The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill directs the agent to install the 'desloppify' health scanner from an untrusted third-party GitHub repository (github.com/peteromallet/desloppify) which is not among the verified or trusted organizations.\n- [REMOTE_CODE_EXECUTION]: The prerequisite check encourages the direct execution of remote code from the git repository using the 'uvx' tool ('uvx --from git+https://github.com/peteromallet/desloppify.git desloppify'), posing a high risk of unverified code execution.\n- [INDIRECT_PROMPT_INJECTION]: The skill instructs the agent to blindly follow 'INSTRUCTIONS FOR AGENTS' provided in the tool's output rather than using its own reasoning ('Follow the scan output's INSTRUCTIONS FOR AGENTS — don't substitute your own analysis'). This creates a significant vector for third-party control if the tool or its output is compromised.\n
Ingestion points: The agent reads its primary workflow tasks and specific refactoring instructions from the outputs of 'desloppify scan' and 'desloppify next'.\n
Boundary markers: There are no mentioned delimiters or safety prompts to prevent the agent from executing malicious instructions embedded in the tool's scan results.\n
Capability inventory: The agent is empowered to perform large-scale code refactors (modifying multiple files), execute 'git' commands (branching, committing, pushing), and use the GitHub CLI ('gh pr create').\n
Sanitization: The instructions explicitly discourage the agent from interpreting or filtering the tool's instructions, effectively bypassing any sanitization logic the agent might otherwise apply.\n- [COMMAND_EXECUTION]: The core functionality of the skill involves the frequent execution of CLI commands, including the 'desloppify' tool itself, alongside 'git' and 'gh' utilities to manage the project state and health records.

desloppify