The Agent Skills Directory

[SAFE]: The skill consists entirely of markdown documentation and prompt templates. It does not contain executable scripts, binaries, or active code components.
[PROMPT_INJECTION]: The templates are designed for legitimate workflow management. There are no instructions that attempt to bypass AI safety guidelines, override core system instructions, or hide malicious intent through role-play or developer mode activations.
[DATA_EXPOSURE_AND_EXFILTRATION]: No sensitive file paths (e.g., .ssh, .aws), hardcoded credentials, or exfiltration patterns (network calls to non-whitelisted domains) were identified. The templates focus on repository-local or research-based data.
[INDIRECT_PROMPT_INJECTION]: The skill provides templates that interpolate external data via placeholders (e.g., {what}, {question}). While this creates an ingestion surface, the risk is mitigated by the structured format (GOAL, ACCEPTANCE, REPORT) which encourages the agent to maintain control over the worker's scope.
Ingestion points: Placeholders in SKILL.md such as {what}, {feature/change}, and {question}.
Boundary markers: Descriptive section headers are used, though strict isolation delimiters (like XML blocks) are not explicitly mandated.
Capability inventory: Dispatched sub-agents typically have file-system access within the project scope.
Sanitization: No explicit sanitization or validation rules are defined in the templates.

delegation-templates