AI coding agents are good at solving problems—but bad at remembering them. You spend time debugging an obscure issue. The agent eventually figures it out. The session ends. When the same issue shows up later, the agent has no memory of the solution, and you start over from scratch. This leads to repeated explanations, duplicated effort, and wasted time.
Skiller adds persistent memory to AI coding agents. When an agent uncovers something non-obvious—such as a debugging technique, workaround, or project-specific pattern—Skiller saves it as a reusable skill. That skill is automatically loaded the next time a similar situation appears. No re-explaining. No rediscovering the same fix.
How It Works • Detects valuable, hard-won knowledge during problem-solving • Stores that knowledge as a named skill • Automatically applies relevant skills in future sessions
Each solved problem makes the agent better prepared for the next one.
Why It Matters • Builds institutional memory over time • Reduces repeated debugging and context-setting • Helps teams share solutions implicitly, not manually
Think of it as a checklist that writes itself—so your future self (and your teammates) don’t have to relearn lessons you already paid for.
Works with Claude Code, GitHub Copilot, Cursor, and other agents that support the Agent Skills standard.
Fork of Claudeception, extended for cross-agent compatibility.
Skiller uses the Agent Skills standard — structured markdown files in your project. Install is the same for every agent: download the skill file into your repo.
Claude Code natively supports skills via SKILL.md files in .claude/skills/.
Project-level (recommended — shared with your team via git):
mkdir -p .claude/skills/skiller
curl -sL https://raw.githubusercontent.com/markdav-is/Skiller/main/SKILL.md -o .claude/skills/skiller/SKILL.mdUser-level (applies to all your projects):
mkdir -p ~/.claude/skills/skiller
curl -sL https://raw.githubusercontent.com/markdav-is/Skiller/main/SKILL.md -o ~/.claude/skills/skiller/SKILL.mdThat's it. Claude Code discovers skills automatically. Say "save this as a skill" or "what did we learn?" in any session to trigger extraction.
Copilot supports three integration points: agent skills (automatic), a custom agent (@skiller), and a prompt file (/skiller).
macOS / Linux:
mkdir -p .github/skills/skiller .github/agents .github/prompts
curl -sL https://raw.githubusercontent.com/markdav-is/Skiller/main/SKILL.md -o .github/skills/skiller/SKILL.md
curl -sL https://raw.githubusercontent.com/markdav-is/Skiller/main/.github/agents/skiller.agent.md -o .github/agents/skiller.agent.md
curl -sL https://raw.githubusercontent.com/markdav-is/Skiller/main/.github/prompts/skiller.prompt.md -o .github/prompts/skiller.prompt.mdWindows (PowerShell):
New-Item -ItemType Directory -Force -Path .github\skills\skiller, .github\agents, .github\prompts
Invoke-WebRequest https://raw.githubusercontent.com/markdav-is/Skiller/main/SKILL.md -OutFile .github\skills\skiller\SKILL.md
Invoke-WebRequest https://raw.githubusercontent.com/markdav-is/Skiller/main/.github/agents/skiller.agent.md -OutFile .github\agents\skiller.agent.md
Invoke-WebRequest https://raw.githubusercontent.com/markdav-is/Skiller/main/.github/prompts/skiller.prompt.md -OutFile .github\prompts\skiller.prompt.md- Automatic: Agent mode loads the skill when context matches
- Custom agent: Type
@skillerin Copilot Chat - Prompt command: Type
/skillerfor a session retrospective - Manual: Say "save this as a skill" or "what did we learn?"
Any agent that reads .github/skills/ can use Skiller:
mkdir -p .github/skills/skiller
curl -sL https://raw.githubusercontent.com/markdav-is/Skiller/main/SKILL.md -o .github/skills/skiller/SKILL.mdThe skill activates automatically when the agent:
- Just completed debugging and discovered a non-obvious solution
- Found a workaround through investigation or trial-and-error
- Resolved an error where the root cause wasn't immediately apparent
- Learned project-specific patterns or configurations through investigation
- Completed any task where the solution required meaningful discovery
Trigger a learning retrospective:
/skiller
or
@skiller review this session
Or explicitly request skill extraction in any agent:
Save what we just learned as a skill
Not every task produces a skill. It only extracts knowledge that required actual discovery (not just reading docs), will help with future tasks, has clear trigger conditions, and has been verified to work.
Extracted skills go to .github/skills/[skill-name]/SKILL.md by default — the cross-agent standard path.
| Path | Scope | Agents |
|---|---|---|
.claude/skills/ |
Project | Claude Code |
.github/skills/ |
Project | Copilot, Cursor, and others |
~/.claude/skills/ |
User | Claude Code |
~/.copilot/skills/ |
User | Copilot |
Modern AI coding agents have a native skills system. At startup, they load skill names and descriptions (about 100 tokens each). When you're working, the agent matches your current context against those descriptions and pulls in relevant skills.
But this retrieval system can be written to, not just read from. So when Skiller notices extractable knowledge, it writes a new skill with a description optimized for future retrieval.
The description matters a lot. "Helps with database problems" won't match anything useful. "Fix for PrismaClientKnownRequestError in serverless" will match when someone hits that error.
Agents use three-level progressive loading to keep things efficient:
- Discovery: Name + description only (~100 tokens per skill) — loaded at startup
- Activation: Full SKILL.md instructions — loaded when context matches
- Execution: Supporting files (scripts, templates) — loaded only as needed
This means you can have dozens of skills installed without overwhelming the context window.
More on the skills architecture from Anthropic, Claude Code docs, and VS Code.
Extracted skills are markdown files with YAML frontmatter:
---
name: [descriptive-kebab-case-name]
description: |
[Precise description including: (1) exact use cases, (2) trigger
conditions like specific error messages or symptoms, (3) what
problem this solves.]
author: Skiller
version: 1.0.0
date: [YYYY-MM-DD]
---
# [Skill Name]
## Problem
[What this skill solves]
## Context / Trigger Conditions
[Exact error messages, symptoms, scenarios]
## Solution
[Step-by-step fix]
## Verification
[How to confirm it worked]See resources/skill-template.md for the full template.
The skill is picky about what it extracts. If something is just a documentation lookup, or only useful for this one case, or hasn't actually been tested, it won't create a skill. Would this actually help someone who hits this problem in six months? If not, no skill.
Fork of Claudeception, extended for cross-agent compatibility.
The idea comes from academic work on skill libraries for AI agents.
Voyager (Wang et al., 2023) showed that game-playing agents can build up libraries of reusable skills over time, and that this helps them avoid re-learning things they already figured out. CASCADE (2024) introduced "meta-skills" (skills for acquiring skills), which is what this is. SEAgent (2025) showed agents can learn new software environments through trial and error, which inspired the retrospective feature. Reflexion (Shinn et al., 2023) showed that self-reflection helps.
Agents that persist what they learn do better than agents that start fresh.
SKILL.md # Main skill definition (all agents)
.github/
agents/skiller.agent.md # Copilot custom agent
prompts/skiller.prompt.md # Copilot prompt file (/skiller)
copilot-instructions.md # Copilot repo instructions
resources/
skill-template.md # Template for new skills
CLAUDE.md # Claude Code guidance
COPILOT.md # Copilot guidance
VSCODE.md # VS Code guidance
VISUAL-STUDIO.md # Visual Studio guidance
Contributions welcome. Fork, make changes, submit a PR.
MIT