The Agent Skills Directory

[COMMAND_EXECUTION]: The skill coordinates various internal tools and sub-skills (e.g., plan, terminus-maximus, browse-qa) using standardized command patterns to automate development phases. These operations are restricted to the local environment and the agent's defined capabilities.
[EXTERNAL_DOWNLOADS]: The documentation references the garrytan/gstack repository as a conceptual blueprint for the workflow design. The skill instructions specifically advise against direct installation or mirroring of the external scaffolding, emphasizing a custom local implementation instead.
[PROMPT_INJECTION]: The skill possesses a surface for indirect prompt injection. 1. Ingestion points: The skill accepts a user-provided goal and a target --url for the browse-qa gear. 2. Boundary markers: No explicit delimiters or instructions to ignore embedded content are provided. 3. Capability inventory: The skill can trigger file writes, git commits, and browser interactions via the sub-skill chain. 4. Sanitization: No sanitization of external URL content or goal input is specified. These risks are mitigated by the skill's multi-stage review structure and human-in-the-loop triggers like git status checks.

build