The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill operates on Go source code files which serve as untrusted ingestion points. An attacker who can influence the repository's code could embed malicious instructions in comments or string literals to influence the agent's behavior during a refactoring pass.
Ingestion points: The skill reads all files matching **/*.go (specified in SKILL.md).
Boundary markers: The skill explicitly requires human sign-off via the environment's question tool for all planning stages and irreversible decisions, preventing autonomous obedience to embedded instructions.
Capability inventory: The agent has permissions to Write and Edit files, and execute Bash commands for go, git, gh, and gopls (specified in SKILL.md).
Sanitization: The skill prioritizes mechanical transformations using gopls and other AST-aware tools over manual edits, which limits the surface for simple textual injection.
[EXTERNAL_DOWNLOADS]: The skill fetches and installs several Go development tools from official and well-known sources during its setup process.
Evidence: Installs golang.org/x/tools/gopls and golang.org/x/perf/cmd/benchstat (official Google/Go repositories). It also references github.com/uber-go/gopatch and github.com/dave/dst, which are well-known tools within the Go ecosystem.

golang-refactoring