Build the eight asks recovered from the previous session's transcript

The punch list was not the whole record. These are the owner's own typed
requests, read back out of the previous session's transcript rather than
guessed at, then each checked against the code before being treated as open.
Three they suspected were already done really were: dialogs already had a
Close-only titlebar, inactive dimming already existed, and the corner resize
hitbox had already been tuned.

Snap layouts on drag, asked for twice. Edge snapping worked but committed
silently on release with nothing shown first, so there was no way to know it
would happen or where. A translucent drop-target preview now follows the
drag, and throwing the pointer at a monitor's top edge drops down the
existing six-cell grid to aim at. The preview calls the same snap_zone that
end_drag does, so the two cannot disagree. Two defects found by screenshot
before landing: moving down onto the flyout closed it, and its labels
overflowed at a fixed cell width - the same "text goes out of view" fault
already fixed once for the context menu.

New File now offers real types, chosen by extension, with the de-duplication
counter placed before the extension so the file stays what it says it is.

Refresh re-reads init.lua and fires a new srd.on("refresh") handler instead
of only re-scanning the icon grid. What refresh means beyond srdwm's own
config stays the config's decision.

general.config_reload_on_write (default on) applies an edited config on save,
via an mtime sweep rather than an inotify watch: no new dependency, same
behaviour on every target, and unaffected by editors that write through a
temp file.

A real bug behind "what happens when our config fails": you lost every
keybinding. do_reload cleared the binding, handler and repeat tables before
re-executing and never restored them, so a syntax error left neither the old
config nor the new one, and the only key still working was the reload combo
nobody thinks to press. The tables are now restored on any failure and
config errors reach notify-send, not just the log.

srd.lock() and a default Mod4+Ctrl+l binding: the built-in lock screen could
not be reached from Lua at all. Native rather than shelling out, because a
lock key that shells out fails silently when the binary is not on PATH.
Default bindings added for srd.window.move and a dynamic/tiling toggle, both
of which existed with no way to reach them, plus srd.layout.get() so the
toggle reads the live workspace rather than the configured default.

Dialogs open centred, and are excluded from remembered geometry in both
directions - that table is keyed by app_id, which a dialog shares with the
window that spawned it, so dialogs inherited an unrelated position and size
and then overwrote it with their own.

theme.decorations.title_bar.button_mode (dynamic by default, or fixed) drops
the Maximize button on a window whose client pinned min == max size, where
pressing it can do nothing. Maximize is removed from the slot list rather
than skipped in place on both the render and hit-test sides, so the remaining
buttons close the gap identically; three tests pin that agreement, which is
what fails silently when it drifts.

Also: the nested backend's screencopy pass now draws both menus, the flyout
and the drag preview. Four investigations in one day started from a
screenshot missing a tier, so that pass carries an explicit list of what it
still omits and the on-screen loop points at it.

512 tests pass, clippy clean.

