Skip to content

Quick Start

Get kedge running in your code repository in five minutes.

Prerequisites

  • kedge installed
  • A git repository with code
  • A docs repository with markdown steering files (or you'll create one below)

1. Initialize

In your code repository root:

kedge init

kedge creates a kedge.toml with sensible defaults. Open it and configure:

  • [triage]: your AI provider (anthropic, openai, or command)
  • [remediation].agent_command: the command that updates docs (e.g., a Kiro or Claude Code agent)
  • [[repos.docs]]: the git URL and path of your documentation repository
[detection]
# exclude_dirs = [".git", "node_modules", "target", ".venv", "__pycache__", ".tox", "vendor"]

[triage]
provider = "anthropic"
model = "claude-haiku-4-5-20251001"

[remediation]
agent_command = "your-agent-command"
auto_merge_severities = ["no_update"]

[repos]
docs = [
  { url = "git@github.com:your-org/docs.git", path = "steering/", ref = "main" },
]

[[agents]]
name = "claude"
global_steering = "~/.claude/docs"
workspace_steering = "docs"
agents_file = "CLAUDE.md"
skill_dir = ".claude/skills/"

2. Stamp provenance

Once your steering files have kedge: frontmatter with anchors pointing to code locations:

kedge link

kedge computes an AST fingerprint for each anchored code location and writes it into the steering file's provenance field. Future drift detection compares against this baseline.

3. Detect drift

kedge check

kedge outputs a JSON drift report to stdout. Exit code 0 means all docs are up to date. Exit code 1 means drift was found. At least one anchor's code has changed since provenance was last stamped.

Save the report to a file:

kedge check --report drift.json

4. Run the full pipeline

kedge update

kedge runs the complete three-layer pipeline:

  1. Detection: compares AST fingerprints at provenance vs HEAD
  2. Triage: classifies each drifted anchor as no_update, minor, or major via AI
  3. Remediation: invokes your agent to update docs and open merge requests, and advances provenance for no_update anchors

kedge outputs a RemediationSummary JSON showing what was remediated, what had provenance advanced, and any errors.

5. Check status

kedge status

kedge shows all steering files with their anchors, groups, and current provenance values.

Optional: distribute steering files to agents

If your agents read docs from specific local directories (e.g., .kiro/steering/), use kedge install to copy or symlink steering files there. This is independent of the drift pipeline above.

kedge install --link                  # dev machine: symlink to global agent dirs
kedge install --workspace             # CI: copy to workspace dirs

See Agent Integration for details on configuring [[agents]] platforms.

Next steps