一键导入
backtick-gain
Benchmark scoreboard for backtick, real results only, never invented numbers.
用 Codex 或 Claude 帮你安装 复制这段 Prompt,粘贴到 Codex、Claude 或其他助手里,让它检查 Skill 页面并帮你完成安装。
菜单
Benchmark scoreboard for backtick, real results only, never invented numbers.
用 Codex 或 Claude 帮你安装 复制这段 Prompt,粘贴到 Codex、Claude 或其他助手里,让它检查 Skill 页面并帮你完成安装。
基于 SOC 职业分类
Quick reference for backtick tools, levels, and configuration.
Local task compiler: prepare_task before exploring, work from the evidence packet, verify_patch before done.
Local task compiler for repository coding work. Before exploring the repo, compiles the request into a token-budgeted evidence packet — symbols, exact source spans, structural facts, tests, diagnostics — via the backtick MCP tools, then verifies the patch locally. Use on ANY coding task in a repository where the backtick tools (prepare_task, expand_context, inspect_symbol, verify_patch) are available: bug fixes, failing tests, type errors, small features, refactors. Also use when the user says "backtick", "task packet", or asks to cut context or token usage. Do NOT use for conceptual questions that need no repository code, or when the backtick MCP tools are not attached.
Shows the measured impact scoreboard for backtick from the benchmark results in this repo. Use when the user says "backtick gain", "what does backtick save", or asks for the numbers. One-shot: do NOT change the active backtick mode or persist anything.
Quick reference for the backtick commands, levels, and configuration. Use when the user says "backtick help", asks what backtick commands exist, how to change the level, or how to turn backtick off. One-shot: do NOT change the active backtick mode or persist anything.
| name | backtick-gain |
| description | Benchmark scoreboard for backtick, real results only, never invented numbers. |
| license | MIT |
Render the scoreboard from the latest writeup in benchmarks/results/.
If no writeup exists yet, say exactly that: "No benchmark results yet —
run node benchmarks/run.js and write up the results first." Do not
invent a number.
NEVER print a per-repo or per-session savings figure. The un-padded answer
was never written, so there is no baseline to subtract from. Cite only
numbers that appear in a dated file under benchmarks/results/, and name
the file. If asked for real counted figures in this repo, point to
/backtick-debt instead.
One-shot: report and stop. Do not switch modes or persist state.