What is cc-harness-skills?

If you use AI coding agents like Claude Code, Codex, or OpenClaw, you have probably hit the same wall: the agent forgets things mid-session, claims a task is done when it is not, or loses track of user corrections after a long conversation. CC Harness Skills is a downloadable pack of six "skills" you install into your coding agent to fix exactly those problems. Think of each skill as a plug-in that teaches your agent a better habit, for memory, for verifying its own work, for managing long context, and for coordinating multiple agents at once. The six skills are: Dream Memory, which consolidates logs and session history into a compact, durable memory index so your agent does not forget what matters, Memory Extractor, which pulls out durable facts about the user, project, feedback, and references from recent turns, Verification Gate, which runs a read-only check after the agent says it is done, to distinguish actual completion from a hopeful claim, Swarm Coordinator, which splits big tasks into research, synthesis, implementation, and verification using bounded workers, Structured Context Compressor, which produces a nine-part summary so long sessions or handoffs do not lose key details, and Kairos Lite, which adds lightweight, time-aware proactive jobs with explicit limits so the agent can do background work safely. You would use this if you are a builder or team relying on coding agents and want better behavior without forking an entire agent product. For example, if your agent keeps "forgetting" your coding style preferences across sessions, Dream Memory and Memory Extractor address that. If you are tired of an agent claiming a bug fix is complete when it never actually ran the tests, Verification Gate is the skill for that. The repo recommends starting with just three, Dream Memory, Verification Gate, and Structured Context Compressor, as the fastest way to feel whether the pack is useful. What is notable is that these skills were distilled from a publicly mirrored coding-agent codebase, then rewritten so they are portable across multiple agent runtimes rather than locked to one vendor. Each skill is a clean bundle: a prompt template, helper scripts, and documentation, designed to be installed with a single command or copied manually. The project explicitly frames itself as "not a source dump" but a curated, tested skill layer.