The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses the eval command in the run_update step to execute the string stored in INSTALL_CMD.
[DYNAMIC_EXECUTION]: The INSTALL_CMD variable is dynamically constructed at runtime by concatenating the PLATFORM variable, which is parsed from a local manifest file (.gsd-install-manifest.json). Executing dynamically generated command strings via eval is a high-risk pattern that can lead to command injection.
[INDIRECT_PROMPT_INJECTION]: The skill processes data from external files (.gsd-install-manifest.json and CHANGELOG.md) that could be modified by other processes or users. The platform field in the manifest is extracted via grep and cut and then used in a shell command. If this field contains shell metacharacters like semicolons or backticks, it could trigger the execution of unintended commands when eval is called.
Ingestion points: Reads .gsd-install-manifest.json in detect_platform and check_current_version steps.
Boundary markers: None present; the parsed values are directly interpolated into shell commands.
Capability inventory: Uses Bash to execute shell commands, eval for dynamic code execution, and npx for package installation.
Sanitization: There is no validation or escaping of the PLATFORM or CURRENT_VERSION values before they are used in shell logic.
[EXTERNAL_DOWNLOADS]: The skill uses npx to download and execute the get-shit-done-multi package from the npm registry. While npm is a well-known service, the use of latest as a version specifier means the exact code being executed is not pinned and can change without notice.

gsd-update