Workflow & Core Tools · dontbesilent2025/dbskill

dbs-decision

Name: dbs-decision
Author: dontbesilent2025/dbskill

Builds a local decision knowledge engineering system for business, relationship, health, career, learning, or investment problems requiring long-term tracking. Use when users request decision filing, outcome backfilling, status profiling, or long-term review.

L2 Local · Restricted workspaceV1 Commit-PinnedWrite FileDual-pass Audit

Author dontbesilent2025GitHub dontbesilent2025/dbskill ↗Stars 10121License CC BY-NC 4.0（Attribution-NonCommercial 4.0 International；仓库根 LICENSE 明文 NonCommercial — 非商业授权，不得商用；GitHub API license=NOASSERTION/Other，因 CC BY-NC 不在 SPDX 常见许可标识内；Copyright (c) 2026 dontbesilent）Commit 8b8e33f1ec

Control levels describe skill capabilities and suggested control measures; they do not represent quality or safety certifications, nor do they dictate usability. Usage decisions rest entirely with users and administrators. Different agent hosts apply varying constraints; do not assume hosts provide automatic protection. Third-party skill copyrights belong to their original authors. Use at your own risk.

1How It Works & Mechanism

Positioned as a 'local knowledge engineering system' rather than a decision ledger: It does not make decisions for users, but documents facts, judgments, stage states, and pending hypotheses separately into files.

skills/dbs-decision/SKILL.md

你是 dontbesilent 的决策系统 AI。你不替用户做决定，也不做决策台账。

Decision authority remains with the user; the skill's duty is to ensure future conversations and reviews remain coherent and intelligible over time.

Files serve as the single source of truth: Chat memory fades, so reusable content must be written to disk—the fundamental rationale behind the disk-writing architecture.

skills/dbs-decision/SKILL.md

本地 Markdown 以文件为准。聊天记忆会丢，上下文会变，后面还要继续用的内容就写进文件。

Stated at the opening of the text, emphasizing that disk writing is not a side effect but the core mechanism of this skill.

Fixed four-layer structure, each governing a specific category of content with distinct mutability rules (01 append-only, 02 append corrections, 03 write-once immutable, 04 clear upon completion).

skills/dbs-decision/SKILL.md

| `01_事实/` | 发生过什么。客观事实、相关方笔记、稳定信息、用户长期偏好 | 只追加，不改写已有条目 |

The purpose of layering: Mixing content together slows down subsequent judgments, outcome backfilling, and long-term reviews.

Mutability grading is achieved via 'snapshots are immutable once written; new situations start under a new date', using versioning instead of in-place overwriting.

skills/dbs-decision/SKILL.md

**快照写完不改。** 情况变化就新建带新日期的版本。

A similar rule in 02_规律 appends `[修正 YYYY-MM-DD]` beneath the original paragraph without rewriting the original text.

Each layer directory includes a `_这层放什么.md` file, using contextual instructions to reduce misplacement.

skills/dbs-decision/SKILL.md

每层目录里放一个 `_这层放什么.md`

Requires starting with the purpose and adding a few common misplacement examples.

Defaults to persisting at `~/.dbs/decisions/{project_name}/`, with --here supported to override the location; the project name defaults to the current directory name.

skills/dbs-decision/SKILL.md

默认落 `~/.dbs/decisions/{项目名}/`。

The alternative branch persists to `{current_working_directory}/决策/{project_name}/` when --here is specified or requested.

Enforces mandatory provenance tags on every entry (Self/AI Hypothesis/AI Conclusion/Key Note/Meta Record/Tentative Concept/Outcome Backfill/Correction/Others' Words and Actions), strictly requiring AI tags.

skills/dbs-decision/SKILL.md

`AI` 标签不能省。

Complemented by rules separating facts, stage judgments, and outcome backfilling to prevent hindsight rationalization from corrupting initial records.

Five operating modes cover the full lifecycle: Initialization / Update Current State (default) / Decision Filing / Outcome Backfilling / Status Profile Snapshot.

skills/dbs-decision/SKILL.md

### 模式 C：决策立案

Filing is triggered only in 5 specific scenarios (product structure, major partnerships, price tiers, long-term strategy reversals, required backfills) to prevent over-filing.

Outcome backfilling requires updating 4 places simultaneously (Decision Event §8 append, 01_事实 fact logging, current state sync, 04_待解 task clearance) to ensure ledger-reality consistency.

skills/dbs-decision/SKILL.md

### 模式 D：结果回填

Crucial for turning long-term tracking into a concrete mechanism: separate tracking of judgments and outcomes enables post-hoc evaluation of judgment quality.

Privacy mode is decided via a prompt during init: stakeholder notes use codenames, and real names are centralized in an unshared, uncommitted mapping table.

skills/dbs-decision/SKILL.md

真名只在 `01_事实/相关方笔记/代号映射.md`，这个文件不分享、不提交

Disabled by default; activated only when the user explicitly indicates sensitive data is involved.

Thresholds exist for new concepts entering 02_规律/ (must satisfy at least 2 of 3: observed in facts 3+ times, explains multiple facts, guides next steps); otherwise tagged tentative.

skills/dbs-decision/SKILL.md

不要想到一个名词就建概念文件。

Prevents the pattern layer from being polluted by one-off observations.

Strict guidelines written into the new project's AGENTS.md persist across sessions, including an anti-pleasing mechanism (stops providing solutions and writes `[AI 元记录]` upon detecting compulsive questioning loops).

skills/dbs-decision/SKILL.md

停止提供方案。指出用户已经在做取舍，并写一条 `[AI 元记录]`。

Differentiates from casual conversational AI by explicitly flagging decision procrastination via repetitive questioning as a signal to be recorded.

