Back to skill

Security audit

HumanAI Convention

Security checks for vulnerabilities and agentic risk

Overview

The skill is not deceptive or destructive, but it directs agents to send potentially sensitive interview transcripts to an external service without enough operational privacy controls.

Review this skill before installing if transcripts may include personal, medical, financial, professional, relational, or third-party information. Use it only with informed participant consent, review the exact JSON payload before sending, redact identifiers and secrets, and confirm the external service's retention and deletion practices.

Vulnerability Patterns
  • 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
  • Unauthorized Access and Privilege EscalationObtains permissions beyond the task's legitimate needs
Findings (1)

other

Warning
Location
SKILL.md:37
Finding

Potentially Sensitive Interview Transcripts Transmitted Without Operational Consent or Data-Minimization Safeguards

Content
View full analysis

Vulnerability Details

File Location: SKILL.md, lines 37–57
Vulnerability Type: Privacy Data Exposure
Risk Level: Medium

Vulnerable Code

bash
curl -X POST https://humanaiconvention.com/v1/agent/participate \
  -H "Content-Type: application/json" \
  -d '{
    "agent_id": "your-agent-name",
    "messages": [
      {"role": "assistant", "content": "Tell me about a moment in the last few days when something shifted or stayed with you. Not a pattern — one specific time."},
      {"role": "user", "content": "[participant response]"},
      {"role": "assistant", "content": "[PIVOT: SENSORY] What were you aware of through your senses right in the middle of it?"},
      {"role": "user", "content": "[participant response]"},
      {"role": "assistant", "content": "[PIVOT: TEMPORAL] Walk me through the moments right before..."},
      {"role": "user", "content": "[participant response]"},
      {"role": "assistant", "content": "[COMPRESSION: kernel] If you could only keep one detail that holds it all — which detail would it be?"},
      {"role": "user", "content": "[participant response]"}
    ],
    "config": {
      "include_gfs": true,
      "include_receipt": true
    }
  }'

Technical Analysis

The documented workflow submits an agent identifier and the complete interview transcript to an external service. The interview prompts solicit recent experiences, sensory details, emotional context, and potentially relational information. Participant responses may therefore contain names, locations, health information, private events, or other identifying and sensitive data.

Although the skill later refers to consent as a general principle, the operational submission procedure does not require the agent to:

  • Obtain informed consent immediately before external transmission.
  • Identify the external recipient and explain the processing purpose.
  • Show the exact payload to the part ...[truncated 2092 chars]
Remediation
View remediation

Remediation Suggestions

  1. Require explicit, informed consent immediately before transmission and clearly identify the external service, purpose, and categories of data being sent.
  2. Present the final payload to the user or participant and require affirmative confirmation before invoking curl.
  3. Redact names, locations, contact information, credentials, account identifiers, and unrelated third-party details by default.
  4. Submit only the minimum excerpts necessary to calculate the requested score rather than the complete transcript.
  5. Document the service's retention duration, deletion mechanism, access controls, secondary-use policy, and applicable privacy terms.
  6. Warn users not to include secrets, authentication tokens, medical records, financial information, or other high-risk data.
  7. Provide an opt-out or local-only scoring mode where feasible.
  8. Add a pre-submission validation step that detects likely sensitive information and blocks transmission until it is removed or explicitly approved.
Vulnerability Patterns
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • 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 (2)

External Transmission

Medium
Category
Data Exfiltration
Confidence
84% confidence
Finding

The skill depends on curl and directs the agent to interact with a third-party HTTPS endpoint, which creates an external data transmission path. In this skill's context, that transmission is central to the feature, but it still carries security and privacy risk because user-provided content is sent outside the local trust boundary to a remote service.

Content

Scanner excerpt · SKILL.md (reported line 9)May include surrounding context.

md
openclaw:
    requires:
      bins:
        - curl
    homepage: https://humanaiconvention.com/agent
    emoji: "∴"
---

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
94% confidence
Finding

The skill instructs users to POST full interview transcripts to an external API, and those transcripts are likely to contain sensitive personal or emotional information. The documentation does not provide any warning, consent language, data minimization guidance, retention details, or privacy safeguards, so an agent could exfiltrate user content off-platform without the user understanding the privacy impact.

Content

No source excerpt is available for this finding.