2 files changed, 155 insertions, 0 deletions

diff --git a/docs/DEFAULTS.md b/docs/DEFAULTS.md
index 3d1a863..a130434 100644
--- a/docs/DEFAULTS.md
+++ b/docs/DEFAULTS.md
@@ -19,6 +19,7 @@ srd.set("general.auto_raise", false)                   -- Default: false - also
 srd.set("general.gpu", false)                          -- Default: false - udev backend only, see "GPU rendering" below
 srd.set("general.desktop_icons", true)                 -- Default: true - see "Desktop icons" below
 srd.set("general.desktop_icons_all_monitors", true)    -- Default: true - mirror icons onto every monitor, not just primary
+srd.set("general.config_reload_on_write", true)         -- Default: true - re-read init.lua when it changes on disk
 srd.set("general.reserve_top", 0)                      -- Default: 0 - static space reserved before any bar/dock connects
 srd.set("general.reserve_bottom", 0)                   -- Default: 0 - see "Startup space reservation" below
 srd.set("general.reserve_left", 0)                     -- Default: 0
@@ -453,8 +454,18 @@ srd.set("theme.decorations.title_bar.button_side", "right")   -- Default: "right
 srd.set("theme.decorations.title_bar.button_order", "")        -- Default: "" (unset)
 srd.set("theme.decorations.title_bar.button_glyph", "hover")   -- Default: "hover"
 srd.set("theme.decorations.title_bar.button_style", "traffic_lights")  -- Default: "traffic_lights"
+srd.set("theme.decorations.title_bar.button_mode", "dynamic")  -- Default: "dynamic"
 ```
 
+`button_mode` sets `"dynamic"` (the default) to show only the buttons a
+window can actually use, or `"fixed"` to always show the full set. Today
+dynamic mode has one rule: a window whose client pinned its minimum and
+maximum size to the same value gets no Maximize button, because pressing it
+can do nothing. GNOME, KDE and Windows all hide or disable maximize in the
+same case. A dialog's Close-only titlebar is a separate rule and applies in
+both modes. Live-settable with `srd set button_mode <dynamic|fixed>`, and
+readable back from `srd settings`.
+
 `text_align` sets `"center"` for the macOS convention (title centered on
 the whole titlebar width, ignoring the button cluster the way real macOS
 does) or `"left"` (default) for the Windows/GTK convention. Any value
@@ -832,3 +843,34 @@ local elapsed = srd.debug.profile_stop()
 This documentation provides a comprehensive reference for all default values and configuration options in SRDWM.
 
 
+
+## Config reloading and what happens when a config breaks
+
+A Lua config is a program, so breaking it is an ordinary event rather than
+an exceptional one. Three things make that safe.
+
+**A failed reload changes nothing.** Reloading clears the keybinding, event
+handler and repeat-key tables before re-running `init.lua`, so that a
+binding deleted from the file really disappears. If the new file fails to
+parse or errors while running, all three tables are put back exactly as they
+were. The last working config keeps running. Whatever the broken run managed
+to register before it failed is discarded rather than merged, because half a
+config is not a config.
+
+**The error is shown, not just logged.** Config failures go to
+`notify-send` as well as the log. Without that the failure is close to
+silent from the user's side: the compositor keeps running and the edit simply
+does nothing.
+
+**Edits apply on save.** `general.config_reload_on_write` (default `true`)
+checks the config directory's `.lua` modification times once a second and
+reloads when one changes. Set it to `false` for a config that does expensive
+work at load time. `Mod4+Ctrl+r` still reloads on demand in either case.
+
+Two limits to know. A reload does not re-register key *grabs* with the
+backend, so a brand new key combination needs a restart before the
+compositor sees that key at all; an existing combination picks up its new
+action immediately. And a reload rebuilds the theme from the config file, so
+it discards live `srd set` theme changes - including titlebar settings
+changed through the right-click menu. That is the correct precedence, but
+with reload-on-write it now happens every time the config is saved.
diff --git a/docs/TODO.md b/docs/TODO.md
index 72b3d20..49f7312 100644
--- a/docs/TODO.md
+++ b/docs/TODO.md
@@ -1,5 +1,118 @@
 # TODO / planned features - master checklist
 
+## Eight asks recovered from the previous session's transcript, all built (2026-08-28)
+
+Reported as "there was more stuff from previous agent", then "please do all of
+them ... i also stated many issues ie titlebar inconsistencies etc". The list
+was rebuilt by reading the owner's own typed messages out of the previous
+session's transcript rather than by guessing, then each item was checked
+against the code before being called open. Three things they suspected were
+already done really were done - dialogs already got a Close-only titlebar,
+inactive-window dimming already existed, and the corner resize hitbox had
+already been tuned. The eight below had not been built.
+
+**1. Windows-style snap layouts on drag.** Asked twice ("if you move the
+window to absolute north it show you layout options", then "why do i still
+not see the windows layout ... when moved to areas of screen like in
+windows"). Edge snapping already worked but committed silently on release
+with nothing shown beforehand, so there was no way to know it was about to
+happen or where. Now: `WindowManager::drag_snap_preview` returns the rect the
+window will land in, drawn as a translucent accent fill with a solid outline
+(`elements::snap_preview_elements`, native solid fills - a preview can be a
+whole monitor in size and changes as the drag moves, so rasterising a bitmap
+per zone change would allocate megabytes); and throwing the pointer at a
+monitor's top edge drops down the existing six-cell Snap-Layouts grid to aim
