The Agent Skills Directory

[PROMPT_INJECTION]: The skill ingests untrusted data from external URLs and local files during accessibility audits, creating a surface for indirect prompt injection. The skill mitigates this risk through strong grounding instructions that require the agent to follow specific tool-provided fix directives and mandates user confirmation before applying more than ten fixes or editing files outside the immediate target.\n
Ingestion points: External URLs accessed via audit_live and local files or HTML strings processed by audit_html.\n
Boundary markers: Instructions specify grounding in tool output, a prohibition on inventing content, and manual confirmation for large scopes.\n
Capability inventory: The skill utilizes Edit, Write, and Bash tools, providing the agent with file modification and shell execution permissions.\n
Sanitization: No explicit sanitization or filtering of external content is described.\n- [COMMAND_EXECUTION]: The skill allows the use of the Bash tool, which permits the agent to execute shell commands. While intended for remediation tasks, this grants the agent powerful system-level capabilities.

accessibility-fix