all installed slugs in one request: ```bash SLUGS=$(openclaw skills list 2>/dev/null | awk '{print $1}' | tr '\n' ',' | sed 's/,$//') curl -s "${CLAWBRAIN_API_URL}/signals?slugs=${SLUGS}" | jq '.' ``` ``` ```markdown ## Step 3 — Write Verdict Signals For each skill installed longer than 14 days, write a verdict signal: ```bash curl -s -X POST "${CLAWBRAIN_API_URL}/signal" \ -H "Authorization: Bearer ${CLAWBRAIN_API_KEY}" \ -H "Content-Type: application/json" \ -d '{ "slug": "skill-slug", "agent_id": "your-agent-id-or-name", "verdict": "keep", "days_kept": 47, "notes": "Optional: why keep/remove/flag" }' ``` ``` ### Technical Analysis The workflow instructs the agent to inspect `SOUL.md` or `CLAUDE.md`, even though the declared functionality only requires installed skill slugs and an `openclaw skills list` command is already provided. Agent instruction and configuration files may contain substantially more sensitive information than the workflow requires. The resulting installed-skill inventory is sent in a URL query parameter to the origin specified by `CLAWBRAIN_API_URL`. The repository does not pin or validate that origin. Optional writes also transmit a bearer credential, an agent-chosen identifier, installation duration, a verdict, and unrestric ...[truncated 1558 chars]
Remediation
View remediation

T09 · Insecure Skill Coding Practices

Error
Location
api/signal.js:18
Finding

Shared Write Credential and Missing Replay Controls Permit Peer-Verdict Poisoning

Content
View full analysis
s.verdict === "keep").length; const flag_count = signals.filter((s) => s.verdict === "flag").length; const remove_count = signals.filter((s) => s.verdict === "remove").length; ``` ### Technical Analysis All contributors authenticate with the same `CLAWBRAIN_API_KEY`, described by the Skill documentation as a shared community write key. Authentication therefore establishes only that the caller possesses the community secret; it does not est ...[truncated 1949 chars]
Remediation
View remediation

T09 · Insecure Skill Coding Practices

Warning
Location
api/signal.js:26
Finding

Unbounded Signal Fields and Unrestricted Slugs Enable Redis Storage and Compute Exhaustion

Content
View full analysis
s.trim()) .filter(Boolean); if (slugs.length === 0) { return res.status(400).json({ error: "No valid slugs provided" }); } if (slugs.length > 50) { return res.status(400).json({ error: "Maximum 50 slugs per request" }); } const results = {}; await Promise.all( slugs.map(async (slug) => { const key = `signals:${slug}`; const raw = await redis.lrange(key, 0, -1); ``` ```javascript const keptDays = signals .filter((s) => s.verdict === "keep" && s.days_kept != null) .map((s) => Number(s.days_kept)); const avg_days_kept = keptDays.length > 0 ? Math.round(keptDays.reduce((a, b) => a + b, 0) / keptDays.length) : null; ``` ### Technical Analysis The write endpoint verifies only the presence of `slug`, `agent_id`, and `verdict`. It does not enforce field types, maximum lengths, a slug character policy, a total request-body limit, or a finite and realistic range for `days_kept`. Although each individual Redis list is trimmed to 1,000 en ...[truncated 1884 chars]
Remediation
View remediation

T08 · Insecure Dependencies

Note
Location
package.json:7
Finding

Unlocked Semver Dependency Range Makes Deployments Non-Reproducible

Content
View full analysis
Remediation
View remediation
Vulnerability Patterns
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code
  • MCP Least PrivilegeUnderdeclared Capability, Wildcard Permission, Missing Permission Declaration
  • MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
Findings (7)

Tp4

High
Category
MCP Tool Poisoning
Confidence
98% confidence
Finding

The description promises a functional service for immediately reading peer signals for any skill, with optional authenticated write-back of verdicts. The actual code chunk does none of that; it only exposes a basic /api/health endpoint for liveness checking. While a health endpoint can be a supporting detail, this chunk by itself does not implement or reflect the declared primary purpose, so the description does not accurately represent the supplied code.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
96% confidence
Finding

The description's primary promise is that the skill allows agents to immediately read peer signals for any skill without API keys or setup, with writes being optional. However, the supplied code chunk only exposes a POST endpoint for appending verdict signals and explicitly requires bearer-token authentication via CLAWBRAIN_API_KEY. There is no corresponding read endpoint or logic shown. While optional write-back is mentioned in the description, the implemented code represents only that optional write path and contradicts the 'works out of the box' no-key read experience. This is therefore a material description-to-behavior mismatch.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
94% confidence
Finding

The README explicitly documents that aggregated agent signals are readable without authentication, which can expose operational metadata such as which skills agents evaluate, retention patterns, and flagging behavior. Even though the endpoint returns aggregates rather than raw records, this still creates an information disclosure risk and can enable profiling, enumeration, or competitive intelligence at scale when no access controls or privacy guidance are present.

Content

No source excerpt is available for this finding.

Undeclared Tool Scope

Medium
Category
MCP Least Privilege
Confidence
74% confidence
Finding

The skill instructs access to environment variables and external network interaction but does not declare an explicit tool scope or permissions boundary. This can cause the host agent to invoke capabilities the user did not clearly authorize, increasing the chance of unintended data access or transmission.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
91% confidence
Finding

The skill directs the agent to submit verdicts, notes, days-kept, and an agent identifier to an external service without a prominent, explicit warning at the action point that this is outbound data sharing. Even if the fields seem low sensitivity, notes and agent identifiers can leak internal operational details or become correlatable telemetry across environments.

Content

No source excerpt is available for this finding.

External Transmission

Medium
Category
Data Exfiltration
Confidence
94% confidence
Finding

The skill includes a POST request that transmits structured data and an authorization bearer token to an external endpoint defined by an environment variable. This is dangerous because a misconfigured or malicious endpoint could collect agent metadata, free-form notes, and shared credentials, and the skill does not constrain destination trust or require verification beyond environment configuration.

Content

Scanner excerpt · SKILL.md (reported line 89)May include surrounding context.

For each skill installed longer than 14 days, write a verdict signal:

bash
curl -s -X POST "${CLAWBRAIN_API_URL}/signal" \
  -H "Authorization: Bearer ${CLAWBRAIN_API_KEY}" \
  -H "Content-Type: application/json" \
  -d '{

Unpinned Dependencies

Low
Category
Supply Chain
Confidence
40% confidence
Finding

Dependencies lack version pinning, allowing potential malicious package updates. Consider pinning versions.

Content

Scanner excerpt · package.json (reported line 8)May include surrounding context.

json
"description": "Peer verdict network for ClawHub skills — powered by Upstash Redis on Vercel",
  "type": "module",
  "dependencies": {
    "@upstash/redis": "^1.34.3"
  }
}