# Migrating from the AgentRig CLI

The document-first AgentRig is a **clean break**, not an update of the npm CLI. Point an agent
at the docsite's `llms.txt`, which links [AGENT-RIG.md](./principles.txt), the native assets, and
the validator from one site build. The agent inspects and reconciles your repository. There is
no new installer, CLI, npm runtime, manifest, or automatic migration. Your existing target files
are **not** deleted or changed by this repository: review any removals or replacements yourself,
especially generated files used by agent hosts.

## From Core 1.0 to Core 2.0

Core 2.0 adds three default rules: `security.md`, `code-review.md`, and
`no-debug-logging.md`, alongside the existing engineering-principles rule.
In a repository already using Core 1.0, follow the updated `llms.txt` index,
copy those three rules into `.agents/rules/` without overwriting local edits,
and rerun the downloaded validator and your project checks. The additional
operational skills are published in the index but remain optional; select them
only for workflows the repository actually uses.

| Former command or surface | Document-first workflow |
| --- | --- |
| `agentrig init [path]` | Point an agent at the docsite's `llms.txt`; it inspects the target, writes/reconciles its `AGENTS.md`, and copies only the Core files linked there. |
| `agentrig update [path]` | Re-read one `llms.txt` index (or raw files from one pinned tag/commit), compare with local instructions, and reconcile intentionally. No ownership database or auto-migration. |
| `agentrig compile [path]` | Use `AGENTS.md` and `.agents/` directly; hand-author only the vendor adapters actually needed. No generic projection/compiler step. |
| `agentrig fix [path]` | Download the validator linked in `llms.txt` to a temporary file, run `node /path/to/downloaded/validate.mjs /path/to/target`, fix reported errors, and rerun. No automatic repair or backup semantics. |
| `agentrig eval --static` | Run the same validator. It checks the small Core contract, not an install score or agent behavior. |
| `agentrig eval`, `--scaffold`, dynamic scoring | Removed from Core; create a separate evaluation project if behavior experiments are needed. |
| `agentrig doctor` | Run the validator plus the target repository's own Build/Test/Lint checks; there is no provider or package health probe. |
| `agentrig dashboard` | Removed; consult the external issue/PR system if the repository uses automated queues. |
| `--skip-agent`, `--dry-run`, `--force`, provider/model flags | No corresponding runtime flags. Ask your agent to preview diffs, preserve local work, and request human approval for risky changes. |

| Former artifact | What replaces it |
| --- | --- |
| `knowledge/` templates, `manifest.json`, `checks.json`, `.agentrig/state.json` | One specification and directly copyable `.agents/` prompts, rules, and skills. No second template tree or ownership state. |
| Triager/developer/reviewer/judge YAML and model tiers | Builder plus the fresh, read-only Paranoid prompt; no model IDs, role YAML, or workflow-state ownership. Add a specialist reviewer for high-risk work if needed. |
| `.agentrig/harness/`, labels, pollers, worktree repair | Optional state-machine/external-record and isolated-worktree recipes when concurrency or automated queues justify them. |
| `.agentrig/eval/`, dashboard, generated vendor instructions, MCP/setup files | Not part of Core. Keep or hand-author *existing target-specific* files that your tools genuinely require; do not delete them as part of a blanket migration. |
| npm CLI build/publish machinery | The Git-versioned specification and its native assets; validation requires only Node.js. A private docs-only package builds a website mirror, not a CLI. Do not infer that a new release tag already exists. |

The published `@doidor/agentrig@0.12.0` package is the **historical CLI path**, not the new
document-first workflow. If you must reproduce an old CLI run, pin that version explicitly
(`npx @doidor/agentrig@0.12.0 ...`); it does not install or upgrade this Core profile. For a
reproducible document-first bootstrap, use raw files from the **same existing tag or commit**.
Otherwise, point an agent at one docsite `llms.txt`. Preserve your own customizations, review
the diff, run your repository checks, then run the validator linked from that site. No clone,
automatic deletion, silent migration, or upstream npm action is part of that process.