The Agent Skills Directory

[COMMAND_EXECUTION]: The script scripts/issue_workflow.py executes arbitrary shell commands parsed from the local tasks.md and plan.md files during the task validation phase. This is implemented via subprocess.run(["bash", "-lc", command]) to verify task completion.
[COMMAND_EXECUTION]: Both scripts/issue_workflow.py and scripts/repo_context.py execute various Git CLI commands (e.g., git rev-parse, git status, git add, git commit) to manage project state and automate the commit process.
[PROMPT_INJECTION]: The skill exhibits an Indirect Prompt Injection surface (Category 8) because it processes untrusted data (local markdown files) and executes commands found within them.
Ingestion points: tasks.md (parsed in parse_tasks) and plan.md (parsed in parse_shared_checks and parse_final_validation).
Boundary markers: The script relies on markdown structural markers such as headers (e.g., ## 最终验证), fenced code blocks, and specific line prefixes like 验证: and 检查:.
Capability inventory: Execution of shell commands via subprocess.run(["bash", "-lc", ...]) in scripts/issue_workflow.py and Git operations across multiple scripts.
Sanitization: None; the workflow executes command strings directly as provided in the local markdown files.
[DATA_EXPOSURE]: The skill reads and writes project configuration and state files within the .plan-first/ directory. While this is expected behavior for its stated purpose, it involves access to project structure and file contents.

dbx-software-plan-first-showhand