The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill instructs the agent to download and install a third-party binary from an external GitHub repository that is not associated with a verified organization.
Evidence: The command go install github.com/Zxilly/go-size-analyzer/cmd/gsa@latest in SKILL.md downloads, compiles, and installs a tool from a third-party user account.
[EXTERNAL_DOWNLOADS]: The skill triggers downloads of external tools and references external container images.
Evidence: Instructions for installing the gsa tool and references to gcr.io/distroless/static-debian12:nonroot in the Dockerfile example. References to well-known registries like gcr.io are generally safe, but the installation of unverified third-party tools is a concern.
[INDIRECT_PROMPT_INJECTION]: The skill reads and processes the content of various project files, creating a risk if those files contain malicious data designed to influence the agent's behavior.
Ingestion points: SKILL.md uses grep to scan configuration files including Makefile, Taskfile.y*ml, .goreleaser.y*ml, Dockerfile, and .github/workflows/.
Boundary markers: There are no instructions or delimiters provided to mitigate the risk of the agent obeying instructions embedded within the scanned files.
Capability inventory: The skill is permitted to run go, gsa, upx, and strip via the Bash tool, providing several paths for executing logic derived from file content.
Sanitization: No sanitization is performed on the data retrieved from project files before it is processed.

go-binary-size