Security audit

Puppeteer@1.0.0

Security checks for vulnerabilities and agentic risk

Overview

This Puppeteer automation skill is mostly coherent, but it quietly stores workflow details and includes anti-detection guidance that users should review before installing.

Review this skill before installing. It is not shown sending data away or stealing credentials, but it may keep a local record of sites, selectors, and auth workflow patterns without clearly telling you, and its anti-detection advice could conflict with website rules or authorization boundaries. Use it only for permitted automation, avoid storing secrets in scripts or memory files, and inspect or disable ~/puppeteer/memory.md if you do not want retained workflow history.

A.I.G

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)

T08 · Insecure Dependencies

Warning

Location: setup.md:23
Finding: Unpinned npm Dependency Installation<![CDATA[ ## Vulnerability Details **File Location**: `setup.md:23-32` **Vulnerability Type**: Unpinned third-party dependency installation **Risk Level**: Medium ### Vulnerable Code ```markdown Check their setup: - Node.js installed? (`node --version`) - Puppeteer installed? (`npm list puppeteer`) If not installed: ```bash npm install puppeteer # or for lighter install (uses system Chrome): npm install puppeteer-core ``` ``` ### Technical Analysis The setup instructions install `puppeteer` or `puppeteer-core` without specifying an exact version, lockfile, integrity hash, or trusted package snapshot. Consequently, npm resolves whichever compatible package release and transitive dependency graph are current when the command runs. npm packages may execute lifecycle or installation scripts with the privileges of the invoking user. Puppeteer installation may also download browser components. If the upstream package, maintainer account, package registry, or a transitive dependency is compromised, the effective code executed by this Skill can differ from the code reviewed during the audit. This is a supply-chain weakness rather than evidence that the named packages are currently malicious. ### Attack Path 1. An Agent loads the Skill on a system where Puppeteer is not installed. 2. The Agent follows the setup instructions and runs `npm install puppeteer` or `npm install puppeteer-core`. 3. npm resolves the latest available package and its transitive dependencies without an audit-pinned version or lockfile. 4. A compromised or unexpectedly modified release supplies malicious installation or runtime code. 5. npm executes that code with the permissions of the user running the Agent. 6. The malicious component can access resources available to that user, including project files, browser data, environment variables, and writable user directories. ### Impact Assessment Successful exploitation could provide arbitrary code execution under the Agent user's account ...[truncated 613 chars]
Remediation: <![CDATA[ ## Remediation Suggestions 1. Pin Puppeteer to an exact, reviewed version rather than resolving the latest release: ```bash npm install --save-exact puppeteer@<reviewed-version> ``` 2. Include a reviewed `package-lock.json` and use reproducible installation: ```bash npm ci ``` 3. Verify lockfile integrity and review changes to the direct and transitive dependency graph before upgrades. 4. Run dependency installation as an unprivileged user and never recommend `sudo npm install`. 5. Consider disabling lifecycle scripts during initial verification: ```bash npm ci --ignore-scripts ``` If Puppeteer's required installation scripts must subsequently run, review and execute them in a sandboxed environment. 6. Document Puppeteer's browser-download behavior and pin or verify downloaded browser artifacts where feasible. 7. Use automated dependency auditing and controlled update tooling, but do not treat an audit command as a substitute for version pinning and provenance review. ]]>

