Install
$ agentstack add skill-leo-atienza-atlas-claude-self-evolve ✓ scanned · ✓ verified — works with Claude Code, Cursor, and more.
Security review
✓ PassedNo issues found. Passed automated security review. · v0.1.0 How review works →
- ✓ Prompt-injection patterns
- ✓ Secret / credential exfiltration
- ✓ Dangerous shell & filesystem operations
- ✓ Untrusted network calls
- ✓ Known-malicious package signatures
What it can access
- ✓ Network access No
- ✓ Filesystem access No
- ✓ Shell / process execution No
- ✓ Environment & secrets No
- ✓ Dynamic code execution No
From automated source analysis of v0.1.0. “Used” means the capability is present in the source — more access means more to trust, not that it’s unsafe.
About
Self-Evolve — Autonomous Capability Growth
The system's meta-skill. When Claude detects it lacks a capability needed for the current task, this skill provides the decision framework to fill the gap — by creating skills, adding MCP servers, or recommending community tools.
Gap Detection Signals
Activate self-evolve when ANY of these occur:
- Missing tool: A task requires an integration or tool that isn't available (e.g., "connect to Jira" but no Jira MCP)
- Repeated pattern: The same multi-step workflow has been executed 3+ times across sessions — check
~/.claude/logs/failures.jsonland memoryINDEX.mdfor patterns - Explicit request: User says "can you learn to do X", "there should be a skill for this", "add a tool for X"
- Tool failure: A tool call fails because the capability doesn't exist, not because of a bug
- Inefficiency signal: Claude catches itself doing >5 manual steps that could be automated
Decision Tree
┌─ Is the gap about a TOOL or INTEGRATION?
│ YES → Search MCP registry
│ │ ├─ mcp-find "{keyword}" → found free server?
│ │ │ YES → Show user: name, capabilities, source → on approval: mcp-add → register in REGISTRY.md
│ │ │ NO → Can a skill wrap existing CLI tools?
│ │ │ YES → Create skill (see Skill Creation Protocol)
│ │ │ NO → Inform user, suggest alternatives or manual install
│ │
│ NO ↓
│
├─ Is the gap about a REPEATED WORKFLOW?
│ YES → Create skill using Skill Creation Protocol
│ │ 1. Capture the workflow steps from conversation history
│ │ 2. Abstract into reusable instructions
│ │ 3. Write SKILL.md with proper frontmatter
│ │ 4. Register in REGISTRY.md
│ │
│ NO ↓
│
├─ Is the gap about DOMAIN KNOWLEDGE?
│ YES → Search Context7 first (no skill needed)
│ │ If recurring → create lightweight reference skill
│ │
│ NO ↓
│
└─ Is the gap about AGENT COORDINATION?
YES → Suggest /flow:smart-swarm (SK-039) for complex multi-agent tasks
MCP Server Addition Protocol
- Search: Use
mcp-findwith relevant keywords
`` mcp-find "jira" / "slack" / "database" / "email" / etc. ``
- Evaluate: Check each result for:
- Free tier available (no API key required, or user has the key)
- Relevant to the task at hand
- Not redundant with existing MCP servers (check REGISTRY.md MCP Servers section)
- Propose: Show the user:
``` Found MCP server: {name} Capabilities: {list of tools it provides} Source: {catalog entry} Requires: {any API keys or auth needed}
Want me to add it? ```
- Install (on user approval):
`` mcp-add "{server-name}" ``
- Verify: Make a test call to confirm tools are working
- Register: Add entry to
~/.claude/skills/REGISTRY.mdunder MCP Servers:
`` | MCP-NNN | {name} | {purpose} | {access method} | ``
- Log: Update
~/.claude/SYSTEM_CHANGELOG.md
Skill Creation Protocol
- Template: Read
~/.claude/skills/self-evolve/templates/skill-template.md
- Write: Create
~/.claude/skills/{name}/SKILL.mdwith:
- Proper frontmatter (name, description with trigger conditions)
auto-generated: truetag in frontmatter- Clear "When to Use" section
- Step-by-step "Process" section
- Expected "Output Format"
- Register: Append to
~/.claude/skills/REGISTRY.mdStandalone Skills table:
`` | SK-NNN | {name} | {purpose} | skills/{name}/SKILL.md | `` Use the next available SK number (check last entry).
- Changelog: Update
~/.claude/SYSTEM_CHANGELOG.md
- Verify: Read back the skill to confirm it's well-formed
Safety Rules — Non-Negotiable
- NEVER install paid services or servers requiring paid API keys without explicit user confirmation
- NEVER add MCP servers that require authentication tokens the user hasn't provided
- ALWAYS verify free tier availability before suggesting
- ALWAYS show the user what will be added and get confirmation for MCP servers
- ALWAYS check REGISTRY.md for duplicates before creating (grep for similar names/purposes)
- MAX 1 auto-created skill per session unless user explicitly requests more
- Skills are local files only — no external downloads or binary installations
- If uncertain whether something is free, ASK the user
Examples
Example 1: Missing MCP Server
User: "Can you check my Asana tasks?"
Gap: No Asana MCP server connected
Action: mcp-find "asana" → found → propose → user approves → mcp-add → register
Example 2: Repeated Pattern
Pattern: Claude has formatted markdown tables 4 times this session using the same 6-step process
Action: Create skill "markdown-table-formatter" with the 6 steps abstracted
Example 3: Domain Knowledge
User: "How do I use the Stripe API?"
Action: Context7 search first → if sufficient, use inline → if recurring need, create reference skill
Source & license
This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.
- Author: Leo-Atienza
- Source: Leo-Atienza/atlas-claude
- License: MIT
Install and usage instructions live in the source repository linked above.
Reviews
No reviews yet — be the first.
Write a review
Versions
- v0.1.0 Imported from the upstream source.