The Agent Skills Directory

[PROMPT_INJECTION]: The skill functions by ingesting design specifications and converting them into executable tasks, creating a surface for indirect prompt injection. A malicious design could influence the agent to generate plans with harmful code or commands.
Ingestion points: External design or architecture documents provided by the user (SKILL.md).
Boundary markers: None provided to isolate untrusted input from the plan generation logic.
Capability inventory: The skill writes implementation plans to the filesystem (docs/plans/) and triggers subagents to perform code changes and shell operations (SKILL.md).
Sanitization: No sanitization or validation of input designs is described before they are processed into task structures.
[COMMAND_EXECUTION]: The skill's templates and best practices documentation include shell commands for testing, version control, and environment setup.
Evidence: Task templates in references/plan-structure-templates.md utilize pytest, git, and bash commands.
[EXTERNAL_DOWNLOADS]: Documentation within the skill provides examples for managing project dependencies using external package registries.
Evidence: Examples in references/best-practices.md demonstrate the use of pip install for project requirements.

writing-plans