Balanov/uncut
UNCUT preserves the full conversation on every model call and fails explicitly when a provider cuts an answer short.
The 7 MB coding-agent harness runs through stdin/stdout for shell pipelines, CI jobs and containers. It provides four tools, with no interface, telemetry, automatic retries or context compaction. It supports OpenAI-compatible endpoints and can resume complete histories from append-only JSONL sessions. Its “lossless” promise does not extend model context limits, and shell commands run with the user’s permissions without sandboxing or approval prompts.
GitHub · LLM repos' note
The 7 MB coding-agent harness runs through stdin/stdout for shell pipelines, CI jobs and containers. It provides four tools, with no interface, telemetry, automatic retries or context compaction. It supports OpenAI-compatible endpoints and can resume complete histories from append-only JSONL sessions. Its “lossless” promise does not extend model context limits, and shell commands run with the user’s permissions without sandboxing or approval prompts.
GitHub · LLM repos' note
score 4