The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted user input to generate a lean canvas and an associated HTML file, creating a surface for potential injection attacks.
Ingestion points: User-provided context, product names, and research data are ingested into the agent's context as specified in SKILL.md under the 'Collect context and evidence' step.
Boundary markers: No explicit boundary markers or isolation instructions are provided in SKILL.md to distinguish user input from the skill's own instructional logic or the HTML template structure.
Capability inventory: The skill possesses file-writing capabilities, specifically writing to the local file system at a path derived from user input (./lean-canvas-{slug}.html) as described in the 'Visual mode' instructions.
Sanitization: The skill lacks explicit sanitization or escaping mechanisms for user-provided data before it is interpolated into the HTML template placeholders ({{PLACEHOLDER}}).
[DYNAMIC_EXECUTION]: The skill dynamically generates a self-contained HTML file from a template at runtime.
Evidence: SKILL.md Step 5 instructs the agent to read references/html-template.html, replace placeholders with user-derived content, and write the resulting file to disk. This process uses computed paths and local template generation based on user input. While the generated file is intended to be self-contained and does not use external assets, the lack of sanitization during the generation process remains a vulnerability.

foundation-lean-canvas