The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill implements a workflow where implementation plans (e.g., docs/plans/feature-plan.md) are read and their contents are interpolated directly into the context for subagents using the delegate_task tool.
Ingestion points: The skill reads external plan files from the docs/plans/ directory to define tasks for the subagents.
Boundary markers: There are no explicit delimiters or instructions provided to the subagents to ignore or isolate potentially malicious instructions that might be embedded within the implementation plans.
Capability inventory: Subagents are granted terminal and file toolsets, allowing them to execute system commands and modify the filesystem based on the plan's instructions.
Sanitization: The plan content is used directly as context without validation or filtering to remove potential prompt injection vectors.
[EXTERNAL_DOWNLOADS]: The skill references standard development packages such as bcrypt and pytest. It also references documentation from the gsd-build/get-shit-done repository on GitHub for its methodology guidance. These references are to well-known services and standard tools in software development.

subagent-driven-development