The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill acts as a pipeline that processes user-provided requirements and research data to build products. This creates an inherent surface for indirect instructions, although the workflow is designed to mitigate this through manual gates.
Ingestion points: User feedback on research, project proposals, and interactive UI prototypes as described in SKILL.md.
Boundary markers: The skill explicitly implements 'Human Locks' that stop the agent's turn and wait for a user response before proceeding, preventing autonomous execution chains.
Capability inventory: The flow skill itself does not perform file writes or network operations; it manages transitions between other task-specific skills.
Sanitization: Input validation is performed socially through the 'Human Lock' mechanism rather than programmatic sanitization.
[NO_CODE]: The skill consists entirely of markdown-based logic and YAML configuration. It does not contain any executable scripts, binaries, or direct shell command invocations, significantly limiting the risk of traditional code-based exploits.
[SAFE]: The skill demonstrates several security best practices, including 'Hard Gates' that prevent unauthorized repository initialization or business code generation until project requirements are stabilized and accepted by the user. It also includes logic to decline the pipeline for out-of-scope requests like simple bug fixes or regex queries.

fancy-flow