Security audit
Security checks for vulnerabilities and agentic risk
This is a local interview-coaching skill with disclosed profile and practice-history storage, and no evidence of hidden network access or malicious behavior.
Install only if you are comfortable with the skill keeping interview-prep profile details, practice answers, scores, and weak areas in local files. Review or delete ~/.openclaw/interview-coach if you no longer want that data retained, and be aware that the broad "help" trigger may activate the skill more often than intended.
Skill manipulates agent memory, state, or stored context. Memory corruption can alter personality, override safety rules, or cause unpredictable behavior.
⭐ STAR BREAKDOWN
━━━━━━━━━━━━━━━━━━
S — Situation: 8/10 ✅ Clear context
T — Task: 6/10 ⚠️ Be more specific about YOUR role
A — Action: 9/10 ✅ Great detail on what you did
R — Result: 5/10 ❌ Add numbers! "Improved efficiency" → "by 35%"
The command list includes the generic trigger 'help,' which is so broad that it could activate during many unrelated interactions. A universal trigger substantially raises the risk of unintended invocation and unnecessary access to local files, especially in systems where skills are auto-selected from user text.
The top-level description says the skill should activate for broad phrases ending with 'or any job interview task,' which is an overly vague catch-all. Broad activation increases the chance the skill intercepts unrelated user requests and unnecessarily gains access to local interview-history/profile data in contexts the user did not clearly intend.
The skill persistently stores sensitive career-related information such as target company, experience, past roles, answers, weak areas, and interview history across sessions. While local-only persistence is less dangerous than exfiltration, retaining this data without clear minimization, retention, and consent controls can expose private user information to other local processes, users, or later unintended skill behavior.
On first message, create data directory:
mkdir -p ~/.openclaw/interview-coach
The trigger list includes common phrases such as 'interview tips,' 'quick prep,' and 'follow up email' that can appear in benign, unrelated conversation. This can cause over-activation, making the skill handle requests outside its intended scope and exposing or modifying persistent local data when not necessary.
The documentation says exec is needed to create the data directory, but the manifest metadata on L0004 only declares read and write. This is an active documentation-to-manifest contradiction about the skill's intended capabilities, especially because the setup instructions explicitly rely on shell execution.
No suspicious patterns detected.