The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill requires the agent to run uvx disambiguate <term>, which downloads and executes code from a remote Python package registry at runtime. Since the package is not pinned to a specific version or hash, it allows for the execution of remote code that is outside the skill's direct control and could be modified at the source.
[COMMAND_EXECUTION]: The skill instructs the agent to execute shell commands (uvx) to process terms from the project's glossary during the specification generation process.
[INDIRECT_PROMPT_INJECTION]:
Ingestion points: The skill ingests untrusted data from the "current conversation" and "codebase understanding" as the primary source for the specifications it generates (SKILL.md).
Boundary markers: The instructions lack delimiters or specific warnings to prevent the agent from accidentally executing instructions hidden within the conversation history.
Capability inventory: The skill has the capability to explore the repository, execute CLI tools, and publish documents to a project tracker (AGENTS.md).
Sanitization: There are no documented steps for sanitizing or filtering input conversation data before it is incorporated into the load-bearing specification documents.
[EXTERNAL_DOWNLOADS]: The skill initiates external downloads of the disambiguate package from a remote registry during its operation.

to-spec