awskills

awskills

Docs · Source · git clone https://github.com/Aitherium/awskills · The Aither World

The Aither World is an operating system for agents — a Linux you can hand to one, the runtimes it works in, and the tools it works with. awnix is the Linux underneath it; awskills is one of its 71 bricks — each installs on its own, runs offline, and needs no account. 5 more are in development and 15 are planned.

Start here: Copy one .md into your agent's skills dir and tell it to use that skill.

Use with Hermes: hermes skills tap add Aitherium/awskills (then hermes skills install Aitherium/awskills/<skill>).

Learn to run coding agents at program scale — from measured telemetry, not vibes.
Free, MIT-licensed. The operating doctrine, plus 40+ battle-tested skills.

Most advice about working with AI coding agents is somebody's feeling. This repo is one
operator's logs: 27,939 prompts across 3,183 sessions over 210 days, re-measured on a
disjoint 34-day window (5,244 human prompts, 9,715 machine-written agent dispatches), then
distilled into rules you can install into your own setup in about ten minutes.

Everything here came out of running a real platform — 260 services, ~166 containers, three
repos — with agents doing most of the typing. The skills are the parts that survived.

📖 The Developer Codex — start here if you want the whole thing

Read it online → · Read it in the repo →

The skills below are procedures. The Codex is the doctrine underneath them — what
makes a codebase run by agents get harder to break over time instead of quietly
rotting while every dashboard stays green.

| | |
|---|---|
| The path — the Aither World Guide | From "what is an agent?" to a terminal that answers you: install, a local brain, your first agent, packs, your own hardware, many agents on one repo. Mirrored from awknowledge. |
| The twenty-nine laws | Every one was a real failure first. Grouped into Enforcement · Silence · Adoption · Deployment · Trust · Configuration · Scaling. |

The laws in one line each:

Enforcement — a rule nothing asserts is a suggestion · make it a check, not a
ticket · watch your gate fail · mutate the test, not just the code

Silence — design for the silence · a check that cannot run must not pass · the
symptom names the innocent · a checker in the wrong place found nothing · detection
without delivery is not detection

Adoption — a gate that floods gets switched off · open green, ratchet down ·
measure it again

Deployment — written is not deployed · you wrote it, that does not mean it ships ·
generate, never copy · the defect lives in the union

Trust — fail closed, then prove the happy path · never trust the caller for an
authorization decision · count the closure, not the edge · a platform you cannot ship
to is not supported

Configuration — a cap that does not survive a restart is not a cap · coordinated
decisions need a channel humans can read

Scaling — a mixture-of-experts model has two budgets, not one · the always-on
context is a budget · a setting in a scope nobody reads does not exist · a hook on
every call spends a latency budget · one sanction file, never scattered · a guard in
one direction is half a guard · portable is files and presets, never credentials

Why this exists. The failures that cost days do not throw. They return 200, render
correctly, log nothing, and leave the container healthy — a missing thing is
indistinguishable from a thing nobody wanted. Twenty-nine laws is what it took to make a
system able to notice.

🧠 Start here — code like this

| Skill | What it teaches |
|---|---|
| awknowledge | The doctrine. 13 measured rules for running an agent at program scale: prompt shape, live-proof gates, plan documents as files not modes, persistent memory with an index, when to orchestrate vs stay solo, when to compact, how to route models. Installs itself into your rules/memory/plans directories without overwriting anything. |
| ramble-driven-development | The prompt-shape law. Median human prompt: 56 characters. But the 5.9% over 1,000 chars carry 78% of everything typed. Ramble to load intent, poke to steer, and put the precision in the harness — 90% of machine-written dispatches name a file path against 6% of human ones. Includes how to mine your transcripts, and the two filters that otherwise inflate your median by 33x. |

git clone https://github.com/Aitherium/awskills
cd awskills
bash scripts/install-awskills.sh          # Windows: pwsh -File scripts/Install-AitherSkills.ps1

Then tell your agent: "use the awknowledge skill".

The one rule to take away if you read nothing else: the careful, fully-specified
prompt still has to exist — you just shouldn't be the one typing it. Put your standards
in a rules file once, add a gate that can fail, and your prompts collapse to
"get it done." You don't type your standards. You install them.

⚠️ Doing this on a bare setup with no gates produces confident garbage at speed. Build
the gate first. Both skills say so up front.

🚀 Then — run the agent on your own hardware, free

If you have a computer and access to an AI agent, you can run your own agent on your own
hardware, for free, today. Install the pack and open the front door:

git clone https://github.com/Aitherium/awskills
cd awskills
bash scripts/install-awskills.sh          # Windows: pwsh -File scripts/Install-AitherSkills.ps1

Then tell your agent: "use the aither-start skill".

