The Agent Skills Directory

[COMMAND_EXECUTION] (MEDIUM): Python code injection vulnerability in scripts/accessibility-check.sh.
Evidence: The script executes a Python snippet via python3 -c that uses shell string interpolation: json.load(open('${PAGE_JSON}')).
Context: The variable ${PAGE_JSON} is constructed using the user-supplied --output-dir argument. A maliciously crafted directory name (e.g., containing single quotes and Python commands) can break out of the open() call and execute arbitrary Python code.
[EXTERNAL_DOWNLOADS] (LOW): Usage of npx for dynamic package execution.
Evidence: The scripts frequently invoke npx @axe-core/cli, npx playwright, npx eslint, and npx tsc.
Context: npx downloads and executes code from the npm registry at runtime if the package is not already cached or installed. While used for legitimate dev tools, it introduces a dependency on external, mutable code sources.
[COMMAND_EXECUTION] (LOW): Use of eval in the orchestration script.
Evidence: In scripts/run-all-checks.sh, the run_check function uses eval "$cmd" to execute the check logic.
Context: While the commands passed to eval are currently hardcoded within the script, the use of eval is a dangerous practice that can be exploited if the script logic is modified to include untrusted inputs.

pre-merge-checklist