agentleFS
apokamoapokamo/kaji/llms.txt

kaji

llms.txt16 starsChanged 28 days agoApache-2.0View on GitHub

Closed-loop agentic development workflows for Claude Code, Codex, and Antigravity CLI. kaji is a Python CLI workflow orchestrator for AI-assisted software development. It turns an issue into a resumable design -> implement -> review -> fix -> verify -> PR loop, with human-in-the-loop gates and artifact-backed verdicts. Current public release: v0.12.0. The supported user-facing entry point is the kaji CLI. The implementation lives in the kaji_harness package. - Closed-loop workflow: explicit review -> fix -> verify cycles instead of one-shot…

The file

We split this llms.txt into the rules it gives an agent

6 rules. About 0 are common, 3 are distinctive, and 3 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈3

    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
  1. Distinctive:

    Run a workflow in a configured target repository:

    line 75
  2. Distinctive:

    Prefer the public README and this file for orientation before reading deeper docs.

    line 146
  3. Very specific:

    Use docs/ARCHITECTURE.md for implementation architecture.

    line 148
  4. Very specific:

    Use docs/dev/workflow-authoring.md and docs/dev/skill-authoring.md for workflow or skill changes.

    line 149
  5. Distinctive:

    Do not infer support for agents beyond Claude Code, Codex, and Antigravity CLI unless a workflow or adapter explicitly exists.

    line 151
  6. Very specific:

    Do not treat legacy/ as the current runtime.

    line 153

≈ “Common” is an estimate: at least 1 in 5 of the 3,592 llms.txt 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.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
llms.txt
# kaji

> Closed-loop agentic development workflows for Claude Code, Codex, and Antigravity CLI.

kaji is a Python CLI workflow orchestrator for AI-assisted software development.
It turns an issue into a resumable design -> implement -> review -> fix -> verify
-> PR loop, with human-in-the-loop gates and artifact-backed verdicts.

Current public release: v0.12.0.

The supported user-facing entry point is the `kaji` CLI. The implementation lives
in the `kaji_harness` package.

## Core Concepts

- Closed-loop workflow: explicit review -> fix -> verify cycles instead of
  one-shot prompting.
- Multi-agent orchestration: workflow steps can route work to Claude Code,
  Codex, or Antigravity CLI.
- Verdict-driven control flow: steps return `PASS`, `RETRY`, `BACK`, or `ABORT`.
- Artifact-backed verdicts: `verdict.yaml` is the primary completion signal when
  available.
- Human-in-the-loop gates: humans keep control of issue readiness, design
  direction, PR review, and merge decisions.
- Interactive terminal runner: kaji can run normal agent CLIs in Herdr panes,
  record raw terminal logs, and advance the workflow when verdict artifacts
  appear. The former tmux backend was removed.

## Start Here

- [README](README.md): public overview, quick start, workflow example, and
  feature summary.
- [Japanese README](README.ja.md): Japanese overview of the same public content.
- [Documentation index](docs/README.md): map of how-to, reference, operations,
  and concept docs.
- [Changelog](CHANGELOG.md): release history and current behavior changes.