| Skill | What it gets you |
|---|---|
| aither-start | The front door. Zero → working agent on your own machine: detect the hardware, install the toolkit, run a model that actually fits, wire it into the agent you already use. Every step ends in a check that can fail. |
| local-inference | A language model on your box for $0 — pick Ollama / llama.cpp / vLLM for the machine you have, size the model so it doesn't OOM, serve it OpenAI-compatible, prove it with a real round-trip. Includes tool-calling setup and the failure modes that look like success. |
| install-skills | Install this pack into any agent — Claude Code, OpenClaw, Hermes, Cursor, Goose, Codex, Gemini CLI. Explains the two layouts (SKILL.md folder vs flat slash-command) and why the wrong one makes an agent "not see" skills that are right there. |
| repo-is-not-a-runtime | The doctrine underneath the cleanup. Two rules: a repo is a source artifact, not a runtime; and every ephemeral an agent creates needs an owner, a TTL and a reaper. Measured: a checkout that was 0.4% .git and 99.6% runtime data + agent debris. Ships repo-hygiene-audit.sh — a gate that fails, because doctrine without a gate is a wish. |
| agent-disk-hygiene | Your agents are quietly eating your disk. A real checkout grew to 1.15 TB — of which .git was 4.7 GB; the rest was agent debris, led by 295 GB of abandoned worktrees. How to reap them safely, and the git log --branches trap that makes every worktree look dirty forever so nothing ever gets cleaned. Ships agent-worktree-reaper.sh. |
| concurrent-safe-git | You do not have this working tree to yourself. The moment two agents (or an agent and a cron loop) share a checkout, ordinary git turns destructive: a bare git commit ships whatever somebody else staged. Two real incidents — a reset --hard that wiped a session's uncommitted work and re-deployed an already-fixed data leak, and a 10-line fix that committed 270 lines. The pathspec commit form, the four commands to never run, and the git update-index --refresh trick for a merge git only thinks is unsafe (measured: 7 blocked files, 1 real, all 7 kept their content). |
| docker-wsl2-disk-reclaim | Your drive is full, you pruned 200GB, and nothing changed. Docker Desktop's VHDX never shrinks — its data disk is mounted without discard. The three-layer accounting model (VHDX 2.0TB ≥ ext4 1.6TB ≥ docker system df 634GB), why fstrim before compaction is mandatory, and why moving the file to another drive doesn't help. Ships docker-wsl2-reclaim.sh. |
| docker-wsl2-build-safety | Stop bulk image builds from crashing Docker Desktop's WSL2 backend and taking every container down — plus the day-distribution check that tells a VM storage collapse apart from a failing disk. They look identical: 44k disk I/O errors in an hour, Device offlined, "Docker Desktop is unable to start". One is a config problem; the other is a hardware purchase. |
| docker-network-ops | Container DNS lies to you. Docker's embedded resolver 127.0.0.11 is a goroutine inside dockerd, not a kernel service — measured failing 38–65% of queries with clean 2s timeouts, sustained, while conntrack sat at 11% and Udp InErrors=0. Why no kernel counter will ever show it, why musl/glibc/nginx each fail differently in the same container, the silent --bind-interfaces race that leaves dnsmasq Up (healthy) serving nothing, and the six measurement traps that each produced a confident wrong answer. Ships docker-net-doctor.py. |
| openclaw | Install OpenClaw, point it at your model instead of a paid API, and connect the AitherOS toolset with one command (aither integrate openclaw). |
| hermes-agent | Install Nous Research's Hermes — self-improving, persistent memory, cron automation — on your own inference. Includes the two config shapes that silently do nothing if you get them wrong. |
| tau | Install Tau — a minimalist Python terminal coding agent — on your own model via ~/.tau/catalog.toml. Includes the folder-only skill layout tau enforces (bare .md is silently skipped) and the /skill:name invocation. |
| deer-flow | Run DeerFlow — ByteDance's LangGraph super-agent harness for long autonomous research/coding runs — on your own endpoint, with MCP servers and per-tool timeouts. |
| ods | Stand up ODS: one installer turns a PC/Mac/Linux box into a private AI server — inference, chat UI, voice, agents, workflows, RAG, image gen, all in Docker, CPU fallback included. |
| ship-an-app-free | Idea → working app → public URL, on free tiers only. GitHub Pages + Actions, Cloudflare Workers when you need a backend. No credit card, no server to rent. |
| github-actions-image-pipeline | Build once, deploy many. GitHub Actions has no cross-workflow dependency, so two workflows that both build an image build it twice on every push — and an ML base is a 40–56 minute build, cold. The division-of-labor fix: one workflow builds, deployers docker buildx imagetools create-retag the :latest in seconds. Plus the 10GB cache that evicts your base, the disk-reclaim an ML base needs on a hosted runner, the actions: read permission github-script silently needs, and the never-exercised-path chain — every step a workflow never ran fails on a real latent bug the first time it runs. |

Everything above is free. No paid API key is required anywhere in that path.

Layout

awskills/
├── codex/      # the Developer Codex — the reading path + the twenty-nine laws
├── skills/     # skill files (.md) — install into any agent, see `install-skills`
├── scripts/    # standalone scripts you can run directly
├── tools/      # MCP tools / CLI utilities
└── packs/      # themed bundles (docker, deploy, security)

Installing into your agent

The Agent Skills standard wants skills/<name>/SKILL.md; Claude Code
slash commands want a flat .claude/commands/<name>.md. This repo ships flat files and the
installer converts per target — so you don't have to care:

bash scripts/install-awskills.sh --list                 # what's detected
bash scripts/install-awskills.sh --dry-run              # show, write nothing
bash scripts/install-awskills.sh --target openclaw      # just one agent
bash scripts/install-awskills.sh --only local-inference # just one skill

Nothing is overwritten without --force, so re-running is safe. Full per-agent path table in
install-skills.

Skills

🐳 recover-docker — un-wedge Docker Desktop's WSL2 engine

Docker Desktop on Windows wedges its WSL2 Linux engine: the docker API returns 500 Internal Server Error, or docker stop/recreate dies with tried to kill container, but did not receive an exit event (common on nvidia-runtime / GPU containers). The GUI looks healthy; the daemon is dead.

scripts/Recover-Docker.ps1 does a complete teardown in the right order — kill the UI + backends → wsl --shutdown → reap vmmem/wslservice zombies → bounce the Windows services → cold-start Docker Desktop → wait for the engine → clean dead containers + restart exited ones. No reboot, container volumes intact, healthy in ~30–60s.

Run it directly:

# one-shot recovery (run elevated for the service bounce)
pwsh -File scripts/Recover-Docker.ps1

# 30s watchdog — auto-recovers on failure
pwsh -File scripts/Recover-Docker.ps1 -Monitor

As a Claude Code skill: copy skills/recover-docker.md into your project's .claude/commands/ and scripts/Recover-Docker.ps1 somewhere on disk, then run /recover-docker (or /recover-docker --monitor). The agent detects the wedge, runs recovery, verifies with docker version / docker ps, and reports which containers came back.

📖 Background: Self-Healing Docker: One Command to Un-Wedge the WSL2 Engine

🛡️ moat-guard — keep private code out of your public package

Open-core release hygiene. Three parameterized tools (nothing project-specific — every rule is a flag) plus a /moat-guard skill that drives them:

| Tool | Job |
|------|-----|
| tools/check_package_leaks.py | Pre-publish gate. Inspect a built wheel/sdist; fail (non-zero exit) if it bundles forbidden files/imports or is missing a required keystone. Drop it in CI before twine upload. |
| tools/find_leaky_releases.py | Audit what already shipped. List index versions below a cutoff and, with --verify, download each wheel to prove the leak. Prints the exact yank checklist (indexes have no yank API). |
| tools/purge_public_leaks.sh | Scrub the public GitHub surface. Delete pre-cutoff releases + tags and filter a leaked file out of the repo's entire history (mirror force-push). Dry-run by default. |

# CI gate: fail the build if it bundles secrets or imports an internal package
python tools/check_package_leaks.py dist/mypkg-2.0.0-py3-none-any.whl \
  --forbid-path '*/secrets*.py' --forbid-import mycorp_internal \
  --require-file '*/licensing.py'

# Audit a published project and prove which versions leak
python tools/find_leaky_releases.py mypkg --cutoff 2.0.0 --verify \
  --forbid-path '*/nanogpt.py'

# Plan a purge (dry-run), then execute once you've read it
bash tools/purge_public_leaks.sh --repo me/mypkg --keep-from 2.0.0 --leak-path src/secret.py
bash tools/purge_public_leaks.sh --repo me/mypkg --keep-from 2.0.0 --leak-path src/secret.py \
  --execute --rewrite-history          # irreversible — breaks pinned installs & forks

As a Claude Code skill: copy skills/moat-guard.md into .claude/commands/ and run /moat-guard check (pre-publish), /moat-guard find (audit), or /moat-guard purge (destructive — always dry-runs first, confirms before force-pushing).

