The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes the glw CLI tool or bun run src/index.ts to interact with GitLab. These executions are legitimate, strictly mapped to user intent, and use a canonical set of commands and flags.
[DATA_EXFILTRATION]: The skill interacts with GitLab's GraphQL API. It manages sensitive information like GitLab tokens via environment variables or a .env file, which is a standard and recommended practice for CLI tools. No unauthorized data transmission to external or untrusted domains was detected.
[PROMPT_INJECTION]: The instructions focus on correctly translating user requests into CLI commands. Technical warnings regarding non-interactive environments (non-TTY) ensure the agent uses appropriate confirmation flags (--yes) only after a mandatory preview step, preventing accidental or malicious bulk mutations.
[DATA_EXPOSURE]: The skill accesses the local file system to read and write configuration files (glw.config.json, .env) and issue content files. These operations are appropriately scoped to the tool's functions and do not target sensitive system directories (e.g., .ssh or .aws).
[SAFE]: The skill implements a robust "Bulk operation protocol" requiring a preview (--dry-run) before execution. It also mandates using --json for parsing CLI output to ensure the agent reasons over structured data rather than human-readable text, which effectively mitigates risks from malformed or malicious data in issue content.

glw