docs: close the gaps in the prose for people #203
Loading…
Add table
Add a link
Reference in a new issue
No description provided.
Delete branch "docs/people-prose-gaps"
Deleting a branch is permanent. Although the deleted branch may continue to exist for a short time before it actually gets removed, it CANNOT be undone in most cases. Continue?
Why
A documentation audit of the prose for people, run on 2026-11-06 at
becb455, found the README,docs/INSTALL.md, anddocs/MAINTENANCE.mdanswering their readers only in part. A person deciding whether to
adopt could not learn what the two hooks do to a session, nor that
.claude/settings.jsonturns off more than the four denied tools,although adoption copies all of it into every user's sessions. A Codex
user following the Quick start never met the hook trust that only they
can give, so their sessions ran without the main-checkout guard. A
repository adopted before
foundry-upgradewas sent to the Quickstart, which only adopts. The maintainer could not tell which tool
versions the checks pass on, while the forge's job pins ShellCheck
0.11.0, whose codes the scripts' directives name, and the README allows
0.7.0.
What changed
A new README section, Hooks and settings, placed after the Quick start
and holding the Codex subsections, says what the main-checkout guard
refuses and what the shell check does, lists the other switches of
.claude/settings.jsonwith the way out for any of them, and says theOpenCode plugins are tested against stubs only; the Description points
to it. The Quick start opens with the pointer to
docs/INSTALL.mdforthe other cases, and its step 3 has a Codex user trust the hooks in the
fresh session that loads the copied files, while adoption step 2 has
the agent ask the user to do so there, as
UPGRADING.mdalready didfor an upgrade.
docs/INSTALL.mdsends a repository adopted beforefoundry-upgradetothe README's Upgrading section, and tells a new repository's step 2 in
three sentences: the empty commit, its push on the user's word before
the adoption's pull request, and the copy or the deletions.
docs/MAINTENANCE.mdnames the versions the checks pass on, as theforge's job pins them, and owns the Node.js requirement, with the test
that runs it, so the README's Dependencies no longer list Node.js for an
adopter.
How to verify
Read
README.mdfrom the top on this branch: the Description points toHooks and settings, which names what each hook does and every switch of
.claude/settings.json, whichjq -r 'keys[]' .claude/settings.jsonlists; on main the Description names the hooks without saying what they
do and only the denied tools. Then read the Quick start as a Codex
user: step 3 has you trust the hooks in the fresh session; on main no
step does. Read
docs/INSTALL.md§ Adopted beforefoundry-upgrade:its link lands on the README's Upgrading section; on main it says "as
the Quick start describes". Read the first paragraph of
docs/MAINTENANCE.mdbeside.forgejo/workflows/maintenance.yml: theversions match its image comment and its
npm installline.Checks run on the tip: the six checks of
docs/MAINTENANCE.md, allpassing, with 225 agent configuration tests and 46 work-folder tests,
and three review rounds of
document-reviewerandsecurity-reviewerwith no finding left open. No regression test changes: the change is
prose only. Not tried: an adoption from Codex or OpenCode, since
neither client is installed here.
Considerations
The Codex hooks documentation does not say whether a running session
lists a
hooks.jsoncreated after it started, only that Codex warns atstartup when hooks need review, so the Quick start and adoption step 2
put the hook trust in the fresh session that loads the copied files,
which holds either way. The guard's sentence gives its gist and leaves
its cases to the script's header and to the refusal's reason. The
audit's rule on when a change needs a migration in
UPGRADING.mdisleft for a decision of its own, and its findings on the workflow pages
and the eval runner's header for a later change.
f87ece4c24cf80a41b2c