Back to skill

Security audit

AI-powered DP Platform Operations Advisor

Security checks for vulnerabilities and agentic risk

Overview

The skill is a coherent DP operations helper, but its instructions include unsafe authenticated API usage that could leak keys or execute attacker-controlled response data.

Review before installing. Use only an HTTPS DP_SERVER_URL, avoid exposing DP_API_KEY in logs or shell history, prefer a minimally scoped read-only key, and do not run the stall-detection Python snippet as written unless it is rewritten to pass API responses as data rather than source code. Require explicit confirmation before any restart action.

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 (3)

T09 · Insecure Skill Coding Practices

Error
Location
SKILL.md:179
Finding

Remote API Responses Are Interpolated into Executable Python Source

Content
View full analysis
Remediation
View remediation
"$tmp1" printf '%s' "$PROG2" > "$tmp2" python3 - "$tmp1" "$tmp2" <<'PY' import json import sys with open(sys.argv[1], encoding="utf-8") as file: p1 = json.load(file) with open(sys.argv[2], encoding="utf-8") as file: p2 = json.load(file) # Process the parsed objects here. PY ``` In addition: - Require HTTPS and certificate verification for the DP server. - Reject malformed or unexpectedly large responses. - Validate the parsed JSON against an expected schema. - Apply response size and request timeout limits. - Run operational tooling with a minimally privileged account. - Rotate the API key if exploitation is suspected. ]]>

T09 · Insecure Skill Coding Practices

Error
Location
skill.json:10
Finding

API Key May Be Transmitted over Plaintext HTTP

Content
View full analysis
Remediation
View remediation
&2 exit 1 ;; esac curl --fail-with-body \ --proto '=https' \ --connect-timeout 3 \ --max-time 15 \ -H "X-DP-API-Key: ${DP_API_KEY}" \ "${DP_SERVER_URL%/}/job/status" ``` ]]>

T09 · Insecure Skill Coding Practices

Warning
Location
SKILL.md:82
Finding

Unquoted Server URL Allows Curl Argument Injection

Content
View full analysis
Remediation
View remediation
&2 exit 1 ;; esac curl -s \ --fail \ --proto '=https' \ --connect-timeout 3 \ --max-time 15 \ -- "${DP_SERVER_URL%/}/homepage" \ && echo "DP Server OK" \ || echo "DP Server NOT running" ``` Apply the same controls consistently to every URL expansion: - Always enclose expansions in double quotes. - Validate the URL with a strict parser or allowlist. - Permit only the HTTPS scheme. - If feasible, allowlist expected hostnames and ports. - Place `--` before positional URL arguments. - Avoid inheriting security-sensitive configuration from untrusted process environments. ]]>
Vulnerability Patterns
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
  • MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
Findings (9)

External Script Fetching

High
Category
Supply Chain
Confidence
90% confidence
Finding

Remote code is downloaded and executed. This bypasses code review and could introduce malicious code.

Content

Scanner excerpt · SKILL.md (reported line 104)May include surrounding context.

bash
# Get all job statuses
ALL_STATUS=$(curl -H "X-DP-API-Key: ${DP_API_KEY}" -s "${DP_SERVER_URL}/job/status")

# Parse and categorize
echo "$ALL_STATUS" | python3 -c "

External Script Fetching

High
Category
Supply Chain
Confidence
90% confidence
Finding

Remote code is downloaded and executed. This bypasses code review and could introduce malicious code.

Content

Scanner excerpt · SKILL.md (reported line 154)May include surrounding context.

md
JOB_ID="$1"  # job ID from user or from health check

# Get per-operator metrics
PROGRESS=$(curl -H "X-DP-API-Key: ${DP_API_KEY}" -s "${DP_SERVER_URL}/job/progress?id=$JOB_ID")

echo "$PROGRESS" | python3 -c "
import sys, json

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
93% confidence
Finding

The manifest description says the skill activates when the user mentions checking the platform, job failures, job status, throughput analysis, fault diagnosis, or operations reports. Several of these phrases are broad operational requests, and the file does not define exclusion conditions or tighter scope, which can cause unintended invocation for general troubleshooting conversations.

Content

No source excerpt is available for this finding.

Intent-Code Divergence

Medium
Category
Not specified by scanner
Confidence
96% confidence
Finding

The skill states that API-key authentication is the only supported method and that no session fallback is allowed, but later instructs automatic re-login when authentication fails. This contradiction can cause an agent or operator to attempt unintended alternate authentication flows, increasing the chance of credential misuse, unexpected session handling, or drift from the documented security model.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
93% confidence
Finding

The skill repeatedly demonstrates use of an API key in shell commands and even echoes a prefix of the key, but does not warn about exposure through shell history, terminal logs, process inspection, copied transcripts, or agent logging. In an ops context, credentials often have broad read or control scope, so weak handling materially increases the risk of credential leakage and subsequent unauthorized platform access.

Content

No source excerpt is available for this finding.

Intent-Code Divergence

Medium
Category
Not specified by scanner
Confidence
94% confidence
Finding

The capabilities section advertises auto-restart policy evaluation and execution, while the guardrails later prohibit auto-restart without user confirmation in production. This inconsistency is dangerous because agents may privilege the capability description and perform disruptive operational actions without approval, potentially restarting production jobs and affecting data processing availability or integrity.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
90% confidence
Finding

The natural-language descriptions for the required environment variables are written only in Chinese, which imposes a specific language on users without any opt-in or alternative locale. This matches the policy category for language or locale constraints in natural-language content.

Content

No source excerpt is available for this finding.

Missing User Warnings

Low
Category
Not specified by scanner
Confidence
76% confidence
Finding

The documented workflow performs multiple curl requests to DP platform endpoints for status, progress, and diagnostics, which may transmit operational metadata about jobs and system state. The README describes the behavior functionally but does not explicitly warn users that running these checks contacts the remote DP server and processes potentially sensitive operational data.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
72% confidence
Finding

The file instructs the skill to always structure responses in a fixed format with English section headers, but it does not offer the user any language or locale choice. Because the skill content and triggers are bilingual, forcing one response convention can violate language-choice policy expectations.

Content

No source excerpt is available for this finding.