The Agent Skills Directory

[PERSISTENCE]: The skill's primary function is to generate new agent skills within the ~/.claude/skills/ directory. These generated skills are described as 'user-level' and are automatically loaded by the agent environment at startup, creating a mechanism for persistent instructions across sessions.
[INDIRECT_PROMPT_INJECTION]: Phase 0 (Context Ingestion) defines a workflow where the agent fetches and analyzes content from arbitrary user-provided URLs and local documents (PDF, DOCX, images). This content is used to extract 'brand signals' which are then used to populate the generated skill. There are no explicit boundary markers or sanitization steps mentioned to prevent malicious instructions embedded in these external sources from influencing the synthesis process.
Ingestion points: SKILL.md (Phase 0) describes ingestion from Website URLs, Newsletters, Social Media profiles, and local documents.
Boundary markers: None specified in the extraction or analysis protocols.
Capability inventory: The skill uses WebFetch for network access, file system read capabilities for local documents, and file system write capabilities to the ~/.claude/skills/ directory.
Sanitization: No explicit sanitization or filtering of external content is defined before the content is used for dimension synthesis.
[DYNAMIC_EXECUTION]: The skill uses a template-based approach (references/brand-skill-template.md) to generate new SKILL.md files. Because these files contain instructions for the agent and are generated using data synthesized from potentially untrusted external inputs (Phase 0), this constitutes the dynamic generation of executable agent instructions.
[COMMAND_EXECUTION]: Phase 4 of the workflow explicitly instructs the agent to invoke an external skill (creator-stack:design-system) to generate additional assets, demonstrating automated tool chaining.

brand-guidelines