Sign InOpen Brain
GitHubGitHub RepoNeeds Review

JuliusBrussee/caveman

Caveman compresses agent inputs and outputs, with local tooling for noisy commands, browsing, memory, and skills. Its own caveat matters: terse workloads can cost more once the skill prompt is included.

GitHub · Trending today
Open Source Open MarkdownOpen JSON
Source Summary

Caveman combines a terse-response skill with a local proxy that compresses agent inputs. A pinned Claude Code test reports **33.2% fewer input tokens** across 54 runs while passing 18 exact-answer checks; the original skill claims **65% fewer output tokens** across 10 tasks.

Practical Implication

Measure it on your own sessions before adopting it. The CLI can inspect local history, rank token sinks, trial changes, compress command output, and revert edits that fail to lower tokens per turn.

Agent-Ready Context
Caveman combines a terse-response skill with a local proxy that compresses agent inputs. A pinned Claude Code test reports **33.2% fewer input tokens** across 54 runs while passing 18 exact-answer checks; the original skill claims **65% fewer output tokens** across 10 tasks.

Measure it on your own sessions before adopting it. The CLI can inspect local history, rank token sinks, trial changes, compress command output, and revert edits that fail to lower tokens per turn.

The output skill adds **about 1–1.5k input tokens per turn**, so already-terse work can regress. The engine-linked runtime is BSL-1.1 rather than OSI open source, and anonymous command and token-count telemetry is enabled by default but can be disabled.
Connected Context · Feed7 Judgment

Caveman adds direct, reversible token optimization to the candidates’ broader context-management approaches: it measures session-level sinks and can test both input compression and terse output rules. Its mixed economics narrow the lesson—compression is workload-dependent, and the skill’s recurring prompt cost can outweigh savings—so adoption should follow local exactness and tokens-per-turn trials rather than headline reductions.

Blast RadiusBoth treat context reduction as measurable and reversible; Blast Radius archives unused history based on predicted reach, whereas Caveman compresses inputs and command output and can revert changes that fail to save tokens.virgiliojr94/book-to-skillbook-to-skill reduces repeated loading through indexed progressive disclosure for stable documents, while Caveman targets live session and tool-output verbosity; they address different token sinks and both require workload-specific evaluation.WTF Is the Context Layer? The Missing Infrastructure for Production Agents — Prukalpa SankarThe shared-context proposal emphasizes portable, versioned knowledge across agents, whereas Caveman optimizes the amount sent per turn; compression can lower transport cost without supplying the governance or shared semantics of a context layer.
Context Map
contextcoding#context-engineering#context-caching#skills
Uncertainty
The output skill adds **about 1–1.5k input tokens per turn**, so already-terse work can regress. The engine-linked runtime is BSL-1.1 rather than OSI open source, and anonymous command and token-count telemetry is enabled by default but can be disabled.