The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill's core functionality relies on processing external user data, which creates a potential vector for indirect prompt injection attacks where malicious instructions could be embedded in input materials.
Ingestion points: The skill reads and reflects on 'initial notes', 'existing sketches', and 'prototype feedback' as part of its 'Interview and Synthesis' process defined in SKILL.md.
Boundary markers: The instructions direct the agent to 'Separate observations, assumptions, proposals, and decisions' and 'Reflect the apparent problem', which provides some logical separation, but there are no explicit delimiters or instructions to ignore embedded commands within the ingested data.
Capability inventory: The skill primarily generates markdown text outputs (pitches and shape maps). It mentions potential use of drawing tools like tldraw if available, but it does not have capabilities for sensitive file access, shell command execution, or network exfiltration.
Sanitization: The instructions do not define any sanitization, validation, or escaping logic for the content provided in user notes or sketches before it is processed by the agent.

map-shape