The Agent Skills Directory

[PERSISTENCE]: The skill creates a persistent execution mechanism by modifying the project's ".claude/settings.json" to register a "SessionStart" hook. This hook executes a shell script (".claude/hooks/session-start.sh") every time the agent session starts, resumes, or clears, ensuring the skill's generated guidelines are injected into the agent's context automatically and persistently.
[PRIVILEGE_ESCALATION]: The skill generates a shell script at runtime and explicitly uses "chmod +x" to grant it execution permissions. This allows the skill to create and authorize its own executable code within the project environment.
[COMMAND_EXECUTION]: The skill makes extensive use of shell commands to manipulate the project environment, including "mkdir", "cat" for file writing, "jq" for JSON configuration editing, and "mktemp"/"mv" for atomic file updates. These operations are used to install persistent hooks and modify project settings.
[INDIRECT_PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection as it ingests untrusted user input (brand descriptions and adjectives) and incorporates them into a 'Creative Brief' that is subsequently designated as the 'design-system authority' for the agent.
Ingestion points: Step 1 captures user-provided strings for project description, target users, and brand adjectives.
Boundary markers: There are no boundary markers or instructions to ignore potential commands embedded within the user-supplied brand descriptors.
Capability inventory: The skill possesses capabilities for file writing, configuration modification, and creating persistent execution hooks.
Sanitization: The skill does not perform any validation or sanitization on the user's input before placing it into the brief or the generated HTML preview.

brand-design