The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes Git and shell commands to manage the worktree lifecycle. It includes multiple safety checks to prevent accidental data loss or system errors:
Validates repository root using git rev-parse --show-toplevel before performing any operations.
Checks the status of a worktree (git status --short) before removal to ensure no uncommitted changes are lost.
Implements a multi-step logic for deleting a worktree the user is currently inside: it resolves the target path, moves the execution context to a safe main worktree, and then executes the removal command.
[INDIRECT_PROMPT_INJECTION]: The skill ingests natural language input (task intent, branch names) which is used to construct shell commands and directory paths. It proactively addresses this attack surface:
Ingestion points: User descriptions of tasks or branch names provided during prompts (rules/worktree-lifecycle.md, Section 2).
Boundary markers: Instructions specify using a "sanitized slug algorithm" to transform user input into safe labels.
Capability inventory: Shell commands including git worktree add, mkdir, and printf are used across the lifecycle rules (rules/worktree-lifecycle.md).
Sanitization: Rule 2 defines a strict sanitization process: lowercasing input, replacing shell-sensitive characters and whitespace with hyphens, and explicitly rejecting reserved labels like ., .., or .git to prevent path traversal or command injection.
[EXTERNAL_DOWNLOADS]: The references/source-survey.md file contains a curated list of external URLs for documentation and community tools. These references target well-known technology companies and services (such as Git-SCM, Anthropic, JetBrains, Microsoft, and official package registries) and are documented neutrally as research rationale with no automated execution logic.

git-worktree