The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill includes an installation command curl https://mise.run | sh in the Error Recovery section. This pattern downloads a script from a remote server and pipes it directly into the shell for execution without prior verification.
[EXTERNAL_DOWNLOADS]: Remote scripts are retrieved from the external domain https://mise.run for system-level installation.
[COMMAND_EXECUTION]: The skill uses the Bash tool to run local task runners (e.g., mise run release:full). These tasks are defined within the repository's filesystem and can execute arbitrary shell logic.
[CREDENTIALS_UNSAFE]: The skill's documentation explicitly mentions the use of GH_TOKEN and GH_ACCOUNT stored in .mise.toml files for authentication purposes during the release process.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection through the repository it operates on.
Ingestion points: The agent reads repository-specific task lists via mise tasks ls and processes contents of files within .mise/tasks/.
Boundary markers: No delimiters or instructions are provided to the agent to ignore potentially malicious instructions embedded in the repository's task definitions or versioning metadata.
Capability inventory: The skill has access to Bash, Read, Edit, and TodoWrite tools, allowing it to execute commands and modify files.
Sanitization: There is no evidence of sanitization or validation of the task names or scripts before they are executed via the Bash tool.

release