⚠️ Purging shrinks exposure but cannot un-distribute what already shipped. If a removed file carried a secret, rotate it.

🗜️ model-quantization — shrink an LLM to 4-bit, locally and free

Make a model fit where bf16 won't — on a smaller GPU, or beside another model on
the same card. tools/quantize_model.py drives
AutoRound and bakes in the gotchas that
otherwise produce a broken or un-loadable artifact.

The default is RTN (round-to-nearest, --iters 0): weight-only, no
calibration data, no forward pass, ~<2 GB peak VRAM, ~1 minute. It runs on your
CPU + GPU together (host-RAM offload) — $0, fully local, and enough for
most models. Calibrated AWQ (--iters > 0) is higher quality but needs a real
GPU, and is refused on architectures whose calibrated path crashes (per-layer
head dims) with a clear steer back to RTN.

# Preview the plan — no weight load, no GPU, no write
python tools/quantize_model.py google/gemma-3-12b-it --dry-run

# RTN 4-bit, local + free (keeps lm_head + multimodal projectors in bf16)
python tools/quantize_model.py google/gemma-3-12b-it -o ./gemma-3-12b-it-awq

# Calibrated AWQ on a GPU (refused on het-head models — use RTN there)
python tools/quantize_model.py mistralai/Mistral-7B-Instruct-v0.3 \
  -o ./mistral-7b-awq --iters 200 --nsamples 128

What it gets right for you: uses AutoRound, not llm-compressor (which silently
downgrades transformers); keeps lm_head + vision/audio projectors in bf16
(vLLM loaders require it); exports compressed-tensors so un-quantized modules
stay plain; and detects heterogeneous head dims to avoid the calibrated-mode
crash. Serve the result with vllm serve <outdir> --quantization awq_marlin.

As a Claude Code skill: copy skills/model-quantization.md into
.claude/commands/ and tools/quantize_model.py onto disk, then run
/model-quantization <model-id> -o <outdir>. The agent dry-runs first, runs RTN
by default, and reports the output path + serve command. Needs
pip install auto-round torch transformers.

🗜️ aither-headroom — cut agent token cost with reversible context compression

Agents burn most of their tokens re-sending bulky context — verbose JSON tool output, retrieved
docs, file dumps. headroom (headroom-ai) crushes that
content with a SmartCrusher pipeline — measured ~46% token savings on an 87 KB tool blob —
while protecting conversation/user text so answers don't degrade. Two ways in: an automatic
pre-send hook at the single LLM chokepoint (flag-gated, graceful no-op — if the sidecar is off,
calls just proceed uncompressed), and agent-callable tools (headroom_compress(content),
headroom_stats()) from the free headroom tool pack.

export AITHER_HEADROOM_ENABLED=true              # turn on the automatic pre-send hook
curl http://127.0.0.1:8788/health                # sidecar healthy? → {"ok":true,"headroom":"0.25.0"}
# prove the savings on a realistic payload before trusting the ratio

Give an adk agent the tools self-service with apply_pack_self("headroom") (free, no entitlement).
Present bulky context as the content blob (conversation text is protected and barely compresses;
below ~800 chars it no-ops). See skills/aither-headroom.md.

🔄 resume-all — bring back every Claude Code session you lost

A reboot, a crash, or a closed terminal and your Claude Code conversations are gone — not
deleted, just unfindable. You reopen N terminals, cd into each project, run claude,
then /resume and squint at a list of UUIDs trying to remember which was which.

scripts/Resume-ClaudeSessions.ps1 reads Claude Code's
own session journals (~/.claude/projects/<encoded-cwd>/<session-id>.jsonl), recovers each
session's AI title, last prompt, working directory, git branch and last-active time, lets
you pick, and reopens them — each in its own terminal tab or tmux window. Read-only against
your history; it never mutates the journals.

# interactive picker — choose which to bring back
pwsh -File scripts/Resume-ClaudeSessions.ps1

# reopen the most-recent session for every project directory, no prompt
pwsh -File scripts/Resume-ClaudeSessions.ps1 -PerDir -All

# only sessions matching some text, from the last day
pwsh -File scripts/Resume-ClaudeSessions.ps1 -Filter payments -LookbackHours 24

# over SSH? resume into tmux — the windows survive a disconnect
pwsh -File scripts/Resume-ClaudeSessions.ps1 -Tmux -Select 1,3

Cross-platform (PowerShell 7): Windows Terminal tabs on Windows, tmux windows anywhere,
Terminal.app on macOS — and if none of those exist it prints the commands rather than
pretending it launched something. -DryRun prints the resolved session ids to stdout.

⚠️ The gotcha this exists to solve: sub-agent sidechains (agent-*.jsonl) and workflow
journals are rewritten constantly, so by write-time they crowd out your real sessions.
A naive "most recently modified N journals" scan surfaces almost none of them. This filters
to genuine top-level conversations before truncating the scan window.

⚠️ Second gotcha, fixed in the engine: resumed tabs used to come up fully monochrome
whenever you launched them from inside a Claude Code session. Claude Code exports
NO_COLOR=1 to its subprocesses so tool output comes back clean — and the terminal this
engine spawns is one of those subprocesses, so the new window, every shell in it, and every
claude inside those shells inherited it. It looks exactly like a terminal-theme problem and
is not one. The engine now scrubs the variable per tab. If you edit the launch command, keep
that scrub — and delete the variable rather than blanking it, since the check is
!("NO_COLOR" in process.env) (presence, not value), so NO_COLOR="" still kills colour.

As a Claude Code skill: copy skills/resume-all.md into .claude/commands/ and the script
onto disk, then run /resume-all (or /resume-all all, or /resume-all <filter text>).

🕸️ omninode-node — join the OmniNode P2P inference mesh in one command

OmniNode Protocol (by SUM-INNOVATION) is a
trustless, peer-to-peer network that pools ordinary machines into a fabric big enough to run models no
single device could hold — any device with a chip can become a node. Standing one up shouldn't be a
ten-step wiki page.

scripts/omninode-node-up.sh takes a fresh machine (Linux / macOS /
Windows-WSL2) from nothing installed → a live, discoverable node: detect hardware → install Rust if
missing → clone + build omni-node → verify two peers discover each other over libp2p/mDNS (or --listen
to run a persistent node). If awdk is present it can also
adk mesh onboard the node so your agents use it — one motion, not two projects.

bash scripts/omninode-node-up.sh          # build + self-verify P2P discovery
bash scripts/omninode-node-up.sh --listen # run a persistent mesh node
bash scripts/omninode-node-up.sh --adk    # + enroll into AitherMesh for adk agents

Verified end-to-end on a 12-core Linux box: clone → build → NODE OK, P2P discovery live. See
skills/omninode-node.md. No credentials, no account, no central server.

