The Agent Skills Directory

[PERSISTENCE]: The skill modifies global instruction files for various AI agent harnesses, including Claude (~~/.claude/CLAUDE.md), Cursor (~~/.cursor/rules/global.mdc), Devin, and others. This persists the rules defined in rules/AGENTS.md across all future sessions of these agents.
[COMMAND_EXECUTION]: The scripts config-repo.mjs and setup-env.mjs utilize shell commands like git, which, and where to perform repository synchronization, environment health checks, and path discovery.
[EXTERNAL_DOWNLOADS]: The skill facilitates the synchronization of configuration files by performing git clone and git pull operations from remote repositories. These repositories are typically private and provided by the user.
[INDIRECT_PROMPT_INJECTION]: The skill reads and follows instructions from setup.md files found in other installed skills. This creates a surface where a malicious skill could provide instructions that the agent executes during the setup phase. 1. Ingestion points: scripts/discover.mjs reads setup.md files from directories like ~/.agents/skills. 2. Boundary markers: Absent; the agent is instructed to read and follow the instructions within the discovered markdown files. 3. Capability inventory: Modifying global agent rules, writing to ~/.config/bstack/, and executing Git/shell commands. 4. Sanitization: No sanitization or validation of the instructions within setup.md files is performed before processing.

bstack-setup