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/av/harbor/annealnpx skills add av/harbor --skill annealgit clone --depth 1 https://github.com/av/harborWhat 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 | $0.00049 | $0.03417 |
| Opus 5 | $0.00024 | $0.01708 |
| Sonnet 5 | $0.00010 | $0.00683 |
| Haiku 4.5 | $0.00005 | $0.00342 |
Grade A, and why
anneal 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 yesterday.
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 — 283 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Anneal
Systematically harden a codebase by fixing AI-introduced code slop over a user-specified duration. The clock decides when work stops, not you.
Your Role
You are the orchestrator. You do exactly two things:
- Manage the clock — check time before every dispatch, stop when the deadline passes
- Dispatch subagents — give them the slop catalog, the progress file path, and get out of the way
You do NOT do any actual work. No code changes, no file edits, no exploration, no analysis, no "quick fixes." All productive work happens inside subagents. Your context is reserved exclusively for the dispatch loop. If you catch yourself doing anything other than checking time, reading the progress file, and dispatching — stop. That work belongs in a subagent.
The Iron Law
YOU DO NOT DECIDE WHEN THE WORK IS DONE. THE CLOCK DECIDES.
Your only job is to keep dispatching useful work until the deadline passes. You have zero authority to judge completeness, sufficiency, or "good enough." The user gave you a duration. You use all of it.
Inputs
The user provides two things:
- Codebase — the project to anneal (defaults to the current working directory)
- Duration — how long to run (e.g., "4 hours", "overnight", "90 minutes")
If the duration is vague ("overnight"), interpret it as 8 hours. If truly ambiguous, ask once.
The Slop Catalog
These are the patterns AI agents actually introduce into codebases. Every subagent receives this catalog as its detection guide. The subagent picks the highest-impact instance it can find — the catalog is a field guide, not a queue.
1. Duplication Instead of Reuse
Reimplements logic that already exists elsewhere in the codebase. The agent lacked full-repo context and produced a new version instead of calling the existing one. Look for: near-identical functions across files, same algorithm implemented with different variable names, utility code that duplicates a library the project already depends on.
What ships with it
1 file 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.
- yesterday First seen · 283 lines · 49 tokens per session scan A e00230717069
anneal is a skill published in the GitHub repository av/harbor (3,198 stars, last pushed 2d ago), licensed Apache-2.0. It adds 49 tokens to every session and 3,417 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 2026-08-30.
Other skills, from other repositories
docker-local-build
Build and test Kurtosis from source on local Docker. Compiles all components (engine, core, files-artifacts-expander), builds Docker images, installs the CLI, and restarts the engine. Use when developing Kurtosis and testing changes locally with Docker.
files-inspect
Inspect, download, upload, and debug Kurtosis file artifacts. View artifacts in an enclave, download them locally for inspection, upload local files, and troubleshoot file mounting issues. Use when services can't find expected files or configs are wrong.
k8s-dev-deploy
Build, push, and deploy Kurtosis dev images to a Kubernetes cluster without creating a release. Rebuilds engine, core, and files-artifacts-expander as multi-arch Docker images with a unique tag, pushes to the logged-in user's Docker Hub, and restarts the engine. Use when testing local code changes on a k8s cluster.
starlark-dev
Develop and debug Kurtosis Starlark packages. Create packages from scratch, understand the plan-based execution model, use print() debugging, handle future references, and test packages locally. Use when writing or troubleshooting .star files.
docker-debug
Debug Kurtosis running on local Docker. Inspect engine, API container, and service logs. Diagnose container crashes, port conflicts, and networking issues. Use when kurtosis commands fail or services aren't reachable on Docker.
enclave-inspect
Inspect and manage Kurtosis enclaves. List enclaves, view services and ports, examine file artifacts, dump enclave state for debugging, and clean up. Use when you need to understand what's running inside an enclave or export its state.