🧩 The Aither substrate — set up and use awdk, awnode, AitherConnect, AitherZero & AitherMesh

Five skills for the coherent substrate the OmniNode node plugs into. Each is a "set it up, then use
it" guide grounded in real commands — standing up compute and having your agents use it is one
motion, not five projects.

| Skill | What it sets up |
|-------|-----------------|
| awdk | The agent toolkit — pip install awdk → adk onboard --quick → adk run. Your model, your loop, your data on your box; manage from the portal. |
| aither-discord-agent | Deploy any awdk agent as a Discord bot with one automated onboarding command — adk onboard --discord installs your pack, validates the bot token live, prints the invite link, verifies identity/tools, and launches. Every DM/@mention runs your agent's own loop. No paid tier needed (hand-rolled fallback). |
| awnode | The body — a local MCP server (adk mcp node) exposing GPU, local inference, ComfyUI, and files to agents; or bootstrap the box as a full inference node. |
| aitherconnect | The seam — adk connect / adk mesh onboard (--headscale behind NAT) to wire a machine, agent, and browser into AitherOS and the mesh. |
| aitherzero | The provisioner — one config.psd1 + bootstrap.ps1 to stand up bare-metal/on-prem/cloud/hybrid, with a generated-from-inventory config editor and az_* agent tools. |
| aithermesh | The fabric — one playbook (Invoke-AitherPlaybook deploy-mesh-agent) to create a private WireGuard mesh, join nodes to the overlay, and deploy agents onto them; nodes defined in config/nodes.yaml. |
| bonsai-27b | A model to run on a node — PrismML's 1-bit Bonsai-27B (Q1_0, 3.8 GB) served on a plain CPU box via the PrismML llama.cpp fork; a 27B model on a laptop. |
| graph-rag-agent | Knowledge on tap — adk ingest a folder/codebase into a knowledge graph, then an agent (adk create-app + recall/search_knowledge) that answers from your material. Local-first graph RAG, no separate vector DB. |
| graph-a-repo | The operational runbook — graph a repo OR a KB end-to-end and prove both halves: the embedder returns the right-dimension vector (rag_verify_embedder, 768 for nomic-embed-text) AND retrieval returns the ingested content (rag_verify_retrieval — an empty graph answers 200 with zero hits). Picks code (CodeRankEmbed + codegraph_*) vs prose (nomic-embed-text) embedder; bakes in the traps (code≠text vector spaces, fleet-vector parity, adk chat not adk query). Front door for the graphrag toolpack. |
| aither-code-intelligence | The operational layer — how to run prospector+codegraph+headroom for real and prove they still work. A 60-second health check, nine silent failure modes with the symptom you actually see (stale paths, orphaned embeddings, mount traps, self-healers killing slow indexers, OOMKilled=false hiding a real OOM), and why hybrid search needs Reciprocal Rank Fusion (measured F1 0.083 → 0.293 → 0.470). Every bug in it reported healthy while broken. |
| aither-codegraph | The what — a call-graph-aware code index. adk run auto-indexes a Python project; agents gain code_search/code_context (symbols, signatures, callers, callees, blast radius). Fleet path: the managed CodeGraph service + codegraph_* MCP tools. |
| aither-prospector | The where — a semantic file-explorer. apply_pack_self("prospector") → map_build(repo) then map_localize("where is auth?") returns the dirs to search first, so CodeGraph/grep only look where it matters. Free pack, dependency-free builder. |
| aither-agent-notebook | The reviewable unit of work — turn "build X" into a runnable .anb Agent Notebook of typed cells (plan / prompt / tool_call / agent_delegate / human checkpoint / result). adk notebook plan "…" → run → status → export to Jupyter; every run is cost-tracked and can be replayed/diffed against a baseline. Six agent tools (notebook_*) proxy the Genesis /notebooks API. The durable counterpart to a one-shot adk forge. |

More skills (drop into .claude/commands/)

Generic, project-agnostic slash commands — pure prompt-skills, no code or dependencies:

| Skill | What it does |
|-------|--------------|
| secretguard | Scan git history for leaked secrets with gitleaks; purge a file from history (filter-repo) or allowlist a false positive. Never echoes secret values. |
| security-audit | Code + dependency + config audit against the OWASP Top 10 — injection, crypto, access control, secret exposure — with severity-ranked findings. |
| dependencies | Audit / update / prune dependencies and check licenses across pip, npm/yarn, and Docker base images (pip-audit, npm audit, safety). |
| performance | Profile and optimize: cProfile/memory_profiler, hotspot hunting, caching, N+1 queries, algorithmic complexity. Measure first. |
| refactor | Apply clean-code refactors — extract method, replace conditionals with polymorphism, simplify nested logic — without changing behavior. |
| compare-versions | Diff two versions of a file/commit/release: structural + behavioral changes, breaking-change risk, and a migration checklist. |

Every skill in the pack

The tables above are the guided tour. This one is the inventory: every published
skill, one line each, generated from the skill's own frontmatter so it cannot fall
behind the directory.

