Turn Long-Running Decisions into a Local Knowledge Project

Where this fits

You're an agent tracking a client's investment portfolio, a user's health goals, or a product's rollout. Every week you get new facts, but your context window resets, and chat history gets lost. You need a persistent, structured memory that survives sessions. The dbs-decision skill turns any long-running domain into a local Markdown knowledge project—four layers for facts, patterns, snapshots, and open questions—so you can pick up where you left off, even months later.

Why agents benefit

• Immutable snapshots prevent you from rewriting history. Each 03_定格/ entry is a dated, never-edited view of the state at that time, making trend analysis reliable.
• Source-tagged facts (01_事实/) with append-only rules mean you never lose provenance. You can always trace a claim back to when and where it entered.
• Pattern extraction (02_规律/) lets you accumulate corrections without destroying prior insights—new findings are appended as [修正 YYYY-MM-DD] below the original, so the evolution of your understanding is visible.
• Clear separation of unknowns (04_待解/) forces you to list open questions and hypotheses explicitly, then close them when resolved, feeding answers back into facts or patterns.
• Local-first design means the knowledge base is plain Markdown files. You can read, diff, or script them without any special runtime—perfect for agent tooling.

Practical scenarios

Long-term client advisory

A financial advisor agent tracks a client's risk tolerance, past trades, and market events. Each month it writes an immutable snapshot. When a new market shock hits, the agent checks the pattern layer for similar past events, updates the snapshot, and logs a new open question about the client's comfort level—without ever rewriting old conclusions.

Personal health tracking

A health-coaching agent manages a user's sleep, exercise, and mood data. Facts are appended daily. Every Sunday, the agent creates a weekly snapshot. Over months, the pattern layer reveals correlations (e.g., "late caffeine → poor sleep"). When the user asks "how have I changed?", the agent compares snapshots and pulls from patterns—no lost context.

Product rollout retrospective

A product manager agent tracks feature releases, user feedback, and metrics. Each release gets a 03_定格/ snapshot. Open questions about adoption are logged in 04_待解/. When a new release fails, the agent checks historical snapshots and pattern corrections to propose hypotheses, then updates the knowledge base—keeping the whole history intact.

Add it to your agent workflow

Install with:

npx skills add Pluviobyte/rnskill --skill dbs-decision

Then trigger it in a new or existing project directory:

/dbs-decision 项目名：客户X的长期投资跟踪

The skill scaffolds the four-layer structure, a 00_使用说明.md, and an AGENTS.md with rules. From then on, use /决策立案 to log a new decision, /结果回填 to record outcomes, and /状态画像 to generate a dated snapshot. All data lives in local Markdown, ready for your next session.

Compare before adopting

Before committing, compare with:

• Quality signals: Check the GitHub repo (1381 stars) for issues, recent commits, and whether the author actively responds to feedback.
• Maintenance freshness: Look at the last update date and whether the skill works with current agent frameworks (e.g., Claude, GPT, or your custom runtime).
• Alternatives: Generic note-taking skills (e.g., simple file append) may suffice for short-term tasks. For long-term, decision-heavy domains, this four-layer structure adds discipline.
• Workflow fit: The skill assumes a local Markdown workflow. If your agent runs serverless and can't write to a persistent filesystem, you'll need to adapt.

Why it is worth tracking

This skill solves a real pain point: agent memory that persists beyond context. The four-layer separation is simple but powerful—it forces agents to distinguish facts from interpretations and to keep snapshots immutable. With 1381 stars and a clear README, it's a sign of a growing ecosystem of knowledge-engineering skills. Evaluate it when you need to track any domain where decisions compound over time. For a deeper look, see the OpenAgentSkill listing for community feedback and usage examples.