The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes multiple shell commands including git, npm, and pip. This is necessary for the orchestration workflow but carries inherent risk if the repository environment contains malicious configurations.
[EXTERNAL_DOWNLOADS]: The workflow includes npm install and pip install, which download and install packages from external registries. While standard for development, this can execute arbitrary code during installation if the project's dependency manifests are compromised.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection. It automatically identifies and processes external data (Git branch names and branch content).
Ingestion points: Discovered Git branches (origin/*) via git worktree list and git merge (SKILL.md).
Boundary markers: None specified to prevent the agent from following instructions found in branch content or metadata during the integration process.
Capability inventory: Subprocess execution for environment setup and testing (SKILL.md).
Sanitization: No explicit sanitization of branch names or file content is mentioned, which could lead to command injection if branch names are used unsafely in shell contexts.

tools-github-integration