The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests external data in the form of git conflict markers and commit messages. Because this data could theoretically contain adversarial instructions, the skill implements a multi-stage defense: Stage 4 performs a mandatory audit to ensure no behavior was invented and that edits remained within the conflicted files, followed by the execution of the project's own validation checks (tests and typechecks).
[COMMAND_EXECUTION]: The skill makes extensive use of local shell commands such as git and gh for branch management and metadata retrieval. It also identifies and executes project-specific build tools like npm install, pnpm install, or cargo to regenerate lockfiles. These actions are standard for the tool's purpose of resolving integration conflicts.
[EXTERNAL_DOWNLOADS]: The skill performs network operations to fetch branch content and pull request metadata from GitHub. These operations use established tools and target well-known services to obtain the necessary context for mending the repository state.

mend