The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from design.md to create structured implementation plans, creating a potential surface for indirect prompt injection. 1. Ingestion points: The skill reads design.md from the repository to extract requirements and architecture. 2. Boundary markers: There are no explicit delimiters or boundary markers specified to separate the untrusted design content from the agent's own instructions. 3. Capability inventory: The skill has the capability to read repository files and write the plan.json artifact; it does not have access to shell, network, or subprocess tools. 4. Sanitization: The skill does not perform sanitization, filtering, or validation on the content ingested from design.md before including it in the resulting plan.

spec-plan