The Agent Skills Directory

[COMMAND_EXECUTION]: The skill deterministically invokes CLI tools like git, gh (GitHub CLI), and curl to manage branches, pull requests, and synchronize state with issue trackers. These operations are restricted to specific, well-known platforms (GitHub and Bitbucket) and are central to the skill's intended purpose.
[INDIRECT_PROMPT_INJECTION]: The skill extracts work items and commands from external roadmap and registry files. To prevent malicious manipulation of these inputs, the selector script (select-next.mjs) performs strict validation, ensuring that document paths are located within the repository and that commands are limited to a safe allowlist.
[DYNAMIC_EXECUTION]: The skill performs dynamic execution in two primary contexts: it uses an embedded Python script in references/read-config.sh for YAML parsing and allows for a user-defined checkCommand in the configuration. Both paths are secured using execution timeouts, process isolation, and path sanitization.
[DATA_EXFILTRATION]: While the skill manages authentication credentials for Jira, GitHub, and Bitbucket, it includes a robust redaction mechanism in references/defer-mutation.js. This module is designed to identify and mask secrets in logs and journals, ensuring that sensitive data is not inadvertently exposed or exfiltrated during automated operations.

develop-next