The Agent Skills Directory

[COMMAND_EXECUTION]: The skill generates and manages shell scripts (worktree-up.sh, worktree-down.sh, dev.sh) that are executed via IDE hooks or manually. These scripts automate system-level tasks including dependency installation, database lifecycle management, and local server execution.
[DATA_EXFILTRATION]: The skill facilitates the movement of sensitive environment files containing credentials (.env, .env.local) between directories. While intended for local environment parity and lacking external network exfiltration, it involves systematic access and manipulation of plaintext secrets.
[DYNAMIC_EXECUTION]: Shell scripts are dynamically assembled by the agent based on user input and repository metadata. These scripts are later executed by the integrated development environment, representing a runtime code generation and execution pattern.
[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted repository content such as lockfiles, database schemas, and directory structures to determine its configuration logic. It relies on AI research to handle unknown backends, which introduces a surface for processing potentially malicious data.
Ingestion points: Repository-specific files (lockfiles, prisma/schema.prisma, convex/schema.ts) and AI-driven research of the codebase.
Boundary markers: The skill utilizes an interactive prompt flow to confirm settings and mandates research summaries before script modification.
Capability inventory: Write access to the repository for script creation, execution of shell commands via IDE hooks, and reading of environment files.
Sanitization: Includes the assert_non_prod_ref utility to prevent operations on production environments and slugifies worktree names for safe database naming.

environments-manager