The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill instructs the agent to install the deslop utility from an unverified source using multiple methods, including cargo binstall deslop, cargo install deslop, and mise use -g github:Nimblesite/deslop@latest. It also explicitly points to downloading pre-compiled binaries from https://github.com/Nimblesite/Deslop/releases/latest. These sources are not included in the trusted vendor or well-known service lists.
[COMMAND_EXECUTION]: The protocol involves extensive use of shell commands to manage the environment and execute code analysis. This includes package managers (cargo, mise, dart pub), version control (git), and language toolchains (dart, flutter). Many of these commands are executed with parameters derived from the local environment or repository structure.
[INDIRECT_PROMPT_INJECTION]: The skill has a significant attack surface for indirect prompt injection as it ingests and processes code from arbitrary Dart and Flutter repositories.
Ingestion points: The deslop tool scans the {target_dir} provided by the user or identified in the workspace.
Boundary markers: None. The skill does not implement delimiters or specific instructions to ignore embedded malicious patterns within the source code it audits.
Capability inventory: The skill possesses powerful capabilities including file system writes (refactoring), binary execution (deslop, cargo), and Git state manipulation.
Sanitization: There is no evidence of sanitization for the data read from the repository before it influences the agent's refactoring logic.

deslop-duplication-audit