The Agent Skills Directory

[COMMAND_EXECUTION]: The script scripts/desloppify_portable.py uses os.execvpe to launch the external desloppify CLI tool with arguments provided at runtime. It also utilizes subprocess.run to execute local git commands for repository state management (e.g., git rev-parse, git check-ignore, and git ls-files). This allows the agent to execute arbitrary sub-commands within the scope of the installed tools.
[EXTERNAL_DOWNLOADS]: The documentation in SKILL.md and references/upstream-and-safety.md instructs users to download and install the core engine from an external GitHub repository (github.com/peteromallet/desloppify) that is not identified as a trusted source. This introduces a dependency on unverified third-party code into the execution environment.
[PROMPT_INJECTION]: The skill is designed to ingest and analyze entire codebases, including source code and comments, which are considered untrusted data. This creates an indirect prompt injection surface where malicious instructions embedded in the analyzed code could potentially influence the agent's behavior during the review or repair process. The skill acknowledges this risk in its guardrails, advising the agent to treat comments as untrusted data.

desloppify