The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill reads code and documentation (e.g., CONTRIBUTING.md, AGENTS.md) from the repository being reviewed to identify bugs and test commands. This content is untrusted and could contain hidden instructions intended to manipulate the agent. The agent has write-access capabilities including git commit and posting review comments to GitHub or GitLab via thurview forge submit. There are no explicit boundary markers or sanitization steps mentioned for the ingested repository data.
[DYNAMIC_EXECUTION]: The skill dynamically identifies and executes test and lint commands found in the repository's configuration files (e.g., Makefile, package manifests). This involves running arbitrary scripts defined within the project being reviewed, which could lead to unauthorized code execution if the repository is malicious.
[EXTERNAL_DOWNLOADS]: The instructions include running npx -y thurview, which downloads and executes the thurview package from the NPM registry. This tool is provided by the skill's author to manage forge interactions and status reporting.
[PROMPT_INJECTION]: User-supplied $ARGUMENTS are directly included in the skill's prompt. This could be exploited to provide instructions that attempt to bypass the skill's safety constraints, such as the directive to never push changes without explicit permission.

thurview-fix