The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data from multiple sources including issue-tracker.md, AGENTS.md, CLAUDE.md, and individual ticket/spec files. \n- Ingestion points: SKILL.md directs the agent to read repository guidance, ADRs, glossary, specs, and all associated tickets. \n- Boundary markers: No explicit delimiters or instructions are provided to the agent to treat content within these files as data rather than instructions. \n- Capability inventory: The agent has the ability to manage git worktrees, branches, and commits, write to the filesystem, interact with PR tools, and execute arbitrary validation and generation commands. \n- Sanitization: There are no mentions of sanitizing or validating the contents of the tickets or specs before processing or executing associated commands.
[COMMAND_EXECUTION]: The workflow involves executing validation commands and generation commands that are defined within the project's specification or tickets. This creates a risk if the source material is modified by an untrusted party to include malicious shell commands.
[DYNAMIC_EXECUTION]: The skill uses a Skill tool to dynamically load and execute other instructions (other skills) at runtime based on the needs of the task, and generates code via commands specified in external ticket data.

implement-spec