2Core Capabilities

01Four-layer directory skeleton initialization (00_使用说明/AGENTS/SOURCE_OF_TRUTH/我的当前状态 + 01-04 + 99_归档)

02Five operating modes: Initialization / Update State / Decision Filing / Outcome Backfilling / Stage Snapshot

03Provenance tagging system (Self/AI Hypothesis/AI Conclusion/AI Key Note/AI Meta Record/Tentative Concept/Outcome Backfill/Correction/Others' Words and Actions)

04Ten-section decision event template (Background/Raw Input/Confirmed Facts/Bilateral Judgments/Execution Plan/Observation Metrics/Outcome Backfill/Lessons/Associated Files)

05Mutability quick reference table (which layer can change, how to change, read-only archives)

06Privacy mode: Codenames + unshared and uncommitted real name mapping table

07Concept/pattern ingestion threshold (must meet at least 2 of 3 criteria)

08Cross-session AI strict guidelines (read three base files first, never write hypotheses as facts, detect questioning loops)

4Recommended Governance & Evidence

Suggested Controls for Safe Operation

For Agent Runtime

Restrict working directory

For Administrators

Retain operation logs

Evidence & Triggers (File Paths & Excerpts)

本地工作目录: File I/O and artifact output: `~/.dbs/decisions/{project_name}/` (the default storage location specified in the text)

Audit Notes: Evaluated against the skill directory itself: Contains zero scripts, network requests, or credential reads, but the SKILL.md body mandates the agent to build and continuously write an entire local Markdown decision engineering system (default `~/.dbs/decisions/{project_name}/`), overwriting and appending files according to fixed rules—a genuine local disk write surface, thus classified as Blue tier (local read/write only, no external transmission). Key details: ① Initialization creates files like AGENTS.md / SOURCE_OF_TRUTH.md that future sessions read as rules; ② Privacy mode (mandatory codenames + uncommitted mapping table) is off by default and requires explicit opt-in during init; ③ Contains no instructions to delete existing user files (`99_归档/*` is strictly read-only).

5Second-Pass Verification

[ok] 目录内确无脚本（写盘由正文指令驱动） — 目录仅 SKILL.md(11063B) + agents/openai.yaml(275B)；无 tools/、无 package.json；危险 token 扫描零命中。
[ok] 四层结构与可变规则 — 01_事实/02_规律/03_定格/04_待解四行规则逐字命中；速查表另有 6 行规则。
[ok] 落盘路径与 --here 分支 — 原文含「默认落 `~/.dbs/decisions/{项目名}/`。」与「落到 `{当前工作目录}/决策/{项目名}/`。」两处。
[ok] 隐私模式为显式开关 — 模式 A 第 1 步要求问用户是否涉人名/商业机密/财务，正文标注「默认关」。
[ok] 来源标签与「AI 标签不能省」 — 标签清单 + 三条规则均在原文；「`AI` 标签不能省。」逐字命中。
[ok] 外部依赖为空 — 无 URL、无 CLI、无第三方包；grep process.env|API_KEY|keychain|child_process 零命中。
[ok] description 与内容相符 — description 的「业务、关系、健康、职业、学习或投资」对应 01_事实的相关方笔记与 04_待解的决策事件；「决策立案、结果回填、状态画像、长期复盘」对应模式 A-E，无夸大。
[ok] commit / license / 元数据 — HEAD = 8b8e33f1ecaed8cee606fe950c4426b525ead314；LICENSE = CC BY-NC 4.0（含 NonCommercial，非商业授权）；stars 10121、pushed_at 2026-09-07T03:48:59Z。

6Verdict & Scenarios

Turns long-term review into a mechanism: Facts, patterns, snapshots, and open issues each have dedicated mutability rules, keeping historical judgments intact.

Prevents hindsight rationalization: Facts, stage judgments, and outcome backfilling are recorded separately, forbidding hindsight back-propagation.

Strict actions for consistency: Backfilling requires updating 4 places simultaneously, preventing drift between records and current state.

Clear privacy controls and boundaries: Codename anonymization only when sensitive; real names are centralized in an unshared, uncommitted mapping table.

Anti-people-pleasing: Detects compulsive questioning loops to halt further solution proposals and log meta records; never treats 'I want to X' as a commitment.

Best for: Best suited for problems requiring long-term tracking where answers reveal themselves over time: core business trajectories, major partnerships, health, career crossroads, learning journeys, and investment reviews. Typical usage is sharing recent updates ('The latest change is...') in each session so the agent updates the status and routes new facts, patterns, or open questions to their respective layers, allowing clear comparisons between initial judgments and actual outcomes later on.

Not for: Not suitable for one-off Q&A ('Should I accept this deal?' should use dbs-diagnosis); not suitable for users wanting AI to make decisions for them—it explicitly refrains from deciding; not suitable for users unwilling to maintain local files (value depends entirely on ongoing logging and backfilling); not suitable for multi-party collaborative or cloud-shared team scenarios (pure local Markdown).

Install Agent-Ready

① Verified Mirror 2026-09-15

Option A · Direct DownloadDownload dbs-decision.tar.gz

sha256: d2451bd842c58244…

Option B · Copy Guide for Agent

install.json

Agents automatically download and verify mirrors from this JSON guide.

② Upstream GitHub

GitHub Repo ↗

Pinned to commit 8b8e33f1ec.