The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes multiple shell commands via git to detect repository state and perform initialization tasks.
Evidence: Commands such as git rev-parse, git log, git branch, git init, git checkout, and git worktree add are used in SKILL.md to manage the repository structure and isolation environments.
[INDIRECT_PROMPT_INJECTION]: The skill automatically generates CLAUDE.md and PRD.md by reading existing project files, which could contain malicious instructions designed to influence the agent's initialization logic.
Ingestion points: Reads manifest files (package.json, pom.xml, etc.), README.md, and source code directories to infer stack and conventions in SKILL.md (Fase 0 and Fase 2.1).
Capability inventory: The skill has the ability to write files, create directories, and execute git branch/worktree operations.
Sanitization: There is no explicit sanitization logic described for the content extracted from existing files before it is interpolated into the new CLAUDE.md instructions.
Boundary markers: The skill includes 'Gate 1' and 'Gate 2' which require the user to explicitly confirm the detection plan and review the generated files before proceeding.

fmway-init