+at. The preview calls the very same `SmartPlacement::snap_zone` that
+`end_drag` does, so preview and commit cannot disagree.
+
+Two defects found by screenshot and fixed before landing: moving down onto
+the flyout closed it (the pointer had left the trigger band), and its cell
+labels overflowed at the fixed 90px width - "Bottom Right" was cut off
+mid-word, the same "text goes out of view" fault already reported and fixed
+for the context menu. The flyout now grows to fit its widest label exactly as
+`open_context_menu` already did.
+
+**2. New File with a real type choice.** "in context menu say new file, user
+can choose what file type is obviously by extension". The desktop menu had
+`New Folder` and `New Text Document` only. Five more types now sit under
+them, each creating an empty file with the right extension. The de-duplication
+counter goes before the extension (`New Shell Script (2).sh`), since a
+suffixed extension would stop being one.
+
+**3. Refresh actually refreshes something.** "does refresh refresh configs in
+a function list in the config ie refresh os, etc, ags/aegis/polybar/waybar".
+Refresh re-scanned the desktop icon grid and nothing else. It now also
+re-reads `init.lua` and fires a new `srd.on("refresh", ...)` handler, so the
+config decides what else to reload. What "refresh" means beyond srdwm's own
+config is deliberately the user's decision - this compositor has no business
+hardcoding whether they run waybar or AGS.
+
+**4. Config reload on write.** "do we suppport update config on write" - no,
+it did not. `general.config_reload_on_write` (default on) polls the config
+directory's `.lua` mtimes once a second and reloads on a change. A `stat`
+sweep rather than an inotify watch: no new dependency, identical behaviour on
+every platform this project targets, and immune to the editor-writes-a-temp-
+file-and-renames pattern that defeats watches on individual files.
+
+**5. What happens when a config fails - a real bug, fixed.** Asked as "what
+happens when our config fails/user does something wrong which can be expected
+since lua programmable config". The answer the code gave was: **you lose every
+keybinding.** `do_reload` cleared `key_bindings`/`event_handlers`/
+`repeat_keys` before re-executing and never restored them, so a Lua syntax
+error - the most likely thing to go wrong with a programmable config - left
+neither the old bindings nor the new ones. The only key still working was the
+hardcoded reload combo, which is the one key nobody thinks to press, because
+nothing said that was the situation. The three maps are now moved out and put
+back on any failure, so a broken edit leaves the last working config running,
+and config errors go to `notify-send` rather than only to a log. Verified live
+in a nested compositor: breaking the config logged "Config edit not applied,
+keeping the last working one", the compositor kept serving IPC, and fixing the
+file reloaded for real.
+
+**6. A lock-screen keybinding.** "do we have a lockscreen binding?" - there
+was no way to reach the built-in lock screen from Lua at all, and the shipped
+config's only lock key ran an external script. New native `srd.lock()`, bound
+to `Mod4+Ctrl+l` by default. Native rather than shelling out to the control
+CLI, because a lock binding that shells out fails silently when that binary is
+not on `PATH`.
+
+**7. Dialogs open centred.** "even dialog/starter windows spawn that side,
+most times should be centered". Dialogs now centre on the target monitor's
+usable area, and are excluded from the remembered-geometry path in both
+directions: `remembered_geometry` is keyed by `app_id`, which a dialog shares
+with the window that spawned it, so a dialog was being given that app's last
+main-window position *and size* - and would then overwrite that memory with
+its own small rect.
+
+**8. Titlebar buttons that follow the program.** "ideally we can also set
+titlebar to have inhouse decorations/buttons of the program/dynamic". New
+`theme.decorations.title_bar.button_mode` (`dynamic`, the default, or
+`fixed`), live-settable via `srd set button_mode` and readable back through
+`srd settings`. In dynamic mode a window whose client pinned min == max size
+gets no Maximize button, because pressing it can do nothing - what GNOME, KDE
+and Windows all do. Maximize is removed from the slot list rather than skipped
+in place, in both the renderer and the hit-test, so the remaining buttons
+close the gap identically on both sides; three tests pin that agreement, which
+is the part that fails silently when it drifts.
+
+Also fixed, and the reason several of these took two attempts: the nested
+backend's screencopy pass was still missing tiers. It now draws the titlebar
+context menu, the desktop menu, the Snap-Layouts flyout and the drag snap
+preview as well as the popups and shadows added earlier today. Four separate
+investigations in one day started from a screenshot that was quietly lying --
+`winit/capture.rs` now carries an explicit list of what it still does not
+draw (border strips and the desktop icon grid), and `winit/render.rs` carries
+a pointer to it at the place a new tier gets added.
+
+**One interaction worth knowing.** A config reload rebuilds `ThemeConfig` from
+the config file, so it discards live `srd set` theme changes - correct
+precedence, and pre-existing, but auto-reload makes it happen on every save
+rather than only when the reload key is pressed. A titlebar customised live
+through the right-click menu reverts the next time `init.lua` is saved.
+
+Full workspace build/test/clippy clean: 512 tests (262 core / 160 wayland /
+43 platform / 34 config / 13 ctl), 0 failed, 0 clippy warnings.
+
 ## Nemo's right-click menu: confirmed working, and two real bugs found doing it (2026-08-28)
 
 The last open punch-list item. It is closed on a real end-to-end repro, not