Back to skill

Security audit

BrowseCTL

Security checks for vulnerabilities and agentic risk

Overview

The skill appears to be legitimate browser automation, but installation and default browser-profile handling create meaningful execution and privacy risks that users should review first.

Install only if you trust the publisher and release pipeline, and prefer an isolated browser profile or test account. Avoid running non-interactive session creation unless you pass --no-copy-data, review any batch files before execution, and treat screenshots, REPL history, persisted sessions, copied cookies, and automation profiles as sensitive local data that may need cleanup.

Vulnerability Patterns
  • Remote Payload Retrieval and ExecutionFetches external code whose behavior can change after review
  • Unauthorized Access and Privilege EscalationObtains permissions beyond the task's legitimate needs
  • 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
Findings (3)

T03 · Remote Payload Retrieval and Execution

Error
Location
scripts/install.js:33
Finding

Unverified Remote Native Binary Is Downloaded and Executed

Content
View full analysis
{ const client = url.startsWith("https") ? https : require("http"); client .get(url, (res) => { if (res.statusCode === 302 || res.statusCode === 301) { return download(res.headers.location, destPath).then(resolve, reject); } if (res.statusCode !== 200) { return reject( new Error(`Download failed with status ${res.statusCode}: ${url}`), ); } const file = fs.createWriteStream(destPath); res.pipe(file); file.on("finish", () => { file.close(); resolve(); }); }) .on("error", reject); }); } async function install() { const tmpDir = fs.mkdtempSync(path.join(os.tmpdir(), "browsectl-")); const archivePath = path.join(tmpDir, archiveName); try { await download(url, archivePath); if (isWindows) { execSync( `powershell -Command "Expand-Archive -Path '${archivePath}' -DestinationPath '${tmpDir}'"`, { stdio: "ignore" }, ); } else { execSync(`tar -xzf "${archivePath}" -C "${tmpDir}"`, { stdio: "ignore", }); } const binaryName = NAME + (isWindows ? ".exe" : ""); const extractedBinary = path.join(tmpDir, binaryName); fs.copyFileSync(extractedBinary, dest); fs.chmodSync(dest, 0o755); console.log(`${NAME} v${VERSION} installed successfully`); } finally { ...[truncated 2852 chars]
Remediation
View remediation

T05 · Unauthorized Access and Privilege Escalation

Error
Location
skills/session.md:243
Finding

Non-Interactive Session Creation Defaults to Copying Browser Cookies and Extensions

Content
View full analysis
Remediation
View remediation

T09 · Insecure Skill Coding Practices

Warning
Location
scripts/install.js:64
Finding

Shell Command Injection Through Temporary Extraction Paths

Content
View full analysis
Remediation
View remediation
Vulnerability Patterns
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Memory PoisoningPersistent Context Injection, Context Window Stuffing, Memory Manipulation
  • Tool MisuseTool Parameter Abuse, Chaining Abuse, Unsafe Defaults
  • Rogue AgentSelf-Modification, Session Persistence
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
Findings (18)

Tool Parameter Abuse

High
Category
Tool Misuse
Confidence
80% confidence
Finding

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).

Content

Scanner excerpt · skills/session.md (reported line 478)May include surrounding context.

md
## 7. `session-delete`

Delete a browser session. This sends a WebDriver `DELETE /session` request to quit the browser, removes any temporary profile directory, and removes the session from the persisted store.

### Parameters

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
89% confidence
Finding

The README describes AI-driven browser automation, session management, profile usage, screenshots, clipboard-style paste, and extraction of page content, but it does not warn that these capabilities may access authenticated sessions, cookies, sensitive page contents, or local browser profile data. In this skill context, the omission increases risk because operators may run the tool against logged-in browsers or persistent profiles without understanding the privacy and data-exposure implications.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
94% confidence
Finding

The README states that installation uses an npm postinstall script to download a platform-specific binary, but it does not clearly warn users that installation triggers network retrieval and execution of externally hosted code. In an AI-agent tooling context, this is particularly risky because users may install the package in trusted environments without realizing they are executing a downloaded binary outside the normal audited npm source tree.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
94% confidence
Finding

The skill explicitly states that sessions persist to .browsectl/sessions.json across invocations and also uses a persistent browser profile directory under ~/.browsectl/<browser>-profile, but the skill description does not clearly warn users about the privacy and security implications. In an AI-driven browser automation context, persisted sessions can retain authentication state, cookies, local storage, browsing history, and other sensitive artifacts that later commands or other local users/processes could reuse unintentionally.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The examples demonstrate automating real third-party login flows, QR-code capture, and message submission to live services without an explicit warning that the tool can act on authenticated accounts and transmit user-provided content. In the context of an agent skill, such examples can normalize high-risk actions and lead users or downstream agents to perform account-affecting operations without informed consent or safeguards.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The screenshot command saves captured page content to disk and automatically creates parent directories, but the skill omits any warning that screenshots may contain credentials, personal data, or proprietary information. In an automated agent setting, this can lead to unintended persistence of sensitive visual data in local files or shared workspaces.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
93% confidence
Finding

The last-message-content command is explicitly designed to extract the latest chat content, including HTML, text, images, and links, yet the documentation provides no warning that this may read sensitive or private on-page data. In an AI-driven browser automation context, this increases the risk of silent data harvesting from chat apps, support portals, or authenticated web sessions.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
93% confidence
Finding

The REPL documentation explicitly says command history is automatically persisted to .browsectl/repl_history, but it does not warn users that typed commands may include secrets such as passwords, tokens, personal data, or internal URLs. In a browser-automation REPL, fill and paste commands are likely to carry sensitive input, so silently retaining them on disk increases the chance of credential leakage to other local users, backups, logs, or source-control commits.

