Keep a WAD current
Generated Markdown for references/process_keeping_wad_current.md.
Open book page Back to the skill graph
# Keep a WAD current
## Purpose
The maintenance sweep that stops the document drifting from reality.
## Prerequisites
- A populated WAD with a recorded baseline
## Flowchart

## Steps
### Step 1: Trigger
Run the sweep when something meaningful lands: code merged, infrastructure changed, a decision made, an incident that revealed the doc was wrong.
### Step 2: Re-run the extractors
`just extract` (or `just wad-extract` in the WCL repo) — generated data re-derives from the live sources; the git diff of `data/generated/` IS the machine-visible drift.
### Step 3: Diff against the baseline
`wcl diff <baseline>:wad.wcl wad.wcl` — read what moved. Update the hand-authored blocks the changes invalidate (a renamed crate's relations, a retired container's deploy targets).
### Step 4: Read the commit history
Have an LLM read the **git log since the last sweep** and list what changed that the WAD hasn't caught up with — renamed modules, new services, retired jobs, decisions visible only in commit messages. Commits are the change log the document must answer to; extractors only cover the machine-derivable slice.
### Step 5: Capture decisions
Anything decided since the last sweep becomes an `adr` now, while the context is fresh. New operational learnings update the SOPs.
### Step 6: Check, render, mini-review
`wcl check`, render, and have the owner spot-check the affected chapters. Commit; if the sweep was substantial, record a fresh reviewed baseline.
> [!TIP]
> **Verification**
> Extractors are git-quiet after a re-run, `wcl diff` against the baseline shows only intended drift, and spot-checked chapters match reality.
## Related
- [Generated vs hand-authored data](../references/concept_generated_vs_hand.md)
[← Back to SKILL.md](../SKILL.md)