The Agent Skills Directory

[SAFE]: The skill provides a high-level operating procedure for 'headless' (autonomous) runs. It reinforces safety by stating that the mode increases stamina but not authority, and explicitly forbids the agent from impersonating individuals. It provides clear decision-making tiers to ensure that ambiguous or novel instructions do not lead to unauthorized actions.
[INDIRECT_PROMPT_INJECTION]: This skill creates an attack surface by instructing the agent to ingest external context and goals (Ingestion point: 'Load context' in SKILL.md). The skill includes boundary markers by directing the agent to strictly follow explicit project rules and patterns. While no specific internal capabilities (like subprocess calls) are defined within the skill's own files, and no explicit sanitization is described, the procedural requirement to log all non-trivial assumptions serves as a tracking mechanism for potentially injected instructions.
[NO_CODE]: The skill consists solely of markdown instructions across multiple language versions. There are no scripts, configurations for external tools, or executable commands included, which prevents direct technical abuse via the skill's file content.

headless