The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes project-level markdown files, such as tasks.md, proposal.md, design.md, and specs/*.md, as its primary source of instructions for long-running tasks. This introduces a risk where malicious instructions embedded in these files could influence agent behavior. The ingestion occurs during initialization and at every task checkpoint. No explicit boundary markers or content sanitization are required by the instructions.
[COMMAND_EXECUTION]: The skill mandates the execution of project-specific shell commands for validation, testing, linting, and type-checking as a prerequisite for task completion. It also uses git commands (git status, git diff) and ripgrep (rg) to scan the workspace. The file migration logic involves reading, writing, and deleting files (rm) to enforce standard locations for progress and findings logs.

do-long-task