The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill's 'Phase 0' project discovery process ingests untrusted project files such as README.md, package.json, and other arbitrary project documentation to build marketing context.
Ingestion points: Phase 0 automated exploration of the codebase.
Boundary markers: The instructions lack explicit delimiters or warnings to ignore embedded instructions within the ingested project files.
Capability inventory: The skill has the ability to write project files (MARKETING.md) and execute tools like Grep and local scripts.
Sanitization: No sanitization or validation of the content read from the project files is described.
[DATA_EXFILTRATION]: The skill instructs the agent to 'search for email/ad/analytics config' within the project root. Accessing configuration files is a data exposure risk as these files frequently contain credentials or sensitive system metadata.
[DYNAMIC_EXECUTION]: The skill relies on and references external executable content, specifically mentioning a Python hook at .claude/hooks/voice-gate.py and commands within a scripts/ directory. While these are local to the environment, they represent execution paths that are not defined within the skill's own code.

kai-write