The Agent Skills Directory

[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection due to its architecture of ingesting and processing data from multiple external sources without defined boundary markers or sanitization.
Ingestion points: The advisor role reads transcript.md, audit reports (.audit/*.md), session records (.records/*.md), office-hours notes (office-hours.md), and learner profiles (learner.md). It also utilizes the existing roadmap.html as a seed for regeneration.
Boundary markers: There are no instructions to use specific delimiters or to ignore embedded instructions when reading these input files.
Capability inventory: The skill can write to various course files, including roadmap.md, unit.md, and roadmap.html, and can invoke subagents for research and interviewing.
Sanitization: The skill does not describe any methods for escaping or validating content before it is processed or written back to the course files.
[PROMPT_INJECTION]: The skill implements a "seed rule" for the roadmap.html file, which is designed to preserve manual modifications to the HTML across regenerations. While intended for design flexibility, this creates a persistence mechanism where injected or malicious content in the HTML file will be carried forward by the agent into all future versions of the roadmap render.

academic-advisor