chore(nvim): close cheatsheet gaps, name which-key prefixes, drop dead config

[Gerrrt]

The two deferred items from the second review pass: documentation drift and dead plugin config. No behaviour change except the harpoon guard.

Cheatsheet

The panel's header claims it lays out every curated binding. An audit against the real keymaps found it didn't — two entire features had no rows:

• Completion (blink.cmp) — new card. The seven completion-menu keys (<C-Space>, <CR>, <C-k>/<C-j>, <C-e>, <C-b>/<C-f>, <Tab>/<S-Tab>) had no representation at all.
• Move lines (mini.move) — new card for <A-hjkl>.

Also added <leader>bn/bp, <leader>cl (Trouble LSP refs/defs), ]t/[t, gsh, gsn.

Two descriptions corrected: <leader>rc said "Edit init.lua" where the keymap's desc is "Edit config"; <leader>e dropped the load-bearing "closes Zen if active" side effