Getting it into your agent
One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.
npx skills add topprismdata/cultivating-ml-agent --skill self-evolving-skillsgit clone --depth 1 https://github.com/topprismdata/cultivating-ml-agentWrote this? Show the measurements
A badge with what this costs and how it scanned, read live from this page, so it follows the numbers instead of freezing them. Markdown for a README, HTML for a documentation site or a project page.
[](https://agentmods.dev/skills/topprismdata/cultivating-ml-agent/self-evolving-skills)<a href="https://agentmods.dev/skills/topprismdata/cultivating-ml-agent/self-evolving-skills"><img src="https://agentmods.dev/badge/skills/topprismdata/cultivating-ml-agent/self-evolving-skills/github.svg" alt="Measured on agentmods" height="20"></a>Or the 80×15 button, for a site that already has a row of RSS and ATOM ones. Only the verdict fits; the numbers stay here.
<a href="https://agentmods.dev/skills/topprismdata/cultivating-ml-agent/self-evolving-skills"><img src="https://agentmods.dev/badge/skills/topprismdata/cultivating-ml-agent/self-evolving-skills.svg" alt="Reviewed on agentmods" width="80" height="20"></a>What it costs to keep this loaded
Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.
| Model | Per session | Once invoked |
|---|---|---|
| Fable 5.1 | $0.00060 | $0.00833 |
| Opus 5 | $0.00030 | $0.00417 |
| Sonnet 5 | $0.00012 | $0.00167 |
| Haiku 4.5 | $0.00006 | $0.00083 |
Grade A, and why
self-evolving-skills scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 7d ago.
A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.
Nothing flagged
None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.
How it starts
The opening of the file, as written. The whole thing — 110 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Self-Evolving Skills (Voyager-Style)
Context
Static skill libraries grow only when humans add entries. Voyager (Wang et al. NeurIPS 2023) showed agents that auto-extract and register new skills grow 100+ skills without manual work. agy verified this as P1 for ML agents: 43+ skills → 100+ with quality gates.
The core insight: every experiment is a candidate skill; quality gating prevents library pollution.
Guidance
Extract from Text/Conversation
from framework.src.skill_evolution import SkillExtractor
ext = SkillExtractor()
candidates = ext.extract_from_text("""
We should always check data leakage before training.
Never use future information in features.
Make sure to use walk-forward validation for time series.
""")
# → 3 candidates: 1 skill, 1 anti-pattern, 1 principle
Extract from Experiment Log
# 假设实验记录在 EXPERIMENTS.md
candidates = ext.extract_from_experiments_md(experiments_md_content)
# 自动按 importance 排序, top 20 入选
Validate Before Registration
from framework.src.skill_evolution import SkillValidator
val = SkillValidator()
for c in candidates:
result = val.validate(c, existing_descriptions=existing)
# verdict: APPROVE / REVISE / REJECT
# issues: ['Too similar...', 'Missing frontmatter...']
# similarity_to_existing: 0.0 - 1.0
Register to MCP Library
from framework.src.skill_evolution import SkillRegistry
registry = SkillRegistry(skills_dir="skills/examples")
for c in candidates:
result = registry.register(c)
if result.verdict.value == "approve":
print(f"Registered: {c.name}")
# 写入 skills/examples/{name}/SKILL.md
Auto-Approve (Production Danger)
# 生产环境慎用 — 应该人工 review
registry.register(candidate, auto_approve=True)
Why This Matters
Without self-evolving:
- Skills grow slowly (43+ → manual additions)
- Lessons forgotten after session ends
- Same mistakes repeated
With self-evolving:
- 100+ skills in 6 months vs 5
- Lessons persist with proper templates
- Quality gates prevent bloat
What this file has done since we first saw it
Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.
- 7d ago First seen · 110 lines · 60 tokens per session scan A 550162106eb3
self-evolving-skills is a skill published in the GitHub repository topprismdata/cultivating-ml-agent (5 stars, last pushed 13d ago), licensed MIT. It adds 60 tokens to every session and 833 once invoked, about $0.0003 per session on Opus 5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-09-03.
Other skills, from other repositories
huggingface-hub
Hugging Face Hub CLI (hf) — search, download, and upload models and datasets, manage repos, query datasets with SQL, deploy inference endpoints, manage Spaces and buckets.
writing
A writing guide for turning verified facts and calculations into finished text for a specific audience. It follows the requested language, structure, and length.
regex-mastery
Use this skill when writing regular expressions, debugging pattern matching,optimizing regex performance, or implementing text validation. Triggers on regex, regular expressions, pattern matching, lookahead, lookbehind, named groups, capture groups, backreferences, and any task requiring text pattern matching.
ws-ckpt
A Linux-only tool for creating and managing workspace snapshots. A snapshot is a saved view of files that can later be compared, restored, listed, or deleted.
food-order
Reorder previous Foodora orders, preview cart contents, and track delivery ETA/status with ordercli. Use when the user wants to reorder food, check delivery status, or browse recent Foodora order history. Never confirm an order without explicit user approval.
mnemon
Persistent memory for MiniMax Code. Recall durable context, store important facts and decisions, and link related memories with the mnemon CLI.