Content

No source excerpt is available for this finding.

Context Window Stuffing

Medium
Category
Memory Poisoning
Confidence
85% confidence
Finding

Skill attempts to fill the context window with filler content, displacing legitimate instructions and safety constraints. This can degrade agent performance or bypass safety boundaries.

Content

Scanner excerpt · skills/repl.md (reported line 129)May include surrounding context.

text
browsectl> fill hello world
browsectl> fill input[name="email"] user@example.com

Context Window Stuffing

Medium
Category
Memory Poisoning
Confidence
85% confidence
Finding

Skill attempts to fill the context window with filler content, displacing legitimate instructions and safety constraints. This can degrade agent performance or bypass safety boundaries.

Content

Scanner excerpt · skills/repl.md (reported line 373)May include surrounding context.

text
browsectl> fill hello world
browsectl> fill input[name="email"] user@example.com

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
92% confidence
Finding

The screenshot command writes page content to a PNG file on disk, but the documentation does not warn that screenshots can capture secrets, session data, QR login codes, personal information, or proprietary content. In this skill context, browser automation is often used on authenticated pages, making saved screenshots a realistic source of data exposure if files are shared, backed up, or left in predictable locations.

Content

No source excerpt is available for this finding.

Session Persistence

Medium
Category
Rogue Agent
Confidence
70% confidence
Finding

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.

Content

Scanner excerpt · skills/session.md (reported line 251)May include surrounding context.

md
| Flag | Type | Default | Description |
|---|---|---|---|
| `--foreground` | `bool` | `false` | Run session creation in the current process (no background worker). |
| `--detach` | `bool` | `false` | Spawn a background worker and return immediately (fire-and-forget). |
| `--copy-data` | `bool` | `false` | Copy cookies and extensions from the real browser profile. Skips the interactive prompt. |
| `--no-copy-data` | `bool` | `false` | Do not copy any user data. Skips the interactive prompt. Conflicts with `--copy-data`. |

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The skill documents that in non-interactive mode, session creation defaults to copying cookies and extensions from the real browser profile. That behavior can silently transfer authentication material and installed extension state into an automation-controlled profile, increasing the risk of unintended access to sensitive accounts or leakage of private browsing context. In a browser automation skill, this is especially risky because non-interactive use is common in agent workflows where users may not realize privileged session data is being imported.

Content

No source excerpt is available for this finding.

Session Persistence

Medium
Category
Rogue Agent
Confidence
60% confidence
Finding

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.

Content

Scanner excerpt · skills/session.md (reported line 609)May include surrounding context.

md
│  2. browsectl status                                        │
│     → Verify chromedriver is reachable (optional)           │
├─────────────────────────────────────────────────────────────┤
│  3. browsectl session-create                                │
│     → Starts driver if needed                               │
│     → Copies profile data if requested                      │
│     → Launches browser, creates WebDriver session           │

Missing User Warnings

Low
Category
Not specified by scanner
Confidence
87% confidence
Finding

The README notes that sessions persist to .browsectl/sessions.json, but it does not frame this persistence as a user-impacting behavior or caution that browser automation state may be retained between runs. Because persistent session data can affect privacy and system state, the skill description should include an explicit warning so users understand the retention behavior.

Content

No source excerpt is available for this finding.

Missing User Warnings

Low
Category
Not specified by scanner
Confidence
92% confidence
Finding

The documentation explicitly describes writing screenshots to local paths and persisting session tab state to .browsectl/sessions.json, but does not warn users that local artifacts containing potentially sensitive page content and session metadata will be created. In an AI-driven browser automation context, this increases the risk of unintentional storage of credentials, QR codes, account data, or browsing state on disk.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
80% confidence
Finding

The example workflow navigates to doubao.com and shows a resulting page title of 豆包, which implicitly centers a specific language/locale experience in the skill documentation. Because no language or locale choice or justification is provided, this can conflict with the policy against forcing a specific language without user opt-in.

Content

No source excerpt is available for this finding.

Missing User Warnings

Low
Category
Not specified by scanner
Confidence
86% confidence
Finding

This markdown file documents create_tab opening a new browser tab and optionally navigating to a provided URL, which can trigger outbound network requests and load remote content. The section explains usage but does not include any warning about privacy, data exposure, or the fact that opening a URL will contact external sites.

Content

No source excerpt is available for this finding.