| Skill | What it is |
|---|---|
| acp-drive | Drive any external ACP agent from your agent. |
| acp-serve | Expose an AitherOS agent to ACP editors. |
| adk-deepseek-coder | Fill-in-the-middle, repo packing, and a reward that can fail. |
| adk-harnesses | One shell that drives every coding shell. |
| adk-unsloth | Local training and serving, wired to your agent. |
| adversarial-verification | Stop grading your own homework. |
| agent-disk-hygiene | Your coding agents are quietly eating your disk. |
| agent-integrations | Integrate awdk and AitherOS with mainstream coding agent harnesses (Claude Code, Cursor, Aider, Cline, Codex, Hermes, Roo Code) — both as a CLIENT of cloud providers (OpenRouter, DeepSeek, Anthropic, Moonshot) and as a … |
| agent-operating-doctrine | The standing instructions, tool order and hook wiring one team runs its coding agents under -- no false blockers, a 150-word report cap, search-then-graph, awgit in a shared worktree, and the PreToolUse/Stop gates that … |
| agents-everywhere | Turn an agent idea into a verified, shareable Aitherium tool across awdk, awsh, MCP, WebMCP, PWA, Codex, Claude Code, and other agent CLIs. |
| aither-agent-notebook | Turn 'build X' into a reviewable, re-runnable plan. |
| aither-code-intelligence | Stand up agent code-search, and prove it actually works. |
| aither-codegraph | A call-graph-aware code index your agents can query. |
| aither-discord-agent | Deploy your awdk agent as a Discord bot (automated onboarding). |
| aither-headroom | Cut agent token cost with reversible context compression. |
| aither-powered-coding | Set up any coding agent (Claude Code, Cursor, Aider, Cline) to use AitherOS as its brain — with automatic model switching, 1272 MCP tools, managed agent dispatch, and transparent failover between DeepSeek/Kimi/local … |
| aither-prospector | A semantic file-explorer that tells agents WHERE to look. |
| aither-retrieval | Aither Retrieval — one surface, every modality. |
| aither-start | Zero-to-working-agent on your own machine, for someone who has never written code. |
| aitherconnect | Awconnect — wire your machine, agent, and browser into AitherOS. |
| aithermesh | Stand up a private mesh and onboard nodes + agents to it. |
| aithernode | Awnode — turn a machine's hardware into something your agents can use. |
| aitherzero | Provision any machine from one config file and a library of numbered PowerShell automation-scripts. |
| awconnect | Wire your machine, agent, and browser into AitherOS. |
| awdk | Run your own AI agent, on your machine, in three commands. |
| awembed | Train an embedding model that knows your corpus. |
| awgit-claude-code | The awgit workflow for Claude Code / an Aither adk agent — how to work when every commit is captured as a semantic edit-op: what to check after a commit, how to read the attribution ledger, how to teleport deltas to a … |
| awgit-setup | Set up awgit (Aither World-Graph git) on a machine — the semantic version-control layer that turns every git commit into a function-level edit-op with a verified GitHub identity, a durable attribution record, and … |
| awknowledge | Install a measured operating doctrine for running an agentic coding tool at program scale — prompt shape, live-proof gates, plan documents, persistent memory, delegation, compaction and model routing. |
| bonsai-27b | Run a 27B model in ~3.8 GB on a plain CPU box. |
| bonsai2-27b | Serve PrismML's Bonsai 2 27B (Ternary-Bonsai-2-27B, PTQ1_0 5.9 GB or PQ2_0 7.2 GB) as an OpenAI-compatible endpoint with the PrismML llama.cpp fork. |
| character-forge | Character Forge — consistent characters, animation, talking avatars. |
| claude-code-hooks-fast | Make Claude Code hooks cheap, especially on Windows, where every shell-form hook quietly starts four processes on every tool call. |
| claude-model-switcher | Switch Claude Code between DeepSeek (native 1M context), Kimi, local AitherOS models, and stock Anthropic — one command, no manual config. |
| code-like-david | Renamed to awknowledge. |
| compare-versions | Compare two versions of a file, commit, or release and report structural and behavioral diffs |
| concurrent-safe-git | Commit safely in a worktree you do NOT have to yourself — several agents, a teammate, and a maintenance loop all editing and committing at once. |
| debt-ledger | Keep one master tech-debt ledger that actually gets written to — severity tables, collision-free ids via the bundled next_debt_id.py, and the discipline that debt found is debt recorded in the same turn. |
| decision-cards | Decision Cards — structured asks that actually reach the owner. |
| deer-flow | Run ByteDance's DeerFlow — a LangGraph-based super-agent harness for autonomous research, coding and content creation with sub-agents, persistent memory and sandboxed execution. |
| dependencies | Manage project dependencies, check for updates, and scan for vulnerabilities |
| dgx-spark-deepseek-v4 | A 284B model at ~30 tok/s on one desk machine. |
| do-my-taxes | Do My Taxes — personal tax preparation workflow. |
| docker-network-ops | Diagnose container DNS and networking without fooling yourself. |
| docker-wsl2-build-safety | Stop bulk builds from killing Docker Desktop (and don't blame the disk). |
| docker-wsl2-disk-reclaim | Your drive is full and docker system df is lying to you. |
| event-loop-debugger | Detect, pin, and fix asyncio event-loop stalls in Python services — a synchronous call or CPU-bound work blocking the loop — including on the free-threaded python3.14t build where py-spy and gdb fail. |
| fail-closed-authz | Review auth, tenancy, entitlement and cache-sharing code against six fail-closed defect classes that ruff and mypy cannot see — four checked mechanically by the bundled security_lint.py, three semantic ones you read … |
| finetune-that-improves | Train your own model without making it worse. |
| floci-local-cloud | Exercise cloud automation (AWS/Azure/GCP/OCI) against local MIT-licensed emulators instead of real accounts — no credentials, no credit burn, no live instance at risk. |
| github-actions-image-pipeline | Run GitHub Actions image pipelines without rebuilding the same images on every push, or paying an hour-long cold build for a typo. |
| github-agentic-c2 | Use GitHub as the control plane for autonomous agents — issues as the task queue, PRs as the review gate, git as the conflict-resolution protocol, Actions on self-hosted runners as your own compute, and Projects as the … |
| github-artifact-quota | The storage limit that silently freezes your deploys. |
| github-runner-fleet | Stand up, verify and repair a fleet of self-hosted GitHub Actions runners on one Windows host. |
| gpu-vram-attribution | When the platform cannot tell you who is using the GPU. |
| graph-a-repo | Turn a repository OR a knowledge base into a graph an agent can actually answer from — pick the right embedder, stand it up, ingest, and PROVE both halves work (the vectors are the right dimension AND retrieval returns … |
| graph-rag-agent | Build a knowledge graph and an agent that owns it. |
| hermes-agent | Install Nous Research's Hermes agent and run it on your own model — a self-improving agent with persistent memory, autonomous skill creation, cron automation and multi-platform messaging. |
| hetzner-fleet | Provision bare metal and VPS directly into your mesh from Hetzner Cloud. |
| install-skills | Install this skill pack into whatever agent you already use — Claude Code, OpenClaw, Hermes, Cursor, Goose, Codex, Gemini CLI, or any Agent-Skills-compatible client. |
| lan-latency | Fix a wired desktop or node that pings its own router at tens of milliseconds with jitter and zero NIC errors. |
| local-image-generation | Give an agent the ability to draw — generating images on the machine it is already running on, via ComfyUI, Sana, SD.Next/A1111 or an in-browser model. |
| local-inference | Run a language model on your own hardware for free — pick the right backend (Ollama, llama.cpp, or vLLM) for the machine you actually have, download and quantize a model that fits in memory, serve it on an … |
| local-research | Delegate codebase and web exploration to the local aither-local MCP server so frontier context stays small. |
| moat-guard | Open-core release hygiene — verify a build doesn't leak private code before publishing, and purge leaks that already shipped (GitHub releases/tags/history + index yank list) |
| model-fitness-benchmark | Decide whether a candidate model is fit to run an agent stack, using hard oracles instead of vibes or vendor tables. |
| model-quantization | Quantize an LLM to 4-bit with AutoRound — RTN runs free on local CPU+GPU; keeps lm_head/projectors in bf16 and dodges the new-architecture crashes |
| nvidia-cuda-mcp | Give your agent current CUDA knowledge. |
| ods | Stand up ODS (Osmantic Deployment System) — one installer that turns a PC, Mac, or Linux box into a private AI server with local LLM inference, a chat UI, voice, agents, workflow automation, RAG and image generation, … |
| omninode-node | Join the OmniNode P2P inference mesh in one command. |
| openclaw | Install OpenClaw (the local-first personal AI assistant) and wire it to your own hardware — point it at a local model instead of a paid API, connect the AitherOS toolset over MCP with one command, and install this skill … |
| performance | Analyze and optimize application performance, identify bottlenecks |
| pooled-inference | Run a model no single box can hold, across every tier you own. |
| prove-it-live | The verification standard — green tests, a 200, and "deployed" are not proof. |
| ps5-bridge | Let an agent play a PS5 game you own: convert your own dumped PS5 executable with AnyPS5, start it with its Aither bridge switched on, and drive it with the ps5-bridge tool pack (watch frames, hold buttons, measure fps). |
| purecss-art | Generate realistic illustrations hand-coded entirely in HTML + CSS (the Diana Smith / "cyanharlow" technique — absolute-positioned divs, gradients, layered box-shadow, clip-path, transforms; no images, no SVG, no … |
| ramble-driven-development | Stop writing careful prompts. |
| recover-docker | Hard-recover Docker Desktop from the WSL2 wedge (API 500 / "did not receive an exit event") and restart exited containers |
| refactor | Refactor code to improve structure, readability, and maintainability |
| repo-intake | Turn "I found this GitHub repo — integrate/adapt/adopt it" into a decided, recorded outcome instead of another unread folder in a research folder. |
| repo-is-not-a-runtime | The architecture doctrine for agent-driven repos. |
| repo-to-website | Turn any GitHub repo into a real website on GitHub Pages — a proper landing page with your README's content, live at a URL, instead of a raw README. |
| resume-all | Reopen your killed Claude Code sessions — snapshot before a reboot, restore after, or pick from every project and resume as terminal tabs or tmux windows |
| safe-bulk-rename | Rename a token across hundreds or thousands of files in this shared worktree — a package, a directory, a product name — without corrupting load-bearing strings, without committing a peer's in-flight work, and without … |
| secretguard | Scan for leaked secrets and purge them from git history |
| security-audit | Perform comprehensive security audits on code, dependencies, and configurations |
| ship-an-app-free | Take an idea to a working app at a public URL using only free tiers — GitHub Pages for the site, GitHub Actions for the build, and a free serverless backend when the app needs one. |
| split-inference | Shard ONE model across MULTIPLE machines' GPUs with llama.cpp RPC — build with -DGGML_RPC=ON, start rpc-server on the backend hosts, launch the main server with --rpc, and PROVE the split is real instead of a silent … |
| tau | Install and run Tau, a minimalist terminal coding agent in Python — point it at your own local model via ~/.tau/catalog.toml instead of a paid API, install this skill pack into its skills directory, and use the /skill: … |
| website-as-code | Deploy a website as code — GitHub Pages frontend + Cloudflare Tunnel backend on any machine + an automated never-show-a-raw-error fallback. |
| windows-kernel-leak | Diagnose a Windows box where every program starts slowly (hooks time out, a bare cmd /c exit takes seconds) by measuring the kernel pool first. |
| worktrees | Make, list and retire agent git worktrees through awgit wt — never raw git worktree add/remove. |
| xy-charts | Render data charts (line/scatter/bar/heatmap/pie/histogram/density) to PNG/SVG/JPEG/WebP/PDF or self-contained interactive HTML via the xy engine — for metrics, benchmarks, report figures, dashboard panels, and any … |

