Security audit
Security checks for vulnerabilities and agentic risk
The skill is a disclosed autonomous coding workflow that writes project progress files and commits changes, with no hidden install, exfiltration, or system persistence found.
Install this only if you want an agent to autonomously modify your repository, create .autopilot tracking files, run checks, use browser validation, perform web/GitHub research when needed, and make commits. For sensitive or private codebases, set clear boundaries about network use, files in scope, and whether commits are allowed before enabling the mode.
Tool parameters are crafted to achieve unintended or unsafe behavior. Parameter abuse can bypass intended safety checks (e.g. shell=True, --force, dangerous glob patterns).
- "GET /api/[resource]/:id returns single item"
- "POST /api/[resource] creates item"
- "PUT /api/[resource]/:id updates item"
- "DELETE /api/[resource]/:id deletes item"
risk: medium
files_hint: ["src/api/[Resource].ts"]
The activation guidance includes broad terms like 'autopilot' and 'autonomous loop' that are likely to appear in normal user requests, increasing the chance the skill activates outside narrowly intended scenarios. Because this skill authorizes unattended iteration, file creation, research, browser use, and commits, accidental activation meaningfully expands the agent's operational scope without a deliberate opt-in.
The skill advertises web/GitHub research and browser-based validation but provides no clear warning that external network access, browsing, and potential data exposure may occur. In an autonomous loop, this can cause code, metadata, URLs, or internal context to be sent to third-party services or loaded into browser sessions without the user's informed consent.
The skill directs the agent to create and maintain project files under .autopilot/ and later to commit changes, but it does not require an explicit upfront warning or consent for repository modifications. In unattended mode, this can lead to unexpected persistence of artifacts, pollution of the working tree, and unauthorized commits of sensitive or low-quality changes, especially because the loop minimizes user intervention.
Skill establishes unauthorized persistence across sessions via cron jobs, startup scripts, or state files. Session persistence allows an attacker to maintain access beyond the current interaction.
### When blocked:
- Add to 🚫 count in status summary
- Add entry to Risk Log
### On completion:
- Fill in Completion Summary section
No suspicious patterns detected.