The Agent Skills Directory

[COMMAND_EXECUTION]: The skill performs extensive shell command execution to manage repository state and branch lifecycles. It uses git for fetching, switching, and committing changes, and runs a localized setup script at ~/.claude/scripts/ensure-gitignore.sh.
[PROMPT_INJECTION]: The skill exhibits an indirect prompt injection vulnerability surface. It processes implementation plans from potentially untrusted files (.plans/*.md, PLAN.md) and interpolates verbatim task text into the prompts of subagents equipped with high-privilege tools such as Bash and file-writing capabilities.
Ingestion points: Implementation plan files located in .plans/, the project root, or legacy documentation paths.
Boundary markers: The implementer prompt template uses Markdown section headers (e.g., # Goal, # Your task) as structural delimiters.
Capability inventory: The parent skill executes git and local shell commands; subagents are granted Edit, Write, and Bash access.
Sanitization: There is no evidence of filtering or sanitization applied to the plan content before interpolation into subagent instructions.

implement-plan