Standalone tools

CLI utilities you can run directly — all parameterized, no AitherOS dependency:

| Tool | What it does |
|------|--------------|
| tools/check_exports.py | Validate a Python package: __all__ entries that don't resolve (ghost exports), __version__ vs pyproject.toml drift, and orphan modules nothing imports. Stdlib-only. |
| tools/validate_compose_ports.py | Lint docker-compose for host-port collisions (across one or many -f files), malformed mappings, and unpublished container ports. --strict to fail CI. |
| tools/Backup-DockerVolumes.ps1 | Snapshot Docker named volumes → timestamped .tgz + manifest.json, via a throwaway Alpine container. -Pattern/-SkipPattern/-DryRun, auto-prunes old snapshots. |
| tools/quantize_model.py | Quantize an LLM to 4-bit with AutoRound. RTN runs free on local CPU+GPU (--iters 0, default); keeps lm_head/multimodal projectors in bf16, exports compressed-tensors, and refuses calibrated mode on het-head architectures that would crash. --dry-run to preview. |

PowerShell dev utilities (tools/powershell/)

Standalone PS7 helpers — no module, no setup, just pwsh -File:

| Script | What it does |
|--------|--------------|
| Invoke-FileGrep.ps1 | Recursive regex content search with context lines. |
| Invoke-BulkReplace.ps1 | Regex find/replace across globs, with -DryRun and backreferences. |
| Invoke-FileDiff.ps1 | Unified diff of two files (or inline strings). |
| Invoke-FileSplice.ps1 | Surgically replace a line range in a text file. |
| New-GitBranch.ps1 | Create a branch with a conventional prefix (configurable). |
| New-GitCommit.ps1 | Stage + commit with Conventional Commits validation. |

More coming

More agent-ops glue is on the way. Star the repo to follow along — and PRs/issues welcome.

License

MIT © Aitherium. Use it, fork it, ship it.

The aw family

Standalone tools that share one idea: replace something you would otherwise have to trust with something you can check.

