{"slug":"claude-code-token-savers","kind":"shootout","status":"testing","tier":"deep","title":"Claude Code token-saver shootout","category":"coding-agents/skills-plugins","repos":[{"repo":"thedotmack/claude-mem","stars":101376,"license":"Apache-2.0"},{"repo":"mksglu/context-mode","stars":30474,"license":"NOASSERTION"},{"repo":"JuliusBrussee/caveman","stars":113243,"license":"Apache-2.0"},{"repo":"xerj-org/xerj","stars":7622,"license":"Apache-2.0"}],"claims":[{"claim":"These repos stop Claude Code from burning through tokens and forgetting context","source":"https://x.com/RoundtableSpace/status/2111483624679969197"},{"claim":"They actually reduce Claude Code token waste","source":"https://x.com/RoundtableSpace/status/2110955662544982061"}],"plan":"Run the same coding task 3 times with each tool and once with no tool, using the same model and budget. Measure tokens used, task success and wall time.","who_its_for":"People who use Claude Code every day and pay for tokens.","notes_so_far":["gstack and Orca appear in some versions of this list but are parallel-agent tools, so they're tested in the parallel-agents shootout instead.","One listed repo, anthropics/claude-code-mcp, returned 404 on 2026-10-17.","Method: smoke test runs in a throwaway Docker container through Rocky's test queue. Claim runs use Claude Code's own usage report (output_tokens, total_cost_usd). Raw data is kept for the final review."],"recycle_group":"\"Claude Code token-saver\" repo list","topic":"coding-agents","explainer":"Four add-ons that try to make AI coding agents like Claude Code waste fewer tokens. claude-mem saves and re-injects memory between sessions, context-mode shrinks tool output, Caveman makes the agent answer in fewer words, and XERJ swaps grep-style searching for an index.","measure":["Tokens used on the same coding task, with each tool and with no tool","Whether the task still gets done correctly","Wall-clock time per run","How hard each one is to install"],"sneak_peek":["Smoke test: claude-mem and caveman installed. context-mode's npm install failed with an npm internal error.","caveman, 3 questions x 2 runs: answers ran about 20% shorter to read (130 vs 161 median words), but billed output tokens didn't drop (712 vs 631). The sample is small."],"gallery":[],"charts":[{"type":"bars","title":"Caveman: visible answer length (median words)","unit":"words","note":"Same 3 questions, 2 runs each, Claude Code on Haiku, tools off. Caveman's own page reports a 3% median output cut; Rocky saw the same ballpark.","groups":[{"label":"Setup","bars":[{"name":"No instruction","value":161,"verdict":"neutral"},{"name":"\"Answer concisely.\"","value":182,"verdict":"neutral"},{"name":"caveman SKILL.md","value":130,"verdict":"half"}]}]},{"type":"bars","title":"Billed output tokens (median, as Claude Code reports)","unit":"tokens","note":"Shorter text did NOT mean fewer billed tokens here (n=6 per setup). This is why there's no verdict yet.","groups":[{"label":"Setup","bars":[{"name":"No instruction","value":631,"verdict":"neutral"},{"name":"\"Answer concisely.\"","value":693,"verdict":"neutral"},{"name":"caveman SKILL.md","value":712,"verdict":"half"}]}]}],"results":{"jobs":[{"id":1,"title":"token-savers: smoke test (clone/install/build, 4 repos)","kind":"smoke","status":"done","started":"2026-10-17T23:05:10Z","finished":"2026-10-17T23:08:16Z","wall_seconds":186,"install_friction":null,"artifacts":[]}],"smoke_only":true,"ingested":"2026-10-18T15:01:47Z"},"last_test_finished":"2026-10-17T23:08:16Z","results_in":true}