The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process untrusted data from external sources, including GitHub security advisories and user-provided vulnerability report drafts. This creates a potential surface for indirect prompt injection if an attacker-controlled advisory or report contains malicious instructions designed to manipulate the agent.
Ingestion points: Processes external repository advisories via gh api and analyzes user-provided report files and evidence packages using the scripts/check_report.py utility.
Boundary markers: The skill does not employ explicit delimiters or 'ignore' instructions when processing external report data.
Capability inventory: The agent can read and write local files, perform network requests via the GitHub CLI, and execute a local Python scanner script.
Sanitization: The check_report.py script masks secrets in its output to prevent token leakage, but it does not sanitize or filter natural language instructions from processed documents.
[COMMAND_EXECUTION]: The skill executes a local Python script scripts/check_report.py to analyze reports for style and potential security leaks. This is a primary function of the skill and is executed using the -I (isolated) flag to prevent the script's directory from being added to the search path, which is a security best practice.
[EXTERNAL_DOWNLOADS]: The skill uses the gh CLI tool to fetch security advisory data from the GitHub API. This targets a well-known and trusted service (GitHub) for legitimate informational purposes related to the skill's primary function.

vuln-report