The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill facilitates the installation of external code through the specify bundle install command. These bundles can contain extensions and commands that are scaffolded into the environment as agent skills, introducing new executable logic.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process external bundle manifests and components from various sources, which presents a surface for indirect instructions.
Ingestion points: External data enters the context via bundle install (using catalog IDs, local paths, or ZIP files) and catalog add <url> commands.
Boundary markers: The instructions do not define specific delimiters or guardrails to distinguish between bundle metadata and potentially malicious embedded instructions.
Capability inventory: The skill has the capability to write files to the .github/skills/ directory and manage external software catalogs.
Sanitization: There are no explicit sanitization or validation steps described within the skill instructions for the content of the bundles being installed.
[COMMAND_EXECUTION]: The skill relies on the execution of shell commands via the specify CLI. This includes potentially sensitive operations such as adding network-based catalogs (catalog add <url>) and modifying the local environment through bundle installation and removal.

speckit-bundle