The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructions provide shell commands, such as curl, jq, npm ls, and govulncheck, which use placeholders for data obtained from external vulnerability scan results. If this data (e.g., a CVE ID or package name) is maliciously crafted by an attacker and not sanitized before being executed, it could result in command injection.
[INDIRECT_PROMPT_INJECTION]: The skill processes data from external security scanners, which represents a surface for indirect prompt injection attacks.
Ingestion points: Vulnerability scan results from external tools like Dependabot, Snyk, and Trivy are ingested during the 'Intake and normalization' phase (SKILL.md).
Boundary markers: Absent. The skill does not instruct the agent to use delimiters or specific safety markers when processing ingested vulnerability data.
Capability inventory: The agent can execute shell commands (curl, npm, pip, mvn, govulncheck) and generates security decisions that influence development workflows.
Sanitization: Absent. There are no instructions for the agent to validate or sanitize external data before it is interpolated into shell commands or used in decision logic.
[EXTERNAL_DOWNLOADS]: The skill fetches vulnerability feeds and scores from www.cisa.gov and api.first.org. These are well-known and trusted organizations, and these network operations are legitimate for the primary purpose of the skill.

cve-triage