The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes user-supplied specifications and requirements to generate implementation plans. Maliciously crafted specifications could potentially inject instructions that the agent incorporates into the generated plan files, which are subsequently executed by other skills like superpowers:executing-plans.
Ingestion points: Requirements and specifications provided by the user in the session context, which the skill uses to build the plan header and task steps.
Boundary markers: The skill uses markdown headers and delimiters to structure the plan, but there is no explicit instruction to the agent to treat input from the spec as data rather than instructions or to ignore embedded commands.
Capability inventory: The skill writes implementation plans to the filesystem (docs/superpowers/plans/). It also explicitly triggers the use of downstream execution skills (superpowers:subagent-driven-development, superpowers:executing-plans) which possess significant capabilities including command execution and file manipulation.
Sanitization: No specific filtering, escaping, or validation logic is defined to handle external content extracted from the input specifications before it is interpolated into the generated plan.

writing-plans