Each installs on its own, works offline, and needs no account.

| | instead of trusting | you check |
|---|---|---|
| awdk | a framework's idea of how your agents should run | one loop you can read, pointed at a backend you already pay for |
| awskills (you are here) | that an agent knows your procedure | the procedure written down, versioned, and loadable by any agent |
| awpack | that the pack you want shipped inside somebody's SDK, under whatever licence that SDK happens to carry | the pack as its own versioned artifact, with its own licence, that any agent runtime can install |
| awm | that memory stayed in its lane | tenant:user:project scopes, so a write cannot cross a boundary |
| awdesk | that the agent is somewhere behind a browser tab | a tray icon, a face on your desktop, and the decision card that pops when it needs you |
| awnode | a vendor's cloud with every prompt | a local gateway routing to backends you chose |
| awgraph | that grep found everything | an AST + tree-sitter call graph an agent can traverse |
| awgit | that no one else is editing this file | a lease, refused at commit time if you do not hold it |
| awdelphi | one agent's confident take on a decision | the round trace, the anonymity, and who dissents |
| awclassify | a filename, a folder, or whoever last touched it | doc_type, visibility, audience and topics, with the evidence lines that decided each |
| awdecide | a hosted classifier's probability that never learns whether it was right | the decision, its probability, and the calibration curve from your own resolved outcomes |
| awtoll | that your tooling is saving you context | the measured token cost of each tool call, and what the alternative cost |
| awseal | that the artifact came from who you think | an Ed25519 seal — the key that verifies is not the key that forges |
| awshare | that the download is intact | content-addressed bundles, verified on fetch |
| awsuite | that an agent holding your mailbox will not send on its own | every send, draft, upload and create returns a dry-run until confirm is true |
| awpermit | that an admitted agent will only do what its owner meant | an Ed25519 permit naming the agent, the principal, the door, the actions and the cap |
| awnest | that there is a person on the other end | a verdict with evidence, where "we could not tell" is not "yes" |
| awrena | a leaderboard someone can edit, and votes nobody counted | a scored duel with both answers kept, and a result bound to them |
| awnboard | a share link anyone who sees it can use | an invitation addressed to one person, for one gate, revocable |
| awnix | that the box is what you left it as | an immutable image you built, with atomic rollback |
| awrecover | that the restore worked | a restore that fully lands or does not land at all |
| awstorage | a du you ran last month, and a peers file that says 3 TB free | an inventory snapshot per node with a diff since the last one, and each tree classified re-fetchable or not |
| awrelay | a SaaS in the middle of your agents | findings, alerts and coordination over your own transport |
| awask | that anyone read the paragraph where you asked | the ask itself, with a button that steers the session that raised it |
| awmail | a mailbox somebody else can read | mail your agents send and receive over your own server |
| awdiscord | a self-bot on someone's account, or a bot that acts without asking | slash commands, rate limits, and approval buttons only your operators can press |
| awswarm | that a model either fits your GPU or it doesn't run at all | a placement plan and an acquisition-probability estimate before you spend on a run |
| awfind | one vendor's idea of the web | results from whichever providers you configured |
| awbrowse | that the page said what you were told | the render, the DOM and the requests it made |
| awvoice | that a cloud vendor may hold your audio | a transcript and a wav from a service you host |
| awvision | a filename and a caption somebody wrote | what a model actually reports about the pixels |
| awscreen | a selector that was true when the page was written | the elements actually rendered, by what they look like |
| awbeads | that a layout your users built survives the next deploy | the arrangement as data you can read back, diff, and hand to another surface |
| awsprite | a chatbot with a face that resets every session | a knowledge base you can list, edit, and watch shape what it becomes |
| awbonsai | that inference always means a request left the machine | a WebGPU model answering on the tab's own GPU, with a consent record logged before it ever loaded |
| gawbbonet | the model to keep a 300-message campaign coherent by itself | campaign facts recalled from scoped memory you can list and edit |
| aitherkvcache | a vendor's quantisation defaults | sub-byte KV cache kernels you can benchmark yourself |
| awrtifact | a hand-rolled split script and a hand-edited worker manifest | byte-verified parts in a release, served with Range + CORS, sizes asserted by a live gate |
| AitherZero | a pile of scripts nobody has numbered | numbered, discoverable automation with declarative playbooks |
| AitherConnect | what a page tells your browser to do | a federated search and desktop bridge you host |
| awreason | a confident paragraph | the phases it went through, and every tool call it made to get there |
| awrecurse | that everything you pasted in was actually read | which slices it opened, and what it concluded from each |
| awprism | the first explanation that fits | the ranked alternatives, and the observation that separates them |
| awrepl | what the agent believes the value is | the value, printed from the live session |
| awreport | that the report you pasted carried no token in it | a redacted report, and the duplicate it merged into instead of filing twice |
| awresearch | a summary of pages nobody opened | every claim against the source it came from |
| awfocus | twelve terminal tabs and a bad memory | one command that names every session, finds any transcript, and opens or steers the one you want |
| awgym | that a world model learned anything from the games it saw | transitions captured from real play, fed back, and the retrodiction score falling on grids it never saw |
| awpredict | a model because it trained without erroring | its prediction against a self-updating lookup, on the rows that are actually novel |
| awevolve | that your optimisation loop is finding anything | every version it kept, the score that version earned, and the edit that produced it |
| awsh | that you already know the name of the command | what it decided your line meant, before it acts on it |
| awmine | that a session's lesson survived the session | a row per outcome, a candidate per lesson, and the transcript line each one came from |
| awrise | that a scheduled agent ran at all, and ran exactly once | a durable record of every wake -- fired, skipped, overlapped or timed out -- each with its reason |
| awkno | that the docs site is up, or that you remember the family | the whole ecosystem in your terminal, with no network at all |
| awwall | that a service only talks to the hosts you think it talks to | an explicit egress allowlist, where a denial names the rule that denied it |
| awembed | a general-purpose embedder that has never seen your code | a held-out split of whole directories, scored teacher vs student vs int8 |
| awtax | a closed tax app's sealed file you can never read again | a plain, provider-neutral schema of every figure, with the page it came from |
| awsettings | that you will remember to re-approve the same thing on every box you work from | one profile, unioned rather than overwritten, with the credentials left behind |
| awavatar | a cloud 3D vendor's opaque task id | a manifest with a sha256, a licence and a rig-audit verdict per file |

awnix is the ground floor — A Linux you can hand to an agent — immutable base, capabilities included.

