The Agent Skills Directory

[SAFE]: The skill is designed for text-based content transformation and does not contain instructions for network operations, privilege escalation, or persistence.
[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted text data in the form of user-provided stories, creating a surface for indirect prompt injection.
Ingestion points: The story content provided through natural language triggers or the --story argument (referenced in the Inputs section of SKILL.md).
Boundary markers: No specific delimiters or "ignore instructions" warnings are defined for the interpolation of the story into the prompt.
Capability inventory: The skill has read-only access to brand profiles located in ~/.claude-marketing/brands/. It possesses no capabilities for file writing, network transmission, or shell command execution.
Sanitization: The skill enforces a --client-safe protocol (active by default) to anonymize sensitive client details and mandates that output must respect brand compliance guardrails.
[DATA_EXPOSURE]: The skill reads from an application-specific hidden directory (~/.claude-marketing/brands/). This access is localized to brand configuration data required for the skill's primary function and does not involve accessing system-level credentials or sensitive user directories like .ssh or .aws.

story-mine