# Lean Go And See

> Bootstrap skill for go and see (genchi genbutsu). Read the live surface and not the account of it; every ruling carries a live check or is marked one-time. Replaced by its node-backed skill on node_work_ledger_append_effect, under the factory plan; replace it when that lands.

- Skill: `omninode-ai/lean-go-and-see` (Agent Skill)
- Install (CLI): `npx get-skill add omninode-ai/lean-go-and-see`
- Raw SKILL.md: https://api.skillmd.com/api/skills/omninode-ai/lean-go-and-see/raw
- Safety review: pending (external: skill-scanner PASS, skillspector PASS)
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: Coding & Dev Tools
- Author: OmniNode-ai (https://skillmd.com/u/omninode-ai)
- Updated: 2026-11-11
- Page: https://skillmd.com/skills/omninode-ai/lean-go-and-see

---


# Lean Go and See (bootstrap)

This skill is a bootstrap skill. It is the hand-run form of the practice, kept so lanes can test and
iterate on the process while the real mechanism is built; it is a tool to make tools. It checks
nothing for you: it asks you to read the live surface by hand before you report on it. It is
replaced by its node-backed skill when that skill comes online.

**Replaced by** the node-backed go-and-see skill on the typed work-ledger append, `node_work_ledger_append_effect`
(plan task T-P4: a `live-check=` cell on every ruling at intake, refused by the typed ledger append when absent, and a
drift check on a schedule) and `node_evidence_ref_resolve_effect` (T-L2, the evidence reference resolver, which makes a
chain's evidence a lookup). These tasks have no child ticket yet; they sit under the factory plan. **Replace this skill with the
node-backed skill when T-P4 is merged and deployed.**

## The practice

Read the live surface, not the account of it. A summary, a ledger row, a memory
note or a ticket status says what someone saw at one time. The live state says what
is true now. Report the live reading with the command that took it and the time.

## When to use

Before you state that something is running, merged, deployed, fixed, drifted or
done; before you record a ruling; before you trust a brief's account of state.

## Steps by hand today

1. **Name the surface.** What exactly governs the claim: a pull request head on the
   remote, a process on a host, a deployed tag, a config value, a check-run.
2. **Read it yourself.** Use the narrowest command that returns the state. Read the
   check-runs for a head and not a rollup; read the remote default branch and not a
   stale local copy.
3. **Classify the result.**
   - **MATCH**: the live state agrees with the claim.
   - **DRIFTED**: it disagrees. Say how, and open an andon stop if it is a class
     (`lean_andon_stop`).
   - **UNREADABLE**: you cannot read it. Say so. Never report UNREADABLE as MATCH.
4. **Report with provenance.** The claim, the command, the output and the read time.
5. **Give new rulings a live check.** When you record a ruling, write the command
   that reads the state it governs, as a `live-check=` cell. If it governs a
   one-time decision with nothing to read afterwards, mark it so. A ruling with
   neither cannot be audited later, and that is the largest gap the plan found:
   most recorded rulings were unreadable.

## Factory task per step

Each step above is replaced by this task of the lean practices plan.

| Step | Factory task |
| -- | -- |
| 1 | no task yet |
| 2 | T-L2 (evidence reference resolver) |
| 3 | T-P4 (drift check) |
| 4 | T-L2 (evidence reference resolver) |
| 5 | T-P4 (`live-check=` cell on every ruling) |

## Common traps

- Reporting from the ledger what only the live surface can show.
- Reading a local clone that is behind the remote.
- Reading a green rollup while a required context is missing or stale.
- Treating "I fixed it" as "it reads fixed".

## What this skill cannot do

It cannot make a ruling auditable after the fact, and it cannot run the drift check.
Until T-P4 lands, the audit of the existing rulings is a manual read.

