The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest external requirements and specifications to generate actionable plans. This architecture creates a potential surface for indirect prompt injection if the input requirements contain malicious instructions that are propagated into the generated plan. However, this is inherent to the skill's primary purpose as a planning assistant. 1. Ingestion points: Untrusted feature specifications or requirement documents provided as input to the skill. 2. Boundary markers: None explicitly defined in the prompt instructions to delimit untrusted content. 3. Capability inventory: The skill generates plans containing Python code blocks and shell commands (git, pytest) intended for execution by subsequent subagents or parallel sessions. 4. Sanitization: No explicit sanitization or validation of the input requirements is performed before they are processed and incorporated into the output plan document.

writing-plans