The Agent Skills Directory

[PROMPT_INJECTION]: Indirect Prompt Injection vulnerability surface. The script update_skill_index.py reads 'name' and 'description' metadata from the frontmatter of SKILL.md files found in searched directories and interpolates them into rule files like cline_overview.md or .clinerules/agents_skills.md. Since these files provide instructions to the AI agent, an attacker who can place a malicious skill file in a scanned directory could inject commands into the agent's rules.
Ingestion points: The script scans multiple directories including ~/Documents/Cline/skills/, ~/.kiro/skills/, and .claude/skills/.
Boundary markers: No delimiters or explicit instructions are used in the generated indexes to prevent the agent from obeying instructions embedded in the skill descriptions.
Capability inventory: The script has file system read/write access and performs directory traversal within user-accessible paths.
Sanitization: No sanitization or escaping of the extracted 'description' text is performed before it is written to the instruction files.

skills-index-updater