The Agent Skills Directory

[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because it ingests and processes untrusted data from external websites and visitor feedback.
Ingestion points: The Intake section in SKILL.md collects external URLs, live copy, CSS, and analytics screenshots. Phase 1 also ingests "real visitor input" such as surveys and support tickets.
Boundary markers: The instructions do not provide explicit delimiters or "ignore embedded instructions" warnings when handling this external data.
Capability inventory: The skill has the capability to write and extend multiple files in the local docs/ directory and invoke various sub-skills.
Sanitization: There is no mention of sanitization, validation, or escaping of the external content before it is processed by the agent.
[COMMAND_EXECUTION]: The skill instructs the agent to suggest shell commands to the user for environment configuration.
Evidence: Operating Rule 4 in SKILL.md directs the agent to offer the command npx skills add wondelai/skills/<slug> --global if a required sub-skill is not found in the environment.
[EXTERNAL_DOWNLOADS]: The skill promotes the downloading and installation of external code packages from a third-party source.
Evidence: The workflow relies on constituent skills from the wondelai/skills/ repository and provides a mechanism to install them using the npx package runner.

improve-website