The Agent Skills Directory

[COMMAND_EXECUTION]: The skill makes extensive use of shell commands like grep, sed, awk, and git to analyze project files and update documentation content. These operations are performed via the agent's subprocess capabilities.
[REMOTE_CODE_EXECUTION]: In several instances (e.g., references/sections-update/usage.md), the skill instructs the agent to extract code examples from the project's README file, save them to temporary locations, and execute them using node or python. This pattern creates a high-risk vector where malicious code embedded in documentation could be executed by the agent.
[EXTERNAL_DOWNLOADS]: The skill uses curl to verify the status of badges from well-known services like Shields.io and GitHub, as well as to check for broken links. While these specific targets are generally safe, the skill allows for the verification of arbitrary URLs found within the documentation.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection as it processes data from external files such as README.md, package.json, and source code. It lacks sanitization or boundary markers when reading these files, which enter the agent's context and could contain malicious instructions designed to manipulate agent behavior.
[CREDENTIALS_UNSAFE]: The skill contains instructions and examples (references/detection/compare-sections.md, references/automation/github-actions.md) that involve scanning for and manipulating environment variables and secrets. Specifically, it suggests inserting environment variables into the README via sed, which could lead to accidental credential exposure.

update-readme