The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill includes external documentation links that have been flagged as malicious by automated reputation scanners.
Evidence: The URL https://jeffallan.github.io/claude-skills/skills/language/php-pro/ is explicitly blacklisted by security services.
Impact: Directing the agent or user to interact with blacklisted domains can lead to the download of malicious payloads or participation in phishing activities.
[INDIRECT_PROMPT_INJECTION]: The skill exhibits an attack surface for indirect prompt injection through the ingestion of external data and user-provided code requirements.
Ingestion points: User requirements for PHP application architecture and implementation are processed at runtime to generate code.
Capability inventory: The skill instructs the agent to execute shell commands, specifically vendor/bin/phpstan, vendor/bin/phpunit, and vendor/bin/pest in the project environment.
Boundary markers: The instructions lack explicit delimiters or warnings to prevent the agent from following instructions embedded within user-provided data or project files.
Sanitization: There is no validation or sanitization mentioned for data incorporated into the shell command execution flow.
[COMMAND_EXECUTION]: The skill documentation encourages the automatic execution of testing and static analysis tools.
Evidence: The 'Core Workflow' section in SKILL.md instructs the agent to: Run vendor/bin/phpstan analyse --level=9 and Run vendor/bin/phpunit or vendor/bin/pest.
Risk: If the project environment is compromised (e.g., through a malicious composer.json or local binary), executing these tools can lead to arbitrary code execution.

php-pro