dickwu/mem-setup · Archived

mem-save

Save domain knowledge insights after completing a task.

First seen Mar 30, 2026

Installation

$ npx skills add dickwu/mem-setup --skill mem-save

Summary

  • Save domain knowledge insights after completing a task.
  • Extracts facts, patterns, and observations from the current work and stores them in /knowledge/ domain folders.
  • Promotes hypotheses to rules at 5+ confirmations, demotes rules when contradicted.
  • Use when finishing a task, or when the user says done/finished/complete.

Stronger alternatives

This repository is archived — consider an actively maintained alternative.

Similar popular skills

Related neighbors and high-traction skills in the same topics — useful to compare before installing.

Also in this package

Other skills from dickwu/mem-setup.

npx skills add dickwu/mem-setup

Browse all from dickwu/mem-setup

More details

Agent compatibility

Declared targets from SKILL.md / docs. Unmarked agents are not listed — the skill may still install via the CLI.

Claude Code Not declared
Cursor Not declared
Codex Not declared
GitHub Copilot Not declared
Windsurf Not declared
Gemini CLI Not declared
Cline Not declared
OpenCode Not declared

Also listed on

Alternate registries and mirrors of this skill.

Repository health

License MIT
Default branch main
Open issues 0
Status Archived

Package contents

Files included with this skill beyond the listing page.

  • skill md SKILL.md 2,899 B
  • docs SUMMARY.md 339 B

History

  1. First seen on skills.sh
  2. First recorded snapshot · 1 installs

SKILL.md

Memory Save

Extract insights from the completed task and persist them to the knowledge system.

Steps

1. Reflect on the Task

Ask yourself:

  • What did I learn about the codebase that wasn't already in knowledge.md?
  • Did any hypothesis get confirmed or contradicted?
  • Did I discover a new pattern that should be tracked?
  • Did I make a mistake that revealed a rule worth recording?

2. Identify the Domain

Match insights to existing domain folders in knowledge/INDEX.md. If no folder fits, create a new domain folder with all three files (knowledge.md, hypotheses.md, rules.md) and add it to INDEX.md.

3. Categorize and Write

For each insight, decide where it goes:

knowledge.md — Confirmed facts and patterns

- **[topic]**: [fact] (observed [date])

hypotheses.md — Patterns that need more data

- **[topic]**: [hypothesis] (confirmations: N/5, first observed [date])

rules.md — Confirmed 5+ times, apply by default

- **[topic]**: [rule] (promoted [date], confirmed N times)

4. Promotion Check

Scan hypotheses.md in relevant domains. If any hypothesis now has 5+ confirmations:

  • Remove it from hypotheses.md
  • Add it to rules.md with (promoted [today's date], confirmed N times)
  • Tell the user: "Promoted to rule: [hypothesis]"

5. Demotion Check

If today's work contradicted an existing rule in rules.md:

  • Remove it from rules.md
  • Add it back to hypotheses.md with (demoted [today's date], contradicted by: [reason])
  • Reset confirmation count to 0
  • Tell the user: "Demoted rule: [rule] — contradicted by [reason]"

6. Update Hypothesis Counts

If today's work confirmed an existing hypothesis (but it's not yet at 5):

  • Increment its confirmation count
  • Tell the user: "Confirmed: [hypothesis] (now N/5)"

7. Summary

Report what was saved:

Saved to [domain]:
  + N new knowledge entries
  + N new hypotheses
  ^ N hypothesis confirmations
  * N promotions to rules
  v N demotions from rules

Ask Before Saving

Before writing, briefly list what you plan to save and ask: "Save these insights?" This gives the user a chance to add, remove, or correct entries before they're persisted.

Writing Good Entries

  • Be specific and actionable — future sessions need to understand and apply these
  • Include dates so staleness can be assessed
  • One insight per bullet point
  • Don't duplicate existing entries — update them instead
  • Skip things derivable from reading the code — focus on non-obvious patterns and gotchas