diff --git a/AGENTS.md b/AGENTS.md
index 942272a..331038d 100644
--- a/AGENTS.md
+++ b/AGENTS.md
@@ -181,16 +181,19 @@ Per chunk:
the PR summary names each fix no round reviewed.
3. Merge the PR (standing authorization, this repo only, granted by the maintainer), report, stop.
-A chunk reads these files in part. One call prints a `## ` section by a fragment of its title:
-`awk -v h='
' '/^## /{p=index($0,h)>0} p' | head -80`.
+A chunk reads these files in part. `grep -n '^#' ` lists the headings. This prints the first
+`## ` section whose title holds the fragment, so one call reads a section whole, and the fragment
+holds no apostrophe: `awk -v h='' '/^## /{if(p)exit; p=index($0,h)>0} p' `. A `###`
+section, or one past 80 lines, is read in ranges.
- `CHANGELOG.md`: `## Unreleased` only, the one section a chunk edits.
- `MIGRATION.md`: the top `## From` section, the release being built.
- `README.md`: `## Goals`, `## Audience` and the sections the chunk changes or cites.
-- `docs/decisions.md`: the entries the chunk cites, found by the titles indexed above, never the file.
+- `docs/decisions.md`: only the entries the chunk cites, by their titles under `## Decisions` above.
- `spec/flavour.md`, `spec/html.md`: the sections the chunk changes or cites.
-- A test file or `ci.sh`: its titles or legs, from `grep -n '^test(\|^leg' | cut -c1-110`,
- then a ranged read of the ones the chunk touches.
+- A test file or `ci.sh`: its tests or legs, from
+ `grep -nE '^\s*test\(|^(\w+=\S+ )*leg ' | cut -c1-110`, then a ranged read of the ones the
+ chunk touches.
Reserved for the maintainer whatever any rule here says: changing `version` in `package.json` (a
bump on `main` publishes, `docs/decisions.md` §Publish on a version bump — every release is the