Security hardening: fix W007 and W011 third-party content exposure in playwright-cli skills

[fugazi]

Summary

The latest security audit for microsoft/playwright-cli (Snyk report via skills.sh) flags 2 issues with overall HIGH risk:

• W007 (HIGH): Insecure credential handling in skill instructions
• W011 (MEDIUM): Third-party content exposure (indirect prompt injection risk)

Audit link: https://skills.sh/microsoft/playwright-cli/playwright-cli/security/snyk

Why this matters

• skills.sh is highly visible and widely used; insecure examples can be copied at scale.
• Current skill/docs patterns may normalize passing secrets in plain command arguments.
• The tool is designed to open and process arbitrary web content, which increases indirect prompt injection risk.
• Keeping security audits green is essential for user trust, safe adoption, and long-term maintenance quality.

Findings (from audit)

1) W007 HIGH — Insecure credential handling

Examples and patterns in skill/docs show literal credentials/tokens/cookie values in commands or generated snippets.

Risk: secrets can leak via terminal history, logs, snapshots, traces, CI output, or shared transcripts.

2) W011 MEDIUM — Third-party content exposure

Skill/docs encourage opening/scraping external sites and processing page content in ways that may influence downstream actions.

Risk: untrusted web content can inject malicious instructions and influence agent behavior (indirect prompt injection).

Proposed remediation plan

• Replace literal credential/token examples with safe placeholders.
• Add clear secret-handling guidance (env vars/secrets file, no plaintext secrets in commands).
• Add explicit warning that external web content is untrusted.
• Add regression tests for W007 and W011 patterns.
• Re-run the audit and confirm PASS.

Acceptance criteria

• No docs/skill examples include plaintext secrets or realistic credential literals.
• Security guidance for secret handling and untrusted content is explicit and easy to find.
• At least one automated test exists for each risk class (W007, W011).
• Audit re-run shows no W007/W011 findings (or reduced risk with documented rationale).

References

• Audit page: https://skills.sh/microsoft/playwright-cli/playwright-cli/security/snyk
• Repository: https://github.com/microsoft/playwright-cli