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 agentmods add skills/jansenanalytics/claudex/self-skillnpx skills add JansenAnalytics/claudex --skill self-skillgit clone --depth 1 https://github.com/JansenAnalytics/claudexWrote 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/jansenanalytics/claudex/self-skill)<a href="https://agentmods.dev/skills/jansenanalytics/claudex/self-skill"><img src="https://agentmods.dev/badge/skills/jansenanalytics/claudex/self-skill.svg" alt="Measured on agentmods" 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 | $0.00056 | $0.00904 |
| Opus 5 | $0.00028 | $0.00452 |
| Sonnet 5 | $0.00011 | $0.00181 |
| Haiku 4.5 | $0.00006 | $0.00090 |
Grade A, and why
self-skill 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 today.
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 — 103 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Self-Skill — Autonomous Skill Creation
When This Runs
Automatically — after every non-trivial task, run the 5-question filter mentally:
- Will this come up again?
- Did I have to look something up or think hard?
- Did I write more than 20 lines of non-trivial code?
- Does it involve a specific tool/API with non-obvious usage?
- Would the user benefit from knowing how to trigger this?
2+ yes answers → create or propose a skill.
See references/heuristics.md for the full decision framework.
Improve an Existing Skill (patch-on-friction)
Creating skills is only half the loop — maintaining them matters just as much. Right after you used a skill this turn, run a quick friction check:
- Missing steps — did you have to do something the skill didn't mention?
- Outdated / wrong — a stale command, wrong path, or wrong flag?
- Boilerplate — did you hand-write something the skill could have provided?
- Unclear — did you misread an instruction that should be clearer?
2+ yes → patch it now. Edit that skill's SKILL.md with the smallest diff that fixes
it, add/refresh a patched: YYYY-MM-DD line in the frontmatter, and keep it to one skill per
turn (don't go on a refactor spree mid-task). Don't wait to be asked — an unmaintained skill
is a liability.
The self-edit gate (a PostToolUse hook) automatically audits every SKILL.md you edit —
frontmatter, ≤15 KB size, external-path leaks, and secrets — and warns you if a patch breaks
something. So patch freely; the gate is your safety net.
Scaffold a New Skill
bash ${CLAUDE_SKILLS_DIR:-$HOME/.claude-agent/.claude/skills}/self-skill/scripts/draft-skill.sh <name> \
--desc "what it does in plain language" \
--resources "scripts,references"
This creates the full directory structure and pre-fills SKILL.md.
Full Skill Creation Checklist
When creating a skill (auto or on request):
[ ] 1. Run draft-skill.sh to scaffold the structure
[ ] 2. Write SKILL.md — description, triggers, quick reference
[ ] 3. Write scripts/ — working, tested, standalone
[ ] 4. Write references/ — detailed docs for complex topics
[ ] 5. Test all scripts before documenting them
[ ] 6. Write <skill-name>-guide.md in workspace (plain language, user-facing)
[ ] 7. Send guide as Telegram file attachment
[ ] 8. Paste raw guide markdown in chat — ONE continuous code block (split max 2 if too long)
[ ] 9. Commit: git add -A && git commit -m "feat: add <name> skill"
What ships with it
2 files beside SKILL.md in the same directory: the scripts, references and assets a skill reads on demand. Not counted in the per-session cost; read them before you install if any of them is executable.
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.
- today First seen · 103 lines · 56 tokens per session scan A cc47c4783e39
self-skill is a skill published in the GitHub repository JansenAnalytics/claudex (5 stars, last pushed 2mo ago), licensed MIT. It adds 56 tokens to every session and 904 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
decision-heuristics
当用户在重大选择上纠结（换工作/买房/搬城/合伙/结婚）、列了利弊表还是拿不定主意时调用。 核心理念: 无法决定就答否; 三个重大决定(住哪/和谁/做什么)值得花一两年; 两个均等选择选短期更痛苦的路。 不适用于: 日常琐碎选择（晚饭吃什么）、信息查询。 Triggers: 纠结/拿不定主意/要不要/该不该/利弊/decision/hesitate/should I.
skill-authoring
Author SKILL.md skills: frontmatter, validator limits, structure.
pentest-playbook
7-phase pentest pipeline from passive recon to exploitation.
torchforge-rl-training
Provides guidance for PyTorch-native agentic RL using torchforge, Meta's library separating infra from algorithms. Use when you want clean RL abstractions, easy algorithm experimentation, or scalable training with Monarch and TorchTitan.
recon-playbook
Use when starting or restructuring an authorized external web and API assessment.
llamaguard
Meta's 7-8B specialized moderation model for LLM input/output filtering. 6 safety categories - violence/hate, sexual content, weapons, substances, self-harm, criminal planning. 94-95% accuracy. Deploy with vLLM, HuggingFace, Sagemaker. Integrates with NeMo Guardrails.