The Agent Skills Directory

[Indirect Prompt Injection] (LOW): The skill reads untrusted data from project files to calculate progress, which could lead to instruction overrides if those files contain malicious content. 1. Ingestion points: 02-projects/{ID}-{name}/01-planning/tasks.md and overview.md. 2. Boundary markers: Absent; the skill lacks delimiters or instructions to ignore embedded commands. 3. Capability inventory: The skill can move directories and write to project-map.md and overview.md. 4. Sanitization: There is no explicit validation of the project name (potential path traversal) or the content of the markdown files being read.

archive-project