The Agent Skills Directory

[COMMAND_EXECUTION]: The skill provides numerous shell commands for the agent to execute to verify and audit website content. Evidence includes multiple curl commands piped to text processing utilities like sed, grep, and wc in SKILL.md (e.g., curl -sf https://example.com/ | sed -e 's/<[^>]*>//g' | tr -s ' \n' | wc -c). The agent is also instructed to use the isagentready CLI tool for rescanning and result verification (e.g., isagentready results <domain> --checkpoint <stable-id>).
[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from external websites, creating a surface for indirect prompt injection attacks. 1. Ingestion points: The agent is instructed to fetch content from user-provided URLs using curl commands defined in SKILL.md across various fix workflows (e.g., sections 3.1, 3.3, 3.6). 2. Boundary markers: There are no explicit instructions or delimiters provided to the agent to treat the fetched HTML content as untrusted data or to ignore any instructions found within that content. 3. Capability inventory: The agent has capabilities to execute shell commands, perform network operations via curl, and modify the local project's source files to implement suggested fixes. 4. Sanitization: The skill does not implement or suggest any sanitization, filtering, or validation of the external HTML content before it is processed by the agent or piped into shell utilities.

content-semantics