The Agent Skills Directory

Command Execution: The skill interacts with the GitHub REST API by executing the gh CLI. Evidence is found in scripts/Get-CodeScanningAlerts.ps1 and scripts/get-code-scanning-alerts.sh. Input parameters like repository owners and names are strictly validated against restrictive regular expressions before execution to prevent command injection.
External Tool Dependencies: To function, the skill requires external tools (gh and jq) to be available on the system path. This requirement is clearly documented in SKILL.md as a prerequisite. These are official or industry-standard utilities for GitHub integration and data processing.
Processing of External Data: The skill retrieves alert descriptions and messages produced by external analysis tools (such as CodeQL). This is a potential surface for indirect prompt injection if an attacker controls the analysis results in a repository. The skill ingests this data through the gh api tool and includes it in the results provided to the agent. While the script sanitizes command arguments, the alert content itself is treated as data for the agent to review. The skill's capabilities are limited to read-only operations, which naturally limits the impact of such injection.

gh-code-scanning