manual: the pages #1545 flagged now say what dev does

[AbirAbbas]

Closes #1545

#1545 listed twenty places where the chat manual (and one resident page) disagreed with dev 837b2b0. Dev has moved since (#1627, #1501, #1634, #1638), so every row was re-read against the manual and the code as they stand on dev b267d39, and each got one decision:

• FIX-PAGE: the page was wrong, and now says what the code does, quoting person-facing strings as the code spells them.
• ALREADY-TRUE: the page was already right, either because dev moved or because the row misread it.
• CODE-BUG: the page describes the intended behaviour and the code falls short. The page is left alone so it does not bless the defect, and the defect goes to its own issue.
• IN-BATCH-2: Consolidate reviewed CodeAF fixes — batch 2 #1632 changes that exact text or behaviour, so it is left for Consolidate reviewed CodeAF fixes — batch 2 #1632.

#	Page and topic	Decision	What is true on dev
1	standing-orders: the standing card's keys	FIX-PAGE	No 2. 1 sets it up, 3 (once) only on checks and watches, 0 declines, esc is later, and o Change… (o Change where… on a rule) corrects it. What "once" does next is #1632's to describe.
2	standing-orders: · N standing orders here — /standing on opening a conversation	CODE-BUG	Only the in-process agent answers StandingHere. The engine road's remote.Agent does not, so plain codeaf never draws the line; --no-host does.
3	standing-orders: a plain "always …" sentence brings a card	CODE-BUG	The remember and stand tool descriptions both claim "a preference". The model can file a standing rule as a memory note, and no card appears.
4	keeping-an-eye: a grant "is written down and shown on the card"	CODE-BUG	The grant is stored and feeds the rope, but no surface draws it. The design (docs/STANDING-ORDERS.md D3, "Wave 1 stores and displays it") agrees with the page.
5	pages/standing-goals (resident)	ALREADY-TRUE	That page is the resident's own corpus. The resident really asks yes, stand this up / change when it runs / once, not standing and tenures after 3 clean firings.
6	running-from-the-terminal: the Model Pool picks the models	FIX-PAGE	Crew routing uses its own prior and this install's outcomes. All it reads of the pool is the model-name aliases in the copy built into the binary.
7	running-from-the-terminal: the crew reads own.json	FIX-PAGE	Only codeaf pool reads it.
8	running-from-the-terminal: codeaf pool verify output	FIX-PAGE	signature good: version <n>, generated <date>, metrics <names>.
9	running-from-the-terminal: codeaf pool status	FIX-PAGE	Adds pending judge:, last sweep:, dropped N and identity set, and counts waiting rows without listing them.
10	running-on-another-machine: codeaf missing over there	FIX-PAGE	Quoted as the code spells it, with the retired name elided for the name law.
11	running-on-another-machine: rehearsing a drop with pkill	FIX-PAGE	The pattern now spans the -o options ssh really gets.
12	tasks and questions: the task proposal card	FIX-PAGE	codeaf asks · start it in 15s on the top edge, ◆ recommended with it starts on its own unless you say otherwise · fairly sure, any key stops the clock · you can still change the answer afterwards, and the foot esc later · o other · ? clarify. The default is 15s.
13	tasks: 3 levels and 20 pieces "for every task"	FIX-PAGE	Those bounds belong to the node belt. The default bash belt splits with plandb split: 256 per batch, 1024 per run, four wakes per composite.
14	delegates: where a program's badge shows	FIX-PAGE	On the proposal card and rows, not on the landed card.
15	team-questions-and-caps: the held Traffic row	FIX-PAGE	It carries the spend and what is waiting, and a different ending after Stop for today.
16	team-manager: team_start always asks	FIX-PAGE	It follows the approval posture: no card under the default ◇ YOLO, a card under ◇ asks. The card's a new conversation; it spends until it stops has been true since #1627.
17	team-manager: making a manager from the switcher	FIX-PAGE (partly already true)	With All shown the chip reads teams ▾ and has no manager row. The menu stays open, and the row flips to ◇ Make an ordinary member.
18	getting-started: the setup header	FIX-PAGE	setting up · 1 of 2 under the wordmark, or setting up with one step.
19	getting-started: first run always shows setup	FIX-PAGE	With a key already in the profile or the environment, plain codeaf takes the engine road, which opens on home with no setup at all. --no-host still shows Models and spending while no daily limit is set.
20	commands: missing rows	FIX-PAGE	Rows for /drafts, /stop, /autonomy (both forms), /workspace (both forms), /effort (/think, /thinking) and /senior-dev. The table is split into ## sections to keep each under the size cap.

No row is wholly IN-BATCH-2. Row 1's "once" consequence is #1632's (it is described there as an approval handoff), so this change says only which cards offer 3. For row 15, the sentence about hovering a Traffic row was dropped, because #1632 changes what that hover says.

Validation contract

• Every quoted string in a rewritten section is the code's own. Each FIX-PAGE row was checked against its source by reading the literal (table above; the review notes name each file and line).
• A person's question about a rewritten topic reaches its page. Eight probes were appended to TestTheChatManualAnswersTheQuestionsPeopleAsk. Two of them (where are my cleared drafts, what does /workspace path do) reached no command row on dev and pass now. The other six pin the rewritten sections. No existing probe was edited. When an edit pulled how do I start codeaf off its page, the page was reworded.
• The name law holds. internal/namelaw is green, and the ssh message is quoted without the retired name.
• No banned or resident vocabulary is added, and codeaf is lowercase in all added prose (internal/manual tests, plus a grep of the diff).
• Each edited ## section stays under 2000 characters. The exception is "The manager's screen", which was 6811 on dev and is 5553 after its team_start part moved to its own section.
• The pages do not collide with Consolidate reviewed CodeAF fixes — batch 2 #1632. With Consolidate reviewed CodeAF fixes — batch 2 #1632's manual diff applied on top of this branch, the tree builds, go vet ./internal/manual is clean, and internal/manual passes, including Consolidate reviewed CodeAF fixes — batch 2 #1632's three new Traffic probes. None of these edits overlaps a Consolidate reviewed CodeAF fixes — batch 2 #1632 hunk.

How it was verified

make pr-ready BASE=origin/dev is green, which covers the light gate, the manual gates and the touched package. go test ./internal/manual/ ./internal/namelaw/ passes, and the combination with #1632 was checked as described above. Each row's decision comes from reading the code on dev.

What it deliberately does not do

• It changes no code. Rows 2, 3 and 4 are code defects and need their own issues:
  • The engine road's agent lacks StandingHere, so the count line, the side column and /standing get nothing there. This probably shares a cause with standing: place scoping mismatch for project orders, stale budget on consent card, and activity label #1555's side-column rows.
  • The remember and stand descriptions overlap.
  • The standing card never draws the grant.
• team_start's tool description still says "The person is asked first", which is false under ◇ YOLO. That is Manager tells the person it asked for approval when nobody was asked #1585.
• Stale code comments are left as they are, at cmd/codeaf/poolrecord.go:5 (the crew reads the own sheet) and internal/tui3/teammenu.go:35 (Remove manager).

🤖 Generated with Claude Code