## Architecture and Workflow

- [Architecture](docs/ARCHITECTURE.md): harness architecture, dispatch paths,
  verdict resolution, state, logging, and artifact layout.
- [Workflow overview](docs/dev/workflow_overview.md): entry point for choosing a
  workflow by issue type.
- [Workflow authoring](docs/dev/workflow-authoring.md): how to write workflow
  YAML, steps, transitions, cycles, and verdict mappings.
- [Skill authoring](docs/dev/skill-authoring.md): how to write skills and their
  verdict contracts.
- [Interactive terminal runner](docs/cli-guides/interactive-terminal-runner.md)
  ([Japanese](docs/cli-guides/interactive-terminal-runner.ja.md)): Herdr-backed
  runner for normal Claude Code, Codex, and Antigravity CLI sessions.
- [Antigravity CLI Guide](docs/cli-guides/antigravity-cli-session-guide.md):
  verified `agy -p` / `agy -i`, policy mapping, plain-text output, and explicit
  resume limitations.
- [Claude Code CLI Guide](docs/cli-guides/claude-code-cli-guide.md)
  ([Japanese](docs/cli-guides/claude-code-cli-guide.ja.md)): verified Claude
  Code CLI options, subcommands, permissions, agents, MCP, plugins, and kaji
  integration notes.
- [Codex CLI Session Guide](docs/cli-guides/codex-cli-session-guide.md)
  ([Japanese](docs/cli-guides/codex-cli-session-guide.ja.md)): verified Codex
  CLI session, resume, structured output, sandbox, MCP, plugin, and cloud
  integration notes.
- [AI-driven strategy](docs/concepts/ai-driven-strategy.md)
  ([Japanese](docs/concepts/ai-driven-strategy.ja.md)): project philosophy for
  human direction and AI-executed development loops.

## CLI and Setup

Install kaji for use in another repository:

```bash
uv tool install git+https://github.com/apokamo/kaji.git
kaji --help
```

Run a workflow in a configured target repository:

```bash
kaji run .kaji/wf/official/dev.yaml <issue-id>
kaji run .kaji/wf/official/dev.yaml <issue-id> --from <step-id>
kaji run .kaji/wf/official/dev.yaml <issue-id> --step <step-id>
```

Workflow files are run from `.kaji/wf/` in each repository and are split by
ownership: `.kaji/wf/official/**` is provided/updated/regression-tested by kaji,
`.kaji/wf/custom/**` is owned by the repository (copy official into custom instead
of editing it in place). Canonical contract: docs/dev/workflow-authoring.md
section "ファイル配置". This repository currently includes
`.kaji/wf/official/dev.yaml` and `.kaji/wf/official/docs.yaml` as official
GitHub-backed workflows, plus custom variants such as
`.kaji/wf/custom/dev/dev-thorough.yaml`. Template distribution
and first-time setup details are tracked in:
https://github.com/apokamo/kaji/issues/242

The `dev.yaml` workflow assumes that a GitHub issue exists, required skills are
available, selected agent CLIs are installed, and `/issue-create` has already
been completed. It runs `issue-start` itself.

Set up kaji for repository development:

```bash
git clone https://github.com/apokamo/kaji.git
cd kaji
uv sync
source .venv/bin/activate
kaji --help
```

Quality checks:

```bash
make check
make verify-docs
make verify-packaging
```

Minimal repository config for the GitHub-backed workflow set:

```toml
[paths]
artifacts_dir = ".kaji-artifacts"
skill_dir = ".claude/skills"

[execution]
default_timeout = 1800

[provider]
type = "github"

[provider.github]
repo = "<owner>/<name>"
default_branch = "main"
```

## Important Paths

- `kaji_harness/`: supported runtime package behind the `kaji` CLI.
- `.kaji/wf/official/`: kaji-provided workflow YAMLs (updated and tested by kaji).
- `.kaji/wf/custom/`: repository-owned workflow YAMLs (outside kaji's pytest scope).
- `docs/`: canonical documentation.
- `docs/cli-guides/`: CLI and runner guides.
- `docs/dev/`: workflow and skill authoring docs.
- `legacy/`: historical reference code, not the current supported runtime.

## Notes for AI Assistants

- Prefer the public README and this file for orientation before reading deeper
  docs.
- Use `docs/ARCHITECTURE.md` for implementation architecture.
- Use `docs/dev/workflow-authoring.md` and `docs/dev/skill-authoring.md` for
  workflow or skill changes.
- Do not infer support for agents beyond Claude Code, Codex, and Antigravity CLI
  unless a workflow or adapter explicitly exists.
- Do not treat `legacy/` as the current runtime.

Download now (ZIP)

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
What did you use it for? e.g. Reviewing pull requests in a Rails app

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.