The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes various build, lint, and test tools via the shell, including npm run build, go build, cargo build, tsc, mypy, eslint, golangci-lint, clippy, and various test runners. While these are standard development tools, they execute scripts and code defined within the project being verified.
[INDIRECT_PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because it processes untrusted data and tool outputs without explicit sanitization or boundary markers.
Ingestion points: Ingests data from handoff.md (generated from git diffs), environment detection, and the raw output of build, lint, and test tools.
Boundary markers: None identified. The skill acts on tool outputs directly to perform "Auto-Repairs."
Capability inventory: The skill can execute arbitrary shell commands (via project build scripts), write to the file system (via the auto-repair feature), and create new subagent contexts.
Sanitization: None identified. The logic interprets error messages to perform automated file modifications.
[DYNAMIC_EXECUTION]: The skill uses "Fixable Auto-Repair" logic to dynamically generate and apply code changes (e.g., adding imports, modifying types, fixing lint errors) based on the runtime output of verification tools. It also dynamically determines which shell commands to run based on project type detection.

verification-engine