ClawHub
Back to skill

Security audit

Parameter Golf Monitor

Security checks for vulnerabilities and agentic risk

Overview

This skill coherently fetches public GitHub pull request data to show a Parameter Golf leaderboard, with no evidence of credential access, hidden persistence, destructive behavior, or unrelated data collection.

Install only if you want an agent to query GitHub for the openai/parameter-golf leaderboard. Be aware that generic leaderboard questions may trigger it, --watch keeps polling until stopped, and PR titles are untrusted GitHub content that should be treated as data rather than instructions.

A.I.G

Vulnerability Patterns
  • 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
  • Embedded Malicious CodeShips malicious scripts inside the skill and executes them locally
Findings (1)

T09 · Insecure Skill Coding Practices

Warning
Location
scripts/monitor.py:127
Finding

Untrusted GitHub PR Metadata Is Rendered Without Terminal-Safe Sanitization

Content
View full analysis
4} {score_str:>8} #{e['number']:<4} {status_str:<8} " f"{e['author']:<18} {e['date']} {e['title']}{marker}{suspect_marker}" ) ``` ### Technical Analysis The PR title and author are obtained from the external GitHub API. The title is truncated to 75 characters, but it is not sanitized before being interpolated into terminal output. Length truncation does not neutralize ANSI escape sequences, C0/C1 control characters, carriage returns, bidirectional text controls, or other display-altering characters. Consequently, attacker-controlled PR metadata may alter how the leaderboard appears in a terminal, conceal legitimate output, or display fabricated status and ranking information. The Skill documentation also directs an AI Agent to invoke this script and interpret its results. In that context, instruction-like text from a malicious PR title enters the Agent-visible tool output without an explicit trust-boundary marker. This creates an indirect prompt-injection surface, although the code does not itself execute PR content as shell commands or Python code. ### Attack Path 1. An attacker creates a pull request in the monitored `openai/parameter-golf` repository. 2. The attacker places crafted control sequences, misleading Unicode formatting, fabricated leaderboard text, or instruction-like content in the PR title. 3. The victim or AI Agent runs `scripts/monitor.py`. 4. `fetch_prs()` retrieves the attacker-controlled metadata from the GitHub A ...[truncated 1311 chars]
Remediation
View remediation

SkillSpector

Vulnerability Patterns
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
  • MCP Least PrivilegeUnderdeclared Capability, Wildcard Permission, Missing Permission Declaration
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
Findings (4)

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
93% confidence
Finding

The skill metadata and README describe broad automatic triggering for generic leaderboard- and competition-related queries, which can cause the agent to invoke this skill in contexts where the user did not clearly request GitHub leaderboard monitoring. Over-broad activation increases the chance of unnecessary external data access and context hijacking, especially when similar terms are used in unrelated conversations.

Content

No source excerpt is available for this finding.

Undeclared Tool Scope

Medium
Category
MCP Least Privilege
Confidence
94% confidence
Finding

The skill instructs the agent to run a Python script that makes outbound GitHub API requests, but the manifest does not declare any tool scope or allowed network capability. This creates an authorization gap: the skill can induce network access without an explicit permission boundary, making review and enforcement harder and increasing the chance of unintended external data access.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
94% confidence
Finding

The activation text uses broad phrases like 'check the competition', 'show me the standings', and generic mentions of competitor scores, which can cause the skill to trigger in contexts where the user did not specifically request this repository lookup. Over-broad invocation increases the risk of unnecessary network calls and incorrect tool selection, especially because the skill is network-capable.

Content

No source excerpt is available for this finding.

External Transmission

Medium
Category
Data Exfiltration
Confidence
60% confidence
Finding

Data is being sent to an external URL. This could be legitimate telemetry or data exfiltration. Manual review is recommended.

Content

Scanner excerpt · scripts/monitor.py (reported line 25)May include surrounding context.

python
import urllib.request
from datetime import datetime

API_BASE = "https://api.github.com/repos/openai/parameter-golf"
BPB_PATTERN = re.compile(r"val_bpb[=:\s]*(\d+\.\d+)")
SCORE_IN_TITLE = re.compile(r"(\d+\.\d{3,})")

Static analysis

No suspicious patterns detected.