The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes local git and gh CLI commands to detect repository facts. These calls are made using spawnSync and execFileSync with array-based arguments, which prevents shell injection vulnerabilities by avoiding shell interpretation of arguments.
[DATA_EXPOSURE]: The skill probes for the existence of the CLAUDE_CODE_OAUTH_TOKEN secret using the gh secret list command. This operation only retrieves secret names and is explicitly documented to avoid reading or exposing the actual secret values.
[INDIRECT_PROMPT_INJECTION]: The skill reads various project files, such as package.json and pnpm-workspace.yaml, to extract metadata. Although these files represent an external data surface, the skill uses structured parsing methods and provides the agent with specific instructions for handling the data, which minimizes the risk of embedded instructions being interpreted as commands.
[SAFE]: The skill demonstrates secure development practices, including the use of an allowlist and specific checks to prevent prototype pollution in its configuration override logic. It uses no external npm dependencies, relying solely on standard Node.js built-ins.

initialise-skills