The Agent Skills Directory

[COMMAND_EXECUTION]: The skill is centered around executing the gsd CLI as a subprocess to perform complex software development tasks, such as milestone creation and autonomous coding based on user-provided specifications.
[CREDENTIALS_UNSAFE]: The skill implements a mechanism for "answer injection" that explicitly involves handling and injecting sensitive information like API keys and database connection strings into the environment of the gsd child process. While necessary for tool function, this requires the agent to manage sensitive plaintext secrets.
[INDIRECT_PROMPT_INJECTION]: The skill presents a vulnerability surface where malicious instructions could be embedded in the data processed by the agent.
Ingestion points: External software specifications (spec.md), project metadata files in the .gsd/ directory, and structured JSON results from the CLI tool.
Boundary markers: The instructions lack explicit boundary markers or warnings for the agent to ignore potentially malicious instructions embedded within the specifications it processes.
Capability inventory: The agent possesses the capability to execute shell commands, manage git repositories, and write to the local file system based on its interpretation of the ingested data.
Sanitization: No sanitization or validation routines are specified for the content of the specifications before they are passed to the underlying development engine.
[DYNAMIC_EXECUTION]: The skill facilitates the autonomous generation, modification, and execution of application code (e.g., running npm test or make test on generated projects), which represents a dynamic code execution pattern where generated scripts are run in the local environment.

gsd-orchestrator