The Agent Skills Directory

[COMMAND_EXECUTION]: The skill employs several Python scripts to manage the lifecycle of a development run, including Git worktree creation, state management, and ticket indexing. These scripts use the subprocess module to execute standard local commands (e.g., git worktree, git commit, git rev-parse, and python3). All subprocess calls observed utilize list-based arguments rather than shell strings, which is a secure practice to prevent command injection.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process Markdown tickets from a user-provided directory. While processing external text files is an ingestion point for instructions, the skill implements a 'Senior Guide / Junior Implementer' architecture where a senior model derives a plan from repository evidence before implementation. Furthermore, the scripts/corroborate.py script automatically verifies completion claims against Git history, mitigating the risk of the agent being misled by falsified status reports in untrusted ticket data.
[SAFE]: The Python utility scripts (index_tickets.py, run_state.py, reconcile_run.py, etc.) rely exclusively on the Python standard library (e.g., argparse, pathlib, json, subprocess). No dependencies on third-party packages or remote script downloads were found.
[SAFE]: The installation process (scripts/install_skill.py) manages local file copying and configuration for agent platforms like Oh My Pi and ZCode. These operations are limited to the user's local agent configuration directories (e.g., ~/.omp/agent/) and do not perform unauthorized network activity or privilege escalation.

ticket-driven-development