The Agent Skills Directory

[OBFUSCATION]: A homoglyph was detected in the references/style-presets.md file. The CSS variable --bg-elevated within the matt-gray preset contains a Cyrillic character 'Е' (U+0415) instead of a Latin 'E' in the hexadecimal color code #EBEBЕ6. This mixing of scripts within a single word is a pattern used to evade plain-text scanning.
[METADATA_POISONING]: There is a major inconsistency between the skill's documentation and its actual implementation. The README.md and evals/evals.json refer to the use of Chart.js v4, whereas the SKILL.md and scripts/screenshot-chart.mjs implement the logic using Apache ECharts v6. Such discrepancies can be deceptive regarding the skill's actual dependencies and behavior.
[COMMAND_EXECUTION]: The skill instructs the agent to execute local shell scripts (scripts/export-chart.sh) and Node.js scripts (scripts/screenshot-chart.mjs) to generate HTML and capture screenshots. These scripts have broad access to the local environment.
[EXTERNAL_DOWNLOADS]: The export-chart.sh script automatically installs the playwright package and downloads the Chromium browser binary (~200MB) from external sources during its first execution. While these are well-known tools, performing installations at runtime is a high-privilege action.
[DYNAMIC_EXECUTION]: The export-chart.sh script dynamically creates a package.json file in a temporary directory and installs dependencies at runtime before executing a Node.js script to capture the chart screenshot.
[INDIRECT_PROMPT_INJECTION]: The skill processes user-supplied data (JSON or CSV) and interpolates it directly into HTML templates without explicit sanitization or boundary markers. This content is then rendered by a headless browser, creating a surface where malicious instructions embedded in the data could influence the agent's behavior during the rendering process.

graphic-chart