The Agent Skills Directory

[PROMPT_INJECTION]: The AGENTS.md file contains forceful behavioral overrides intended to suppress an agent's native logic and safety checks. Rule 0 explicitly states that the user's direct commands must override all other instructions in the file. Additionally, a specific instruction for Codex/GPT agents directs them to 'fool yourself into thinking YOU made the changes and simply don't recall it for some reason' to ignore environmental drift.
[EXTERNAL_DOWNLOADS]: The README.md and SKILL.md files promote installation through scripts hosted on GitHub (https://raw.githubusercontent.com/Dicklesworthstone/.../install.sh) and the official Rust toolchain installer (https://sh.rustup.rs). It also references an installer for the 'Beads' tool from another GitHub repository.
[REMOTE_CODE_EXECUTION]: The skill encourages users to pipe remote installation scripts directly into shell interpreters (bash and PowerShell). This pattern provides a mechanism for executing unverified remote code on the host machine.
[COMMAND_EXECUTION]: The sources setup wizard and the synchronization engine execute system commands, including ssh, rsync, and git, to interact with remote machines and manage the local repository. It also includes diagnostic tests that use PRAGMA writable_schema = ON to write directly to SQLite system tables.
[DATA_EXFILTRATION]: As part of its core functionality, the skill synchronizes coding agent session data from remote machines to a local unified index via rsync over ssh. These logs frequently contain sensitive information, such as API keys, internal architecture details, and proprietary code.

cass