The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill is configured to retrieve the typoshift utility from the author's own GitHub repository using standard package runners like uvx or pipx if the tool is not already installed on the system.\n- [REMOTE_CODE_EXECUTION]: The instructions facilitate the download and execution of the reconstruction tool from a remote Git source (git+https://github.com/g4lb/typoshift-skill) when the local environment lacks the necessary binaries.\n- [COMMAND_EXECUTION]: The skill executes shell commands to interact with the reconstruction utility, passing data via standard input and processing the resulting JSON output.\n- [INDIRECT_PROMPT_INJECTION]:\n
Ingestion points: User input provided through the $ARGUMENTS parameter in SKILL.md is processed for layout reconstruction, potentially generating new instructions.\n
Boundary markers: The skill uses a quoted shell heredoc (<<'TYPOSHIFT_EOF') in SKILL.md to safely pass input to the tool, which prevents shell interpolation of the raw input during the command phase.\n
Capability inventory: The instructions in SKILL.md direct the agent to follow high-confidence reconstructions as valid user requests, which could trigger any capability or tool available to the agent.\n
Sanitization: The skill includes safety logic in SKILL.md requiring manual confirmation for reconstructions involving known destructive keywords (e.g., delete, force-push) or those with low confidence scores.

typoshift