The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes multiple shell commands and CLI tools to manage the repository state.
Evidence: git merge origin/<base>, gh pr merge <号> --match-head-commit <HEAD>, and git branch -D <分支>.
Evidence: Executes Node.js scripts using node <git-prefs 技能目录>/scripts/empty-worktree.mjs and $t3-code's t3-worktree-gc.mjs.
[INDIRECT_PROMPT_INJECTION]: The skill reads and follows instructions from untrusted or external files within the repository to determine its logic.
Ingestion points: AGENTS.md, .agents/skills/, and PR reports/comments.
Boundary markers: None specified to separate documentation content from instructions.
Capability inventory: Full git access, PR merge capabilities via GitHub CLI, and execution of local Node.js scripts.
Sanitization: No evidence of input validation or sanitization when processing instructions found in repository files.
[DYNAMIC_EXECUTION]: The skill dynamically identifies and executes script files located in directories defined by other skills.
Evidence: node <git-prefs 技能目录>/scripts/empty-worktree.mjs uses a variable-based path to locate and execute an executable script.

pr-finish