kronk
AGENTS.md868 starsChanged 13 days agoApache-2.0View on GitHub
Your name is Dave and developers will use your name when interacting with you. The manual has been split into separate chapter files in the .manual/ directory. Read only the chapters relevant to your task. Each chapter file begins with its own section index — open the chapter file when you need section-level detail. You will want to look at Chapter 20: Developer Guide for detailed information about the project structure, code, and workflows. Project skills live in .agents/skills/. Load…
- Installs packages
Where files are headingNew AGENTS.md files have more rules about keeping secrets out of the code, and fewer about code style.
Of the AGENTS.md files we read, the 32,671 first seen in the last six months, against the older ones: the kind of rule that gained most share, and the kind that lost most. About files like this in general, not this one, and an estimate, since a kind is a pattern in the words.
In this repository5 other files it gives its agents.
- .agents/default/AGENTS.mdAGENTS.md
- kronk-pr-reviewSkill
- kronk-service-architectureSkill
- probing-kronkSkill
- use-modern-goSkill
The file
We split this AGENTS.md into the rules it gives an agent
4 rules. About 0 are common, 1 is distinctive, and 3 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈1
Distinctive · estimate
This author's own choices. Worth a look.
3
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Very specific: line 11
After modifying any .go file, always run go vet and gofmt -s -w on the changed files.
- Very specific: line 12
After modifying any .go file, always run staticcheck and go fix on the changed package.
- Very specific: line 15
Never perform a full repo test run and never run tests from this location sdk/kronk/tests.
- Distinctive: line 25
Verify documented behavior against the code, but describe the public contract rather than how the code implements it.
≈ “Common” is an estimate: at least 1 in 5 of the 41,581 AGENTS.md files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
AGENTS.md
Your name is Dave and developers will use your name when interacting with you.
The manual has been split into separate chapter files in the .manual/ directory. Read only the chapters relevant to your task. Each chapter file begins with its own section index — open the chapter file when you need section-level detail.
You will want to look at Chapter 20: Developer Guide for detailed information about the project structure, code, and workflows.
Basic Rules
- After modifying any
.gofile, always rungo vetandgofmt -s -won the changed files. - After modifying any
.gofile, always runstaticcheckandgo fixon the changed package. - You need this env var to run tests
- export GITHUB_WORKSPACE=
<Root Location Of Kronk Project>
- export GITHUB_WORKSPACE=
- Never perform a full repo test run and never run tests from this location
sdk/kronk/tests. - Download model fixtures with
kronk model pull <SOURCE> --local(and--projwhen needed), not direct downloads into scratch directories. Keep them in Kronk's normal model store so developers can reuse them after verification.
Documentation Rules
- Reserve
BREAKING_CHANGES.mdfor incompatible changes to Kronk's existing APIs or observable behavior, including wire formats, configuration, and defaults; describe the incompatibility and required migration. Document compatible additive features and opt-in options in the manual instead. - Upgrading Kronk requires upgrading its bindings and matching native libraries together. A mandatory dependency or native-library upgrade, including an upstream ABI change, is not a breaking change by itself when Kronk's public APIs and observable behavior remain compatible. Keep matching-version and installation instructions in README or the manual, not
BREAKING_CHANGES.md. - When updating README's compatibility table, use
sdk/kronk/kronk.go:Versionwithout the-rcsuffix for the first row (for example,1.33.3-rcbecomes1.33.3). Update that version's row in place while it is a release candidate; do not add a separatemainorunreleasedrow or treat the existing row as a completed release. - Focus user-facing documentation on Kronk's semantics and specification: what features mean, their observable behavior, supported inputs and outputs, defaults, constraints, and limitations. Do not explain internal code structure, private functions, or implementation mechanics.
- Be concrete for CLI usage, installation, and configuration. Provide actionable steps, exact commands and flags, file locations, configuration examples, and expected results so users can perform the task without reading the source code.
- Verify documented behavior against the code, but describe the public contract rather than how the code implements it. Keep implementation details in developer-focused documentation, such as Chapter 20: Developer Guide, when they are needed for contributors.
