Skip to content

mnemebrain/mnemebrain-skills

Repository files navigation

MnemeBrain Skills

Claude Code skills for working with the MnemeBrain belief memory system.

Skills

mnemebrain-lite-integration

Helps developers integrate with MnemeBrain Lite using the official Python and TypeScript SDKs. Covers the four core operations (believe, revise, retract, explain), WorkingMemoryFrame for multi-step reasoning, conflict policies, embedding providers, and the Brain high-level API.

  • Triggers on: MnemeBrain SDK usage, belief memory integration, truth states, evidence polarity, WorkingMemoryFrame, confidence ranking
  • Eval results: 100% pass rate with and without skill. Skill provides 47.6% token savings and 64.8% time savings by eliminating codebase exploration. See evaluation report.

mnemebrain-lite-proof

Proves that mnemebrain-lite's belief memory is better than flat key-value agent memory through runnable side-by-side comparisons. Runs a bundled proof script (proof_demo.py) that computes real numbers across 6 scenarios, then frames the output for the target audience (developer or investor).

  • Triggers on: "why mnemebrain", "prove it", "show me the difference", "what's wrong with a dict", value proposition questions, demo prep
  • Eval results: 100% pass rate with skill vs 79% without. The skill's primary differentiator is running the proof script for verified computed values (0/3 without-skill runs executed it). 48.4% token savings. See evaluation report.

Structure

mnemebrain-skills/
  README.md                              (this file)
  mnemebrain-lite-integration/
    SKILL.md                             SDK integration skill
    references/                          API reference docs
  mnemebrain-lite-integration-workspace/
    REPORT.md                            Evaluation report
    grade.py                             Grading script
    iteration-1/                         Eval results
  mnemebrain-lite-proof/
    SKILL.md                             Proof/demo skill (v3)
    scripts/proof_demo.py                Bundled proof script (6 scenarios)
    evals/evals.json                     Test case definitions
  mnemebrain-lite-proof-workspace/
    REPORT.md                            Evaluation report
    README.md                            Workspace overview
    iteration-1/                         Failed (rate limit)
    iteration-2/                         First successful eval
    iteration-3/                         Final eval (all weaknesses fixed)

Evaluation Summary

Skill With Skill Pass Rate Without Skill Pass Rate Token Savings
mnemebrain-lite-integration 100% (15/15) 100% (15/15) 47.6%
mnemebrain-lite-proof 100% (19/19) 79% (15/19) 48.4%

Both skills provide ~48% token savings. The proof skill additionally provides a correctness advantage by forcing execution of the proof script, ensuring responses contain verified computed values rather than hand-estimated numbers.

About

No description, website, or topics provided.

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

 
 
 

Contributors

Languages