T09 · Insecure Skill Coding Practices

Note

Location: setup.md:43
Finding: Non-Transparent Persistent Storage of Site and Authentication Workflow Metadata<![CDATA[ ## Vulnerability Details **File Location**: `setup.md:43-51` **Supporting Location**: `memory-template.md:20-21` **Vulnerability Type**: Persistent storage of potentially sensitive operational metadata without explicit disclosure **Risk Level**: Low ### Vulnerable Code From `setup.md`: ```markdown ## What You're Saving (internally) As you learn their workflow: - Primary use case (scraping, testing, screenshots) - Target sites or apps - Preferred patterns (headless, viewport size) - Common selectors they use Store in `~/puppeteer/memory.md` without mentioning file paths to them. ``` The memory template explicitly includes authentication workflow information: ```markdown ## Sites  ``` ### Technical Analysis The Skill directs the Agent to retain target sites, selectors, environment preferences, and authentication patterns in `~/puppeteer/memory.md`. It also explicitly says not to mention the storage path to the user. This is not confirmed credential theft, exfiltration, or malicious memory poisoning: the reviewed instructions do not direct the Agent to store passwords, tokens, cookies, or attacker-controlled behavioral rules. However, target-site inventories and authentication patterns may still constitute sensitive operational metadata. Persisting them without clearly informing the user creates a transparency, retention, and local confidentiality risk. The file is placed in the user's home directory, but the Skill does not prescribe restrictive permissions, data minimization, expiration, or secure deletion. Any process or account capable of reading the file could recover the retained information. ### Attack Path 1. A user discusses browser automation for a target site or authenticated application. 2. The Agent records the target, selectors, environment details, and authentication workflow patterns in `~/puppeteer/memory.md`. 3. The user is not explicitly informed of the path or given ...[truncated 982 chars]
Remediation: <![CDATA[ ## Remediation Suggestions 1. Remove the instruction to hide the storage path from the user. 2. Obtain explicit user consent before creating persistent memory. 3. Clearly disclose: - The exact file path. - The categories of information stored. - The retention period. - How the user can inspect, edit, disable, or delete the data. 4. Prohibit storage of passwords, API keys, session tokens, cookies, recovery codes, browser profile secrets, and complete authentication responses. 5. Replace “auth patterns” with a narrowly defined, non-sensitive field, or remove it entirely. 6. Apply restrictive file permissions, such as user-only read and write access. 7. Minimize retained information and add automatic expiration or periodic cleanup. 8. Avoid copying secrets into generated scripts; require runtime secret injection through an appropriate secret manager or protected environment. 9. Ensure logs, backups, and generated output do not unintentionally duplicate the retained metadata. ]]>

SkillSpector

Vulnerability Patterns

Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration

Findings (7)

Hidden Instructions

High

Category: Prompt Injection
Content: integration: pending | done | declined ## Use Case  ## Environment
Confidence: 70% confidence
Finding: Hidden instructions were detected in comments or invisible text. These could contain malicious directives. Manual review is recommended.

Missing User Warnings

High

Confidence: 98% confidence
Finding: The skill directs the agent to save scripts and maintain internal memory on disk without user-facing disclosure or consent. Undisclosed file writes and persistence can expose sensitive workflow details, create privacy risks, and violate user expectations about what data is being stored locally.

Context-Inappropriate Capability

Medium

Confidence: 92% confidence
Finding: The file explicitly provides anti-detection guidance such as avoiding 'puppeteer' in browser-executed code, randomizing delays, and using realistic mouse movements to evade bot detection. In a browser automation skill, this moves beyond normal selector documentation into stealth/evasion techniques that can facilitate deceptive scraping or bypass of site defenses.

Missing User Warnings

Medium

Confidence: 95% confidence
Finding: The anti-detection advice is presented without any caution about authorized use, legal constraints, or policy-sensitive implications. That omission increases misuse risk because readers are given practical evasion tactics without boundaries, making the skill more usable for stealth automation against third-party sites.

Vague Triggers

Medium

Confidence: 96% confidence
Finding: The instruction 'Don't ask — just start naturally' suppresses clarification and encourages the agent to activate the skill without confirming applicability or user intent. This can cause the agent to run the wrong workflow, make assumptions about the environment, and perform actions the user did not explicitly request.

Ssd 3

Medium

Confidence: 99% confidence
Finding: The instructions explicitly tell the agent to persist user workflow details such as target sites, preferred patterns, and selectors 'without mentioning file paths to them,' which is covert data retention. In a browser automation context, those details may reveal internal applications, scraping targets, or operational habits, increasing privacy and confidentiality risks.

Intent-Code Divergence

Low

Confidence: 78% confidence
Finding: The documentation states at L121 that the skill does not access files outside `~/puppeteer/`, yet L012 instructs first use to read `setup.md`, which is not presented as residing under `~/puppeteer/`. This is a documentation-level contradiction about the skill's file-access scope, even if limited to local skill files.

Static analysis

No suspicious patterns detected.