The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill instructs the agent to ingest content from external files like .agents/product-marketing.md, .claude/product-marketing.md, and product-marketing-context.md at runtime. If these files are populated with untrusted content, they could influence the agent's behavior.
Ingestion points: The skill body in SKILL.md directs the agent to read multiple potential product marketing context files.
Boundary markers: None present; the agent is not instructed to treat this external data as untrusted or to ignore embedded instructions.
Capability inventory: The skill primarily performs text analysis, planning, and template generation based on provided data.
Sanitization: None present; the skill lacks validation or filtering for the external context files.
[METADATA_POISONING]: The description field in the frontmatter of SKILL.md contains an exhaustive list of synonyms and keyword variants (e.g., "pSEO," "gerar 100 páginas," "landing pages templatizadas") designed to capture a wide range of user intents. This extensive keyword list can interfere with proper agent routing if other skills share overlapping scopes.

programmatic-seo