Back to skill

Security audit

The Symposium

Security checks for vulnerabilities and agentic risk

Overview

The skill is mostly coherent for running an AI philosophy dialogue, but its command template and optional web-app install path create review-worthy execution and supply-chain risk.

Install only if you are comfortable sending symposium questions and generated dialogue context to OpenRouter under your API key. Avoid putting secrets, private documents, or sensitive personal data in prompts. If using the optional web app, review or pin the repository and dependencies before running bun install, and do not run setup commands with elevated privileges.

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 (2)

T09 · Insecure Skill Coding Practices

Error
Location
SKILL.md:102
Finding

Shell and JSON Injection Through Unescaped Dialogue Content

Content
View full analysis
", "messages": [ {"role": "system", "content": ""}, {"role": "user", "content": "Question: \n\nPrevious speeches:\n\n\nGive your speech."} ], "temperature": 0.8, "max_tokens": 1500 }' ``` ### Technical Analysis The instructions place the user-controlled question and model-generated previous speeches directly inside a single-quoted shell argument. No JSON serialization, shell escaping, or safe parameter-passing mechanism is specified. If an agent implements this template through direct textual substitution, an input containing a single quote can terminate the shell argument. The remainder of the input can then be interpreted as shell syntax. Independently, unescaped quotation marks, backslashes, or control characters can corrupt the JSON request. This issue affects both: - ``, which is directly controlled by the user. - ``, which contains externally generated model output and therefore must also be treated as untrusted. ### Attack Path 1. An attacker invokes the Skill with a philosophical question containing a single quote followed by shell syntax. 2. The agent directly replaces `` in the documented `curl` command. 3. The single quote terminates the `-d` shell argument. 4. The injected shell metacharacters cause an additional command to execute. 5. That command runs with the same local privileges and environment as the agent's Bash tool. 6. The injected command may read accessible files, modify the working directory, or send environment variables such as `OPENROUTER_API_KEY` to ...[truncated 807 chars]
Remediation
View remediation
request.json curl --fail-with-body --silent --show-error \ https://openrouter.ai/api/v1/chat/completions \ -H "Authorization: Bearer $OPENROUTER_API_KEY" \ -H "Content-Type: application/json" \ --data-binary @request.json ``` ]]>

T08 · Insecure Dependencies

Warning
Location
SKILL.md:183
Finding

Mutable Remote Repository and Unpinned Dependency Installation

Content
View full analysis
Remediation
View remediation
bun install --frozen-lockfile --ignore-scripts ``` The exact Bun flags should be verified against the supported version, and required scripts should be enabled only after review. ]]>
Vulnerability Patterns
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
Findings (3)

Ae1

High
Category
analysis-evasion
Confidence
100% confidence
Finding

Referenced artifact was not completely inspected

Content

Scanner excerpt · SKILL.md (reported line 41)May include surrounding context.

md
- "SKILL.md"

Credential Access

High
Category
Privilege Escalation
Confidence
60% confidence
Finding

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

Content

Scanner excerpt · SKILL.md (reported line 185)May include surrounding context.

bash
git clone https://github.com/dyz2102/symposium.git
cd symposium && bun install
cp .env.example .env.local  # add your OPENROUTER_API_KEY
bun dev

External Transmission

Medium
Category
Data Exfiltration
Confidence
95% confidence
Finding

The skill sends user-provided questions and the full accumulated dialogue to an external third-party service (OpenRouter) using the user's API key. This is a real data egress behavior: any sensitive information a user includes in the prompt, or that appears in prior speeches, is transmitted off-box to a remote model provider, which may violate user expectations or organizational data-handling rules.

Content

Scanner excerpt · SKILL.md (reported line 102)May include surrounding context.

User message: The question + all previous speeches as context.

bash
curl -s https://openrouter.ai/api/v1/chat/completions \
  -H "Authorization: Bearer $OPENROUTER_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{