The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: Fetches standard Go development tools such as golangci-lint from its official repository and govulncheck from golang.org/x. These are well-known, community-standard resources for the Go ecosystem.
[COMMAND_EXECUTION]: The skill instructs the agent to execute standard Go toolchain commands, including go test, go build, go vet, and gofmt, to perform code validation, testing, and performance measurement.
[DATA_EXFILTRATION]: Performs network operations to collect profiling data from remote endpoints (e.g., http://host:6060/debug/pprof/profile). While these are standard diagnostic practices, they involve sending requests to non-whitelisted domains and could potentially expose process metadata.
[INDIRECT_PROMPT_INJECTION]: The skill ingests and processes Go source files and evaluation test cases (evals/files/). It possesses the capability to execute shell commands and write profiling output to the file system based on this input. Although no explicit sanitization or boundary markers are present, the threat surface is limited to established development tooling.
Ingestion points: Go source files, go.mod, and evaluation files located in evals/files/.
Boundary markers: None identified.
Capability inventory: Shell command execution (go, gofmt, golangci-lint), file system writes (for profiling and benchmark outputs), and network access (targeting pprof endpoints).
Sanitization: None identified.

go