Skills
Project skills live in .agents/skills/. Load the matching skill before the work it covers.
| Skill | Load when |
use-modern-go | Writing, editing, or reviewing any .go file: modern syntax for the Go version in go.mod. |
kronk-service-architecture | Writing, editing, or auditing Go under cmd/, sdk/, or examples/: ownership and import direction, HTTP request/response types, type-set enums, SDK errors and errs.FromSDK, lifecycle and inference invariants, branching style. |
kronk-pr-review | Any review request ("review this PR", "review my changes", "pre-commit review", one aspect such as tests or security). Claude Code; read-only; reviews only the change unless a full review is asked for; runs CI's Go static checks, plus focused race tests for a maintainer's PR only; writes reports to .reviews/. |
MCP Services
Kronk has an MCP service and these are settings:
"mcp": {
"Kronk": {
"type": "remote",
"url": "http://localhost:9000/mcp",
"type": "streamableHttp",
"apis": [
{
"api": "web_search",
"desc": "Performs a web search for the given query. Returns a list of relevant web pages with titles, URLs, and descriptions. Use this for general information gathering, research, and finding specific web resources."
}
],
}
}MANUAL Index
Open a chapter file to see its section-level table of contents.
| Chapter | Topics |
| Chapter 1: Introduction | What is Kronk (SDK + Server), key features, supported platforms, architecture overview |
| Chapter 2: Installation & Quick Start | Prerequisites, CLI install (Homebrew, go install, binary, Docker / OCI), libraries, downloading models, starting server, model config file |
| Chapter 3: Model Configuration | GPU config, KV cache, flash attention, NSeqMax, VRAM estimation, GGUF quantization, MoE vs dense vs hybrid performance, speculative decoding |
| Chapter 4: Batch Processing | Slots, sequences, request flow, memory overhead, concurrency by model type |
| Chapter 5: Message Caching | Incremental Message Cache (IMC), hybrid model IMC, multi-user IMC, cache invalidation |
| Chapter 6: Speculative Decoding & MTP | User guide for speculative decoding: separate-GGUF vs auto-detected MTP drafters, when to choose each, acceptance/adaptive throttle, observability, known limitations |
| Chapter 7: YaRN Extended Context | RoPE scaling, YaRN configuration, context extension |
| Chapter 8: Model Server | Server start/stop, configuration, model caching, config files, catalog system |
| Chapter 9: API Endpoints | Chat completions, Responses API, embeddings, reranking, tool calling |
| Chapter 10: Request Parameters | Sampling, repetition control, generation control, grammar, logprobs, cache ID |
| Chapter 11: Multi-Modal Models | Vision models, audio models, media input formats |
| Chapter 12: Security & Authentication | JWT auth, key management, token creation, rate limiting |
| Chapter 13: Browser UI (BUI) | Web interface, downloading libraries/models, catalog browsing, model management, key/token management, apps, model playground |
| Chapter 14: Client Integration | OpenCode install + agent bundles in .agents/ (default + rote), make targets, model config and swapping models, OpenWebUI, Python SDK, curl, LangChain |
| Chapter 15: Observability | Debug server, Prometheus metrics, pprof profiling, tracing |
| Chapter 16: MCP Service | Brave Search, MCP configuration, OpenCode client setup, curl testing |
| Chapter 17: Troubleshooting | Common issues, error messages, debugging tips |
| Chapter 18: Bucky (Audio Transcription) | whisper.cpp-based transcription: install/libs, model catalog, server/pool config, CLI, BUI, API endpoint, SDK quick-start, streaming API (Feed/Events/Reset, VAD, live-mic example), languages, troubleshooting |
| Chapter 19: Malina (Image Generation) | Experimental stable-diffusion.cpp SDK: libraries, curated model bundles, text-to-image, image-to-image, FLUX.2 components, progress/logging, AVI encoding, examples, limitations |
| Chapter 20: Developer Guide | Build commands, project architecture, BUI development, code style, SDK internals, MTP engine internals, Bucky/Malina ownership, CI workflows (lint/test, goreleaser, container build & publish), version stamping |
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

