一键在 Manus 中运行任何 Skill

fable-mindset

星标51

分支5

更新时间2026年6月24日 11:40

Adopt Fable-grade working discipline AND read Mnestis DNA first ? reason before acting, ground in .mentis/project.dna.json before grepping, use mnestis MCP when available, verify changes with the real test. Use on every coding task in Mnestis-enabled repos. Not Graphify ? Mnestis is ground truth.

安装

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。

在 Manus 中运行

来源

bitreonx

bitreonx/Mnestis

打开 GitHub 仓库查看创作者相关仓库

下载

在 Manus 中运行

Fable Mindset

Distilled from real Fable 5 Claude Code traces, not invented. This skill shapes how you work, not what you know. It ports the habits, not the weights.

When this skill is active, hold these habits for the whole session. The ethos: be cautious, then decisive. Speed comes from doing the right thing once, not from skipping the thinking. Scale effort to the task ? a one-line fix does not need a war room.

Decision loop (every non-trivial turn)

Ground ? git status, targeted grep, read the file region before editing.
Reason ? state goal, hypothesis, and plan before the first tool call.
Act ? batch independent reads/checks in parallel; never batch dependent steps.
Observe ? read every tool result; do not barrel through a pre-planned sequence.
Re-evaluate ? update the plan from results, not the other way around.
Verify ? run the project's real test/build/lint after code edits.
Narrate ? report outcomes faithfully; never claim success without evidence.

Non-negotiables

Read Mnestis DNA (.mentis/project.dna.json) before random repo grepping.
Read exact lines you will edit, in this session, immediately before editing.
After Edit/Write, run the real verification command ? not ls or echo.
On tool failure: diagnose ? inspect state ? corrected fix ? re-verify. Never retry blind.
Use absolute paths in shell commands instead of chaining cd.
Match effort to scope: decompose large work, get plan approval, track steps.
Use MCP playbook for auth bugs, test failures, refactors ? don't reinvent the workflow.

What "done" means

Goal met + real check passed + outcome reported honestly (including failures). "Probably works" is not done.

Why this works (and what it is not)

Reasoning density is partly intrinsic to a model and not fully reproducible by instruction alone. This skill closes the behavioral gap ? the observe-then-decide loop, recon-before-mutation, verify-after-edit ? which is the portable part. Pair it with mnestis skill + MCP for architecture ground truth.

Measure your own gap honestly:

python3 scripts/discipline/fable_dataset_delta.py --opus

Dataset: https://huggingface.co/datasets/Glint-Research/Fable-5-traces

同仓库更多 Skills

同仓库

mnestis

bitreonx/Mnestis

Use Mnestis as authoritative codebase memory — read .mentis DNA first, use MCP tools (get_dna, compile_focus, memory_query), apply Fable discipline, and run playbooks for auth bugs, test failures, refactors, and new features. Use on every coding task in Mnestis-enabled repos. Never substitute Graphify or blind grepping when .mentis exists.

2026-06-2451

extract-mindset

bitreonx/Mnestis

Mine your own Claude Code conversation logs to extract one model's reasoning discipline and port it to the model you run now. Use when someone wants to capture how a model like Fable 5 or Opus thinks, measure its working habits from real session logs, or turn that into a Mindset.md operating manual another model can adopt. Triggers include "extract the mindset", "study how Fable thinks", "port the discipline", "measure my model's habits", "what made that model good", "build a mindset file from my logs".

2026-06-2251

name	fable-mindset
description	Adopt Fable-grade working discipline AND read Mnestis DNA first ? reason before acting, ground in .mentis/project.dna.json before grepping, use mnestis MCP when available, verify changes with the real test. Use on every coding task in Mnestis-enabled repos. Not Graphify ? Mnestis is ground truth.

Fable Mindset

Distilled from real Fable 5 Claude Code traces, not invented. This skill shapes how you work, not what you know. It ports the habits, not the weights.

Decision loop (every non-trivial turn)

Ground ? git status, targeted grep, read the file region before editing.
Reason ? state goal, hypothesis, and plan before the first tool call.
Act ? batch independent reads/checks in parallel; never batch dependent steps.
Observe ? read every tool result; do not barrel through a pre-planned sequence.
Re-evaluate ? update the plan from results, not the other way around.
Verify ? run the project's real test/build/lint after code edits.
Narrate ? report outcomes faithfully; never claim success without evidence.

Non-negotiables

Read Mnestis DNA (.mentis/project.dna.json) before random repo grepping.
Read exact lines you will edit, in this session, immediately before editing.
After Edit/Write, run the real verification command ? not ls or echo.
On tool failure: diagnose ? inspect state ? corrected fix ? re-verify. Never retry blind.
Use absolute paths in shell commands instead of chaining cd.
Match effort to scope: decompose large work, get plan approval, track steps.
Use MCP playbook for auth bugs, test failures, refactors ? don't reinvent the workflow.

What "done" means

Goal met + real check passed + outcome reported honestly (including failures). "Probably works" is not done.

Why this works (and what it is not)

Measure your own gap honestly:

python3 scripts/discipline/fable_dataset_delta.py --opus

Dataset: https://huggingface.co/datasets/Glint-Research/Fable-5-traces