The Agent Skills Directory

[SAFE]: The skill is a meta-instruction set for authoring other skills. It defines a rigorous workflow for testing how agents handle conflicting pressures (Time, Authority, Sunk Cost, etc.).
[COMMAND_EXECUTION]: While the skill instructions frequently mention 'running' tests and 'executing' workflows, it does not contain any direct shell command execution or script files. The logic is entirely handled via natural language instructions to the agent.
[DATA_EXFILTRATION]: The skill references local state files such as .beads/onboarding.json, STATE.json, and HANDOFF.json for context management. This is consistent with its stated purpose of managing multi-step development cycles and does not involve sending data to external domains.
[PROMPT_INJECTION]: The skill uses 'HARD-GATE' tags to enforce procedural compliance. While these are strong behavioral constraints, they are used to ensure the reliability of the skill-writing process and do not attempt to bypass the underlying AI safety filters.
[DATA_EXPOSURE]: The skill instructions allow for the inclusion of external 'Source Material' and 'Exact rationalization' quotes into log files. While this creates a surface for Indirect Prompt Injection, the risk is negligible as the skill focuses on documenting these inputs rather than executing them as commands.

beo-writing-skills