The Aitherium ecosystem

Every repository here is public. Each publishes an aither-manifest.json beside its page, so any surface can read every sibling's — the network is browsable from any node in it.

| repo | what it is | pages |
|---|---|---|
| awdk | Build AI agent fleets — 3 lines, any backend, local or cloud | docs |
| awskills (you are here) | Portable agent skills — self-contained procedures an agent loads on demand | docs |
| awpack | First-party agent packs — the ones we build, versioned and installable on their own | docs |
| awm | A portable, scoped agent memory | docs |
| awdesk | Aither World Desk -- the desktop body of AitherOS Online: tray, avatars, decision cards, the Living Desktop as an overlay | docs |
| awnode | A lightweight local gateway — bridges your apps to the AI backends you chose | docs |
| awrun | A priority-aware queue and dispatcher for agentic runs and ad-hoc CI builds. It also judges whether the runner pool is big enough for the queue it is draining, and can ask a host to grow it -- reserving capacity is zero-sum, so a saturated pool needs more of it, not a different share of it | docs |
| awpool | The free half of the elastic work pool: dispatches public-safe work onto the free GitHub-hosted runners of the public aw* mirrors, so their idle CI minutes become compute. awrun remains the paid/self-hosted lane; this is its free twin | docs |
| awgraph | A semantic code graph for agents — AST + tree-sitter, call graphs | docs |
| awgit | Semantic version control on top of git — edit-ops and leases | docs |
| awdelphi | Anonymous multi-round expert panels — a converged answer with a trace | docs |
| awclassify | Classify any document -- what it is, who may read it, who it is for, what it is about | docs |
| awdecide | One typed-decision contract -- choice / score / bool with a probability -- over a ladder of backends you already run (rules, tiny local models, an LLM's logprobs), fail-closed, with a Brier ledger that resolves every decision against its outcome | docs |
| awtoll | What every tool call costs you in context, measured from your own transcripts | docs |
| awseal | Sign an artifact so a stranger can verify it | docs |
| awshare | Publish an artifact and fetch it back verified | docs |
| awsuite | Your Google Workspace as agent tools, and no write happens without a yes | docs |
| awdit | An append-only audit trail whose gaps are DETECTABLE | docs |
| awbac | Role-based access control that fails closed and explains itself | docs |
| awiam | Who is this caller? A directory and session store that fails honestly | docs |
| awtunnel | Reach a service that has no public address | docs |
| awpermit | Let an agent through your door with a scope, a spend cap, and someone accountable | docs |
| awnest | Prove there is a human before you let them into the nest | docs |
| awrena | Put two agents head to head and get a verdict you can check | docs |
| awnboard | A front gate you can put in front of anything, and hand someone the key to | docs |
| awnix | A Linux you can hand to an agent — immutable base, capabilities included | docs |
| awrecover | Labelled snapshots with an all-or-nothing restore | docs |
| awstorage | Every drive on every node, indexed, classified and diffed -- so you can see what you own before you delete it | docs |
| awrelay | Portable agent messaging — findings, alerts, coordination | docs |
| awask | Your agent asks you a question — and acts on your answer | docs |
| awmail | Give an agent an email address — send, and actually receive | docs |
| awdiscord | Put any agent on Discord as a real bot, with an operator in the loop | docs |
| awnet | The agentic web — agents host a mesh, and agents join one | docs |
| awswarm | Run one model too big for any single GPU across a pool of small ones | docs |
| awfind | A portable search client — query, results, ranking | docs |
| awbrowse | A portable browser client — navigate, console, network, DOM, screenshot | docs |
| awvoice | Hear and speak — transcribe audio, synthesize a voice | docs |
| awvision | See an image — describe it, ask it a question, compare two | docs |
| awscreen | See this machine — what is on screen, and where to click it | docs |
| awkit | Render an agent panel from a tool result — one component, any React app | docs |
| awbeads | A spatial canvas for a page — arrange things, connect them, and keep the arrangement | docs |
| awsprite | Hatch a companion that grows only from what you teach it, then take it home | docs |
| awbonsai | Run a real model in the visitor's own browser — no server round trip, no upload | docs |
| awknowledge | How to run a coding agent so the result survives — the laws, with evidence | docs |
| awbrain | Your history as a wiki of linked markdown — claims pinned to the evidence | docs |
| gawbbonet | GobboNet campaigns with a real agent brain — scoped memory, graph recall | docs |
| aitherkvcache | Near-optimal KV cache quantization for LLM inference — sub-byte compression | docs |
| awrtifact | Deliberately chunk artifacts into GitHub release assets — the productized aitherkvcache mirror lane | docs |
| AitherZero | PowerShell 7+ automation framework — numbered, self-describing scripts | docs |
| AitherConnect | Browser extension — federated AI search, page context, and the Living OS overlay | docs |
| awreason | A portable reasoning client — sessions, phases, thoughts, and the chain that produced the answer | docs |
| awrecurse | Answer a question over a context far larger than the window — recursively, with the trace kept | docs |
| awprism | Turn a failure into ranked hypotheses — and say what would confirm each one | docs |
| awrepl | A REPL an agent can actually use — state that survives between turns | docs |
| awreport | File a bug report that has already scrubbed your secrets and collapsed the duplicate | docs |
| awresearch | Ask a research question, get a cited report you can check | docs |
| awfocus | See, search and steer every Claude session from one command | docs |
| awgym | An ARC training gym — a game a world model can watch, and six roles that play through it | docs |
| awpredict | Predict what your environment does next, and how surprised you were | docs |
| awevolve | Point an agent at a file and a command that scores it, and let it improve | docs |
| awsh | Your terminal answers you -- type a question where a command would go | docs |
| awmine | Mine what your agents did -- outcomes, lessons and procedures out of the transcripts they left behind | docs |
| awrise | Wake an agent on a schedule, let it do one thing, and put it back to sleep | docs |
| awkno | The man page for the Aither World — every brick, stack and law, offline | docs |
| awwall | Say what a workload may reach, and watch everything else fail closed | docs |
| awrouter | OpenRouter for your own fleet: pick a model backend by cost/latency/ capability, fail over, fit the context window, stream. Standalone, OpenAI-compatible, no Aither-specifics required to be valuable | docs |
| awembed | Train an embedding model that knows your corpus, and prove it beats the big one | docs |
| awtax | Turn any tax PDF -- returns, W-2, 1099, statements, even scans -- into structured data you can check | docs |
| awflow | A deterministic workflow runtime — chain agent calls with journal replay and budget control | docs |
| awsettings | Your agent's permissions and config, following you to the next machine | docs |
| awavatar | One character spec in, a rigged, animated, multi-style avatar pack out | docs |

Built on REA (Reverse Engineer Anything) · llama.cpp · vLLM · ComfyUI · CentOS Stream · Podman · Docker · LanceDB · WireGuard · FFmpeg · Blender + Rigify · headroom · SANA · Hunyuan3D · repowise · Playwright · Chromium · Next.js · React.

Part of the Aitherium estate

awskills is one of several open repositories cut from the same system. They
share one set of engineering rules, published as the Developer Codex.

New here? → Read the Codex — eighteen laws for building software
that tells you when it is wrong, plus a four-chapter ramp if you have never
worked with a coding agent. Contributing? → CONTRIBUTING.md

• aitherium.com — the system these repos are cut from
• The Codex — how we work here, and how to contribute
• awdk — the Python agent runtime
• AitherZero — PowerShell automation
• aither-kvcache — KV cache compression
• The Blog — what we measured, and what it cost

README from the reviewed commit 5c5c1a6; it updates when the author re-pins.