The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses npx lighthouse to perform performance audits on target URLs. This involves executing a command-line tool that fetches and analyzes remote content based on user input.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection (Category 8) because it ingests untrusted data from external websites via Lighthouse reports without sanitization or boundary markers.
Ingestion points: Data from lighthouse-report.json is processed and incorporated into the agent's context as described in SKILL.md.
Boundary markers: None identified; external report content is mixed with agent instructions.
Capability inventory: The skill uses npx for command execution and has instructions for modifying build configurations (Webpack, Vite, Next.js).
Sanitization: No validation or filtering of the external audit results is performed before processing.
[SAFE]: The skill metadata (author: cloudflare, source: cloudflare/skills) contradicts the provided author context (oimiragieo). This is classified as deceptive metadata (Category 7), which can lead to misjudgment of the skill's provenance and safety profile.
[SAFE]: The 'Memory Protocol' section instructs the agent to persist state (learnings, issues, and decisions) to the .claude/context/memory/ directory. While this is a form of persistence that influences future behavior, it is a standard state-management pattern for certain agent environments.

web-perf