The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill's workflow in SKILL.md involves downloading external source code using git clone --depth 1 <repo-url> into the /tmp directory for analysis.
[REMOTE_CODE_EXECUTION]: The instructions explicitly direct the agent to "run code" to verify technical claims. This constitutes a remote code execution surface as the code originates from external repositories that are not verified for safety before execution.
[COMMAND_EXECUTION]: The skill employs shell commands like git clone to interact with the file system and network for its research tasks.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection (Category 8) because it ingests untrusted data from the web and GitHub repositories without sufficient safety constraints.
Ingestion points: WebFetch tool and git clone operations mentioned in SKILL.md.
Boundary markers: The skill lacks explicit instructions to treat fetched content as data only or to ignore embedded instructions.
Capability inventory: The agent is granted the capability to execute shell commands and run code based on the researched content.
Sanitization: No sanitization or validation of the fetched code or documentation is specified before the agent interacts with or executes it.

technical-research