shadok-ledgerlisted
Install: claude install-skill shadok-ai/shadok-ai
# shadok-ledger
A shared **state table** across this instance's agents: **one current status per
entity** (a bug, a task, a campaign, an extraction job, a decision…). It exists
because agents run siloed and keep re-raising things another agent already
handled. It is a *table*, not a log: re-recording an entity **supersedes** its
row, so it stays small.
**Source of truth first.** For code and pull requests, the truth is `git` / `gh`
— check those, not this ledger. Use the ledger for everything that has **no
system of record**: infra changes, marketing actions, site activity, ops, tasks.
## Before you assert or act on a status
Run `check` and let it steer you:
```
node ~/.claude/skills/shadok-ledger/ledger.mjs check "<topic, in your own words>"
```
**Ask in plain language — don't try to guess how a sibling worded the row.** The
search ranks rows by word overlap, so `check "claude launcher stub spawn
failure"` finds `claude-launcher-stub-breaks-spawns`. Describing the topic in a
few words beats one lucky keyword; extra words that turn out to be absent cost
you nothing.
- a row says **resolved / done** → do NOT re-raise or re-act; report it as handled.
- **nothing recorded** → treat as **UNKNOWN**, not as "not done": ask the human
or hedge ("from memory, still to do — tell me if it is already handled"),
don't assert.
- **related rows — they may NOT answer your question** → exactly what it says:
the search found no row that answers you, only rows sharing some words. Y