The Agent Skills Directory

[DYNAMIC_CONTEXT_INJECTION]: The skill uses the !command`` syntax in SKILL.md to execute shell commands automatically when the skill is loaded by the agent.
Evidence: !curl -sf https://www.cisa.gov/sites/default/files/feeds/known_exploited_vulnerabilities.json | python3 -c ... ``
Context: This is used to display the current version and entry count of the CISA KEV catalog in the agent's context.
[COMMAND_EXECUTION]: The skill executes shell commands involving curl and python3 during its initialization phase.
Evidence: A piped command sequence fetches a JSON file and parses it using a Python one-liner (python3 -c).
[EXTERNAL_DOWNLOADS]: The skill automatically performs a network request to an external domain upon loading.
Evidence: Fetches the Known Exploited Vulnerabilities catalog from www.cisa.gov.
Note: The download originates from a well-known government organization.
[PROMPT_INJECTION]: The skill possesses a surface for indirect prompt injection because its primary function involves processing untrusted data from external security scanners.
Ingestion points: The skill is designed to ingest vulnerability scan reports from tools like Qualys, Tenable, Rapid7, and Snyk.
Boundary markers: The skill includes a 'Prompt Injection Safety Notice' that explicitly tells the agent to disregard instructions embedded in the scan output or external advisories.
Capability inventory: The agent is granted access to tools such as WebFetch, Read, Grep, and Glob.
Sanitization: The skill relies on natural language instructions for the agent to maintain safety rather than programmatic sanitization or strict schema validation of the input data.

cve-triage