Skill Evolution
Myrm’s Skill Evolution engine automatically turns your successful interactions into reusable skills. Unlike simple prompt storage, the engine runs a full pipeline: trace analysis → extraction → screening → sandbox validation → LLM review → human approval via GUI.
How it works
- Trace Analysis — After each conversation, the
TraceAnalyzer extracts execution trajectories using 5-level progressive disclosure, identifying patterns worth turning into skills.
- Structured Extraction — The
StructuredExtractor uses an LLM to capture skill definitions with confidence scores, ensuring only high-quality patterns are promoted.
- Automatic Screening — Low-quality or duplicate extractions are filtered before reaching you.
- Sandbox Validation — Candidate skills are executed in a safe sandbox environment to verify they actually work.
- LLM Review — An automated reviewer checks for edge cases, security issues, and correctness.
- Human Approval — Nothing deploys without your say-so. Review in the Pending Evolutions Dashboard.
Review pending skills
- Go to Settings → Skills → Pending Evolutions.
- Browse proposals in List or Grid view.
- For each proposal:
- Approve — Deploy the skill immediately.
- Reject — Discard the proposal.
- Revise — Edit the skill in a Monaco diff editor before approving.
- Approved skills become available to the Agent in future conversations.
Variant testing
When Myrm detects a skill could be improved, it generates variants and runs A/B comparisons. You can review variant results in the Shadow GUI panel before promoting a variant to production.
Version history & rollback
Every skill change is tracked:
- History panel — Full evolution timeline with diffs.
- Versions panel — Compare any two versions side by side.
- One-click rollback — Revert to any previous version instantly.
Batch optimization
For large skill libraries:
- Go to Settings → Skills → Batch Optimization.
- Click Run Optimization — the engine evaluates all skills for improvement opportunities.
- Review the batch snapshot and approve/reject changes individually.
- Mid-batch rollback is supported if something goes wrong.
Background review
Myrm runs 4 types of background async tasks during idle sessions:
- Silent skill extraction from successful interactions
- Compression of verbose skills for lower token usage
- Frustration detection → automatic fix variant generation
- Pruning of low-quality skills
All background changes still require your approval before deployment.
Per-session skill scope
When your Agent Profile has many skills bound, not every skill is relevant to every conversation. Per-session skill scoping lets you visually select which skills to load for the current chat, reducing token consumption and improving AI focus.
- Open any chat in Agent mode.
- Click the sparkle icon (✦) next to the message input — the Session Skills toggle.
- A popover lists all globally enabled skills. Uncheck skills you don’t need for this conversation.
- Your selection is persisted server-side and survives page reloads, conversation compaction, and even conversation forks.
- To restore all skills, click Clear Override at the top of the popover.
When you switch to a different Agent, the session skill override is automatically cleared — so you always start fresh with the new Agent’s full skill set.
Built-in prebuilt skills
Beyond self-evolving skills, Myrm ships with 36+ prebuilt skills ready to use out of the box. Notable examples:
- Data Analysis — A six-phase workflow (Ingest → Clean → Explore → Analyze → Visualize → Report) with automatic data quality checks and risk alerts. Supports CSV, JSON, Excel, and database queries.
- Data Analysis Pipeline — Multi-agent version: four specialized roles (Collector → Analyst → Visualizer → Reporter) collaborate in parallel. Outputs HTML reports, PDF, Jupyter Notebooks, Markdown, or PPT outlines.
- Deep Research — Multi-step web research with source verification.
- Code Review — Automated code review with security and performance checks.
- Frontend Development, Test-Driven Development, Systematic Debugging — Development workflow skills.
Prebuilt skills are automatically synced on startup and can be customized, disabled, or reset to defaults via the Settings GUI.
Matplotlib charts render inline with zero-copy vault:// pointers — no Jupyter server required. This gives you Jupyter-quality visualization without the infrastructure overhead.
Key numbers
| Metric | Value |
|---|
| Evolution engine tests | 385 passed |
| Safety layers | 15 (Harness 6 + Server 6 + Service 3) |
| Patch formats | 3 (fuzzy match with 8-level progressive + Unicode normalization) |
| Routing strategies | 4 (including Correction Propagation) |
| Background task types | 4 |
Skill marketplace
Myrm integrates 7 discovery sources in a unified GUI — search, filter by tags, sort by stars/downloads, preview with security scan, and install with one click:
| Source | Coverage |
|---|
| ClawHub | Largest open skill marketplace |
| GitHub | Public skill repositories |
| LobeHub | LobeChat plugin ecosystem |
| ModelScope | 80K+ Chinese AI community models/skills |
| Aliyun | Alibaba Cloud Agent Explorer |
| skills.sh | Curated skill directory |
| Prebuilt | 36+ built-in skills shipped with Myrm |
Additional marketplace capabilities:
- URL Import — Install from any URL (GitHub repo, ZIP archive, etc.)
- Auto-update — Checks for upstream updates with quarantine-based safe update flow
- Skill Sync — Push/pull synchronization across devices
- Curator — Automatic lifecycle management (stale detection, archive, cleanup)
- Consolidation — Merges duplicate or overlapping skills automatically
- Export with Redaction — Download skills as ZIP with automatic secret stripping
vs competitors
| Capability | Myrm | Hermes |
|---|
| Auto-extract skills | Full pipeline | Single CLI command |
| Sandbox validation | Yes | No |
| GUI review dashboard | Yes | No (auto-applies) |
| Confidence scoring | Yes | No |
| Variant A/B testing | Yes | No |
| Version rollback GUI | Yes | CLI only |
| Background optimization | Yes (4 types) | Basic |
| Cross-agent experience sharing | Yes | No |
| One-click extract from chat | Yes (message-level button) | No |
| Skill retrieval layers | 3-layer (semantic+env hybrid, deterministic scorer, progressive loading) | 2-layer (semantic search, playbook synthesis) |
| Execution transparency | Full (skill badges + reason + tool chain + error diagnostics) | Partial (tool list only) |
| Marketplace sources | 7 integrated sources | 3 (ClawHub + GitHub + custom tap) |
| Marketplace UI | Full GUI (search/filter/sort/install) | CLI only |
| Auto lifecycle management | Curator (stale/archive/cleanup via Settings + Smart Prune — not chat-only text suggestions) | Manual |
| Skill consolidation | Auto-merge duplicates | No |
| Per-session skill scope | Visual toggle (pick skills per chat, ~1800 tokens/turn saved) | No |