Getting it into your agent
It runs from inside its repository, so the clone comes first — what it calls does not travel with the file alone.
git clone --depth 1 https://github.com/ShaftHQ/SHAFT_ENGINEnpx agentmods add skills/shafthq/shaft_engine/self-improveWrote 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/shafthq/shaft_engine/self-improve)<a href="https://agentmods.dev/skills/shafthq/shaft_engine/self-improve"><img src="https://agentmods.dev/badge/skills/shafthq/shaft_engine/self-improve/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/shafthq/shaft_engine/self-improve"><img src="https://agentmods.dev/badge/skills/shafthq/shaft_engine/self-improve.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 | $0.00032 | $0.00901 |
| Opus 5 | $0.00016 | $0.00451 |
| Sonnet 5 | $0.00006 | $0.00180 |
| Haiku 4.5 | $0.00003 | $0.00090 |
Grade A, and why
self-improve 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 — 90 lines — stays where its author put it; the contents beside it link to each section on GitHub.
self-improve — ChaosEngine learning & adapting
Lean CE-native skill informed by Task Observer methodology (Eoghan Henn / rebelytics, CC BY 4.0 — see LICENSE and UPSTREAM.md). Not a blind clone.
When
- Primary: root-owned Learning Session after confirmed delivery.
- Secondary: explicit operator request mid-session.
- Not: every casual turn — keep always-on cost low.
Dual track
- Harness — skills, hooks, MemPalace, Graphify, Headroom, installer/doctor.
- Product — enhancements for the product under development (queued issues).
Details: references/observation-taxonomy.md. Product track: references/product-track.md (product issues/evals + ChaosGauge link; CLI/doctor/silent-verify over essay tickets). Activation: references/activation.md. Adopt/reject research: references/research-adopt-reject.md. Roadmap: ../../references/self-improve-master-plan.md. Research/explore isolation (harness): ../../references/context-firewall.md.
How (wraps learning.py)
- Classify each finding (harness vs product; category allow-list).
- Write minimal fields only:
category,title,lesson,proposedChange,benefit,estimatedTokens. - Queue through
learning.pyso privacy gates + GitHub filing invariants hold. - Never auto-install skill patches; stage proposals for human/CI review.
- "Nothing durable" is a valid outcome.
Example:
cat > /tmp/learning-candidate.json <<'EOF'
{
"category": "tooling",
"title": "Doctor headroom fix-next missing",
"lesson": "Operators lacked a single install command after pin landed",
"proposedChange": "Surface uv tool install pin in doctor fix-next",
"benefit": "Faster Headroom provisioning on adopter hosts",
"estimatedTokens": 120
}
EOF
python3 .chaos-engine/learning.py queue \
--state .chaos-engine-state/learning \
--upstream Owner/ExampleRepo \
--candidate /tmp/learning-candidate.json
What ships with it
6 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 Changed · +25 lines 0ae43b1ff610
- yesterday First seen · 65 lines · 32 tokens per session scan A 26c1f9c4ff63
self-improve is a skill published in the GitHub repository ShaftHQ/SHAFT_ENGINE (408 stars, last pushed yesterday), licensed MIT. It adds 32 tokens to every session and 901 once invoked, about $0.0002 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 2026-09-07.
Other skills, from other repositories
flowproof-config
Configure flowproof's SAP GUI, Fiori, and AI authoring credentials by walking the user through flowproof config sap / flowproof config fiori / flowproof config ai. Use when the user wants to set up, change, or check their SAP/Fiori login or model authoring key, or when a flow run fails because SAPUSER, FIORIPASSWORD…
gws-gmail
Gmail: Send, read, and manage email.
dws
A command-line guide for DingTalk, a workplace platform with chat, documents, calendars, approvals, tasks, and other business tools.
web-demo-video
A process for making product-demo videos from a real website or web app by displaying it in an embedded browser page, controlling it with real mouse events, capturing the frames, and combining them with ffmpeg.
kakaotalk-cli
A command-line tool workflow for reading and sending KakaoTalk messages, the chat service widely used in South Korea.
bookmarks-management
Manages X/Twitter bookmarks — organize by category with auto-tagging, clear all bookmarks in bulk, and export/scrape bookmarks with full metadata. Supports keyword-based categorization, CSV and JSON export, keep-filters, and folder creation (Premium). Use when users want to export, organize, or clear their X bookmarks.