The Agent Skills Directory

[PROMPT_INJECTION]: The skill instructions create an indirect prompt injection surface by directing the agent to ingest and analyze HTML content from external web pages. A malicious website could attempt to influence the agent's behavior or code generation by embedding hidden instructions within its HTML or metadata.
Ingestion points: SKILL.md directs the agent to "Read the real login page HTML" and "inspect the ARIA snapshot" in the Pre-flight section.
Boundary markers: There are no explicit instructions or delimiters provided to ensure the agent ignores natural language instructions found within the ingested HTML.
Capability inventory: The agent can write to local configuration and script files (codecept.conf.js, steps_file.js) and execute shell commands via npx codeceptjs.
Sanitization: No sanitization or filtering of the external HTML content is described.
Remediation: Users should instruct the agent to treat all HTML content strictly as data for locator identification and to disregard any potential instructions or commands found within the page source.
[COMMAND_EXECUTION]: The skill contains instructions for executing shell commands, specifically npx codeceptjs run, which is used to verify and run the generated automated tests.
[SAFE]: The skill follows security best practices for credential management by explicitly forbidding hardcoded passwords. It mandates the use of .env files for environment-driven configuration, requires the secret() wrapper for masking sensitive data in logs, and specifies that sensitive files must be excluded from version control via .gitignore.

codeceptjs-auth