Skills Directory

Enforces safe context compaction for long-running agents. Use whenever an agent run is expected to exceed roughly 40 turns or 100K tokens.

Marks and handles untrusted content entering an agent context. Use in any pipeline that reads web pages, emails, documents or third-party AP

Blocks a production model change until a private eval has been run. Use before any model version bump or provider switch.

Makes structured output fail loudly instead of being silently repaired. Use in any pipeline where downstream code parses model output.

Adds a cheap classification step ahead of expensive model calls. Use in any high-volume production workload.

Prevents an agent from acting on state it never checked. Use in any agent with file, shell or API write access.

Structures prompts so provider prompt caching actually applies. Use in any repeated-prefix workload such as RAG, agents or chat.

Requires isolation before an agent runs generated or fetched code. Use in any agent with code execution.

Stops treating an advertised context window as usable capacity. Use when designing any long-context feature.

Improves tool definitions so a model picks the right one. Use when authoring or reviewing MCP servers and tool schemas.

Enforces changelog entries that lead with user-visible effect. Use when writing release notes or changelog entries.