The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill relies on reading and processing data from external files (e.g., .project/strategy/problem-definition.md). This creates a surface for indirect prompt injection if those files contain malicious instructions.
Ingestion points: User inputs during 'discovery' and existing files within the .project/ directory.
Boundary markers: The skill uses structured templates and explicit validation checklists to delimit content, though these do not fully prevent adversarial instructions.
Capability inventory: The agents primarily perform text analysis and file writing/reading via the project structure.
Sanitization: The workflow includes a mandatory 'Human Validation' step where a human review is required before final publication/usage, significantly mitigating the risk of automated exploitation.
[COMMAND_EXECUTION]: The skill includes shell script snippets in SKILL.md and JavaScript files in the tests/ directory. These are used for project governance and CI/CD validation (e.g., checking if a 'degraded mode' deadline has passed or verifying the existence of required marketing documents). The paths used in these scripts are protected against traversal attacks using normalized relative path checks.

direction-marketing