The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill's core functionality involves ingesting data from past campaign outcomes, channel-specific drafts, and external briefs to inform its 'Compound' and 'Optimise' stages, creating a vulnerability to malicious instructions embedded in that data.
Ingestion points: Reads content from .agents/marketing-strategy.md, .agents/marketing-learnings.md, and campaign-specific files stored in the .agents/campaigns/ directory.
Boundary markers: The skill lacks explicit instructions or formatting (such as XML tags or specialized delimiters) to isolate ingested data from the agent's system instructions, which could allow malicious data to influence agent behavior.
Capability inventory: The skill has permissions to write to the local filesystem (.agents/marketing-learnings.md) and sequences multiple other skills for execution, including advertising and email tools.
Sanitization: No sanitization or validation logic is specified for data retrieved from campaign drafts or external strategy files before it is processed by the agent.

compound-marketing