The Agent Skills Directory

[COMMAND_EXECUTION]: Several core scripts, such as run_experiment.py (autoresearch-agent) and result_ranker.py (agenthub), utilize subprocess.run with shell=True to execute user-provided evaluation and build commands. This is an intended feature of the toolkit to allow developers to run arbitrary test suites and benchmarks.
[COMMAND_EXECUTION]: The git-worktree-manager and agenthub utilities perform shell-based operations to manage Git worktrees, branches, and tags. These actions are clearly documented as part of the skill's primary purpose of parallel development orchestration.
[PROMPT_INJECTION]: The agenthub and autoresearch-agent skills define prompt templates that interpolate user-supplied strings (such as task descriptions or file content) into agent instructions. While this creates a surface for indirect prompt injection, it is managed by the platform's underlying guardrails and is necessary for the skill's collaborative logic.
[EXTERNAL_DOWNLOADS]: Utilities like dependency-auditor and git-worktree-manager can trigger package installations (e.g., npm install, pip install) if the user enables dependency management flags. These operations are performed via standard package managers based on local project manifests.

engineering-advanced-skills