The Agent Skills Directory

[DYNAMIC_EXECUTION]: The skill frequently employs eval and source <(command) patterns to execute shell instructions generated by local binaries at runtime. Examples include setting project slugs via gstack-slug and configuring search paths through gstack-paths, representing the execution of dynamically generated code.
[COMMAND_EXECUTION]: The skill performs extensive shell operations for workflow orchestration. This includes managing git worktrees, creating and committing changes to files like CLAUDE.md, and spawning background processes. It also uses the open command to launch external URLs for documentation.
[INDIRECT_PROMPT_INJECTION]: By ingesting codebase content and user intent into generated specifications, the skill presents a surface for indirect prompt injection. It mitigates this risk by using structured boundary markers (e.g., <<<USER_SPEC>>>) and an instruction boundary in its calls to external models, as well as a multi-stage review process.
[DATA_EXFILTRATION]: The skill transmits potentially sensitive technical data to external services. This includes filing issues via the GitHub CLI and sending drafts to OpenAI Codex for quality scoring. To protect sensitive information, it includes a mandatory redaction step (gstack-redact) that scans for credentials and PII before dispatching data to these external sinks. Additionally, it features an opt-in telemetry system for usage statistics.

spec