The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill instructs the agent to install external software using Go commands (go install github.com/steveyegge/gastown/cmd/gt@latest and go install github.com/steveyegge/beads/cmd/bd@latest). These sources are personal repositories and are not from a recognized organization or official service provider, making the execution of these binaries a high-risk operation.
[COMMAND_EXECUTION]: The skill requires broad shell access (Bash(cmd:*)) to manage the 'gastown' engine. It performs various system operations including installing tools, creating workshop directories (~/gt), and modifying local configuration files like routes.jsonl.
[EXTERNAL_DOWNLOADS]: During the setup flow, the skill logic encourages the ingestion of third-party projects via GitHub URLs provided by the user. This involves fetching and potentially executing content from external sources.
[PROMPT_INJECTION]: The skill creates a surface for indirect prompt injection by processing content from external repositories.
Ingestion points: External GitHub repositories provided by the user during rig configuration (references/examples.md).
Boundary markers: None identified in the instructions to distinguish between trusted system commands and untrusted project data.
Capability inventory: The skill has access to Bash, Read, Write, Edit, and WebFetch tools, which could be exploited if malicious instructions are encountered in a project repo.
Sanitization: No validation, sanitization, or filtering steps are defined for external project content before the agent acts upon it.

gastown