skills.rest
melandlabs
melandlabsOFFICIAL
Explore all skills by this author →
𝕏@AlloomiAIGitHub

continual-learn

Persist learning across turns by maintaining a MENTAL_MODEL.md workspace file.

1.0k|46|Updated Apr 10, 2026
Search Tags:#benchmarking#continual learning#agent memory#Debugging#mental model#iterative tasks#workspace notes
One-click install
npx skills add https://github.com/melandlabs/openloomi --skill continual-learn
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: continual-learn
Source: https://github.com/melandlabs/openloomi/tree/main/benchmark/continual-learning-bench/skills/continual-learn
Command: npx skills add https://github.com/melandlabs/openloomi --skill continual-learn

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve?

AI agents lose context between turns in long iterative tasks, repeating mistakes and forgetting feedback. This Skill enforces durable, file-based memory so lessons, hypotheses, and state survive across an entire session.

Core Features & Use Cases

  • Mandatory file persistence: Requires an actual filesystem write to MENTAL_MODEL.md before every response, even when the final answer must follow a strict JSON schema.
  • Structured note-taking: Records goals, feedback, durable lessons, commands tried, failures, hypotheses, and the next concrete action.
  • Hygiene rules: Keeps the file compact, prunes stale notes, and forbids secrets or large data dumps.
  • Use Case: During a multi-round debugging or benchmarking session, the agent reads its notes at the start of each turn and updates them before replying, so it never re-runs a failed command or loses track of a confirmed hypothesis.

Quick Start

Ask the agent to work through an iterative debugging task while maintaining its findings in a MENTAL_MODEL.md file in the workspace.

Frequently Asked Questions about continual-learn

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I make an AI agent remember lessons across turns?▼

Use a durable workspace file like MENTAL_MODEL.md that the agent reads at the start of each turn and updates before every response. This Skill mandates real filesystem writes so memory persists outside the conversation context.

What should an agent write into a persistent memory file?▼

Record the task goal, current plan, feedback received, durable lessons, commands tried, observed failures, active hypotheses, and the next concrete action. Keep notes terse and prune stale entries to keep the file compact.

Does the memory file write conflict with strict JSON output requirements?▼

No. The file write happens before the final response, and the final response still obeys the requested schema exactly. The two operations are sequential, not mixed.

When should I use file-based agent memory instead of conversation context?▼

Use it for long iterative tasks like debugging, benchmarking, or feedback loops where context windows overflow or sessions restart. It is less necessary for single-turn questions with no follow-up state.

What should never be stored in the MENTAL_MODEL.md file?▼

Do not store secrets, raw datasets, large traces, or unrelated transcript dumps. The file is a compact scratchpad for high-signal notes, not a log archive.

Related Skills

Explore More Software Engineering Skills→
idontcarlidontcarl

playwright-interactive

Maintains a persistent Playwright session to debug web and Electron apps across iterations.

Community
Advanced
az9713az9713

handoff

Summarize the current conversation into a handoff document saved to a temporary path.

Community
Intermediate
alex-voloshin-devalex-voloshin-dev
2

ralph

Execute a Read-Act-Learn-Feedback loop until a binary success condition is met.

Community
Advanced
christian-bromannchristian-bromann
3

langchain-vector-stores

Index and retrieve embedded documents across multiple LangChain vector stores.

Community
Advanced