The Agent Skills Directory

[SAFE]: The skill instructions and shell commands are standard for development and analysis workflows. It uses utilities like find, ripgrep, and git to inventory and process local and external files according to its stated purpose.
[EXTERNAL_DOWNLOADS]: The skill facilitates the cloning of external repositories from GitHub using git clone. These downloads are intended for analysis and are scoped to well-known editor projects. This is standard behavior for a code harvesting tool and utilizes a trusted service.
[COMMAND_EXECUTION]: Shell commands are used to automate the identification and classification of tests within target repositories. These commands are executed locally and are restricted to file discovery and metadata extraction, presenting no elevated security risk.
[INDIRECT_PROMPT_INJECTION]: The skill possesses an attack surface for indirect prompt injection because it reads and processes external, untrusted source code. However, the methodology explicitly mandates the extraction of 'behavior invariants' and the writing of fresh local proofs, which significantly mitigates the risk of the agent obeying malicious instructions embedded in the analyzed codebases.

editor-test-harvester