The Agent Skills Directory

[PROMPT_INJECTION]: The skill exhibits a surface for indirect prompt injection as it is designed to ingest data from untrusted external sources, such as feature specifications and technical design documents, and interpolate them into instructions for an AI agent.
Ingestion points: Technical design documents (e.g., specs/design-{feature-name}.md) and product requirement skills (e.g., skills/prd-{feature-name}/SKILL.md).
Boundary markers: The templates use structured markdown headers and XML-like tags (e.g., <background_information>, <instructions>) to delimit sections, though they lack explicit warnings to ignore embedded instructions within the source data.
Capability inventory: The generated task commands are granted access to powerful tools including Bash, Write, Edit, and Skill (used for invoking other agent capabilities like code committing).
Sanitization: No explicit sanitization or escaping of the ingested source data is performed before it is written to the command files.
[COMMAND_EXECUTION]: The skill generates implementation tasks that include executable shell commands within a 'Verify' section. These commands are intended to be run via the Bash tool to validate code changes before they are committed.
Evidence: The Task Command template in SKILL.md includes a placeholder for verification commands ({verification command 1}) and specifies Bash as an allowed tool in its frontmatter.
Context: This dynamic execution of commands is a core feature of the skill, providing a programmatic way to ensure task completion through automated testing and build checks.

writing-implementation-tasks