The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill instructs the agent to read and prioritize instructions from local project files (e.g., brand-context.md). This establishes a surface where a maliciously crafted file in the user's environment could attempt to influence the agent's behavior. However, the skill lacks dangerous tools or command execution capabilities, limiting the potential impact to textual output generation.
Ingestion points: Files named brand-context.md located at the project root, ./.claude/brand-context.md, or ./docs/brand-context.md are read in full.
Boundary markers: The skill does not implement delimiters or specific escaping for this content; instead, it explicitly directs the agent that the file's content 'overrides an agent's own defaults'.
Capability inventory: No shell execution, tool usage, or network operations are defined in the analyzed files.
Sanitization: No validation or sanitization is performed on the ingested file content.
[SAFE]: The skill consists entirely of natural language instructions and does not contain any code, scripts, or external dependencies.

abm-ops