The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes various system and project-local tools such as grep, git, tsc, go, and cargo to analyze code and run tests. It prioritizes project-local binaries to ensure correct tool versioning.
[DYNAMIC_EXECUTION]: In its highest verification tier, the skill synthesizes and executes minimal reproduction scripts to verify behavioral claims. This execution is performed within a throwaway git worktree to prevent persistent side effects or corruption of the working directory.
[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted behavioral claims and source code as data for processing. Ingestion points: claim argument and target files. Boundary markers: [receipt] and [gate] blocks. Capability inventory: Command execution via subprocess for tools like grep, tsc, go test, and pytest. Sanitization: Use of git worktree for isolation, default denial of network access, and mandatory cleanup via shell traps.

verify-behavior