The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructs the agent and its subagents to execute several shell commands, including git worktree for isolation, gh pr create for pull request management, and dynamically discovered test commands such as npm test, bun test, pytest, or go test (SKILL.md, Phase 2 and Phase 3).
[EXTERNAL_DOWNLOADS]: The README.md provides installation instructions using npx skills add vltansky/skills/skills/batch, which fetches the skill's code from a remote source controlled by the vendor.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because it derives execution logic from untrusted data within the codebase being modified.
Ingestion points: The agent researches the codebase to find file patterns, conventions, and test recipes in files like package.json or Makefile (SKILL.md, Phase 1).
Boundary markers: Absent; there are no instructions to use delimiters or ignore instructions embedded within the discovered project files.
Capability inventory: The skill can spawn background subagents, execute arbitrary shell commands for testing, and perform network operations via the gh CLI (SKILL.md, Phase 2).
Sanitization: Absent; the skill lacks mechanisms to validate or sanitize the commands it extracts from the project files before executing them.

batch