Engineering AgentMemoryAutomation

Self-Improving Agent

Curate auto-memory into durable project knowledge, promote patterns to rules.

$skillfish add ~skills/self-improving-agent

Real Examples Get Started Models & Context

Real Conversations

See how the Self-Improving Agent skill responds to actual requests. These are representative examples of what your AI agent will produce.

You

I just installed self-improving-agent. Can you explain how the memory to CLAUDE.md promotion lifecycle works? What's automatic vs. what needs my approval?

Agent

Here's the full lifecycle — nothing gets promoted without your approval:

1. Claude discovers a pattern
   └── Auto-writes to MEMORY.md (automatic)

2. Pattern recurs 2–3 sessions
   └── /si:review flags it as a promotion candidate

3. YOU review candidates and approve
   └── /si:promote writes it to CLAUDE.md or .claude/rules/

4. MEMORY.md entry removed
   └── Frees space for new learnings (after your approval)

The key distinction:

MEMORY.md — Claude's scratchpad: "I noticed this project uses pnpm"
CLAUDE.md — enforced instruction: "Use pnpm, not npm. Always."

Promoted rules load in full every session. Memory entries are truncated at 200 lines and may not load if the file gets full.

Automatic: capturing patterns, detecting candidates, flagging stale entries, the error-capture hook

Requires your approval: every promotion, every deletion from MEMORY.md

Run /si:status to see current memory health, or /si:review for a full audit with candidates queued. Want to run a status check now?

You

A new developer is joining next week and will be using Claude Code with our project. I want to make sure MEMORY.md has clean, accurate information — not stale debugging notes from 3 months ago.

Agent

Smart prep — running a full review before onboarding avoids the new hire learning outdated patterns.

Memory Health:

MEMORY.md:     162/200 lines (81%)  Warning
Topic files:   2 (debugging.md, api-patterns.md)
CLAUDE.md:     87 lines

Stale Entries Found (7):

Line 34: References src/api/v1/users.ts — file deleted in November refactor
Line 67: "Use npm run migrate:old" — script removed; correct command is pnpm db:migrate
Lines 89–91: Three entries about old Webpack config — project is on Vite now
Line 134: "Redux is installed but not used yet" — Redux removed in December

Promotion Candidates (3 — new hire should see these as rules, not notes):

"Always run pnpm lint before committing" — seen 6 sessions
"Auth token in Authorization: Bearer header, not a cookie" — seen 4 sessions
"Use pnpm db:seed:test for test data, not manual SQL inserts" — seen 3 sessions

Recommendation:

Delete 7 stale entries (frees 18 lines)
Promote 3 candidates to CLAUDE.md — the new hire sees them as enforced rules from day 1
MEMORY.md drops to ~144/200 lines — healthy

Delete stale entries and start the promotion workflow?

Get Started

How to install and use this skill in your preferred environment.

Skills are designed for AI coding agents (Claude Code, Cursor, Windsurf) and IDE-based workflows where the agent can read files, run scripts, and act on your codebase. Web-based AI can use the knowledge and frameworks, but won't have tool access.

Models & Context

Which AI models and context windows work best with this skill.

Recommended Models

Best

Claude Opus 4 Claude Sonnet 4 GPT-4.1 Gemini 2.5 Pro Grok 3 Kimi K2

Good

Claude Haiku 4.5 GPT-4.1 mini Gemini 2.5 Flash Grok 3 mini

Larger models produce more detailed, production-ready outputs.

Context Window

This skill's SKILL.md is typically 3–10 KB — fits in any modern context window.

8K Skill only

32K+ Skill + conversation

100K+ Skill + references + codebase

All current frontier models (Claude, GPT, Gemini) support 100K+ context. Use the full window for complex multi-service work.

Pro tips for best results

1

Be specific

Include numbers — users, budget, RPS — so the skill can size the architecture.

2

Share constraints

Compliance needs, team size, and existing stack all improve the output.

3

Iterate

Start with a high-level design, then ask follow-ups for IaC, cost analysis, or security review.

4

Combine skills

Pair with companion skills below for end-to-end coverage.

Ready to try Self-Improving Agent?

Install the skill and start getting expert-level guidance in your workflow — any agent, any IDE.

$skillfish add ~skills/self-improving-agent

← Browse all 169 skills