The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill's instructions create a vulnerability surface for indirect prompt injection by directing the agent to load and follow context from external files that could be controlled by an adversary.
Ingestion points: The SKILL.md file (Initial Assessment section) instructs the agent to search for and read files such as .agents/product-marketing.md, .claude/product-marketing.md, or product-marketing-context.md to gather product context before generating content.
Boundary markers: The instructions do not provide delimiters or specific guidelines to separate the external data from the agent's core instructions, increasing the likelihood that the agent might follow malicious directives embedded in these context files.
Capability inventory: The skill facilitates the generation of public-facing marketing copy, SEO strategy documents, and structured YAML data files based on the ingested content.
Sanitization: There are no mechanisms described for validating, filtering, or sanitizing the content of the marketing files before it is processed by the agent.
[NO_CODE]: The skill does not contain any executable scripts, binaries, or automated shell commands. It relies entirely on natural language instructions, structured data templates, and markdown documentation, which limits the risk of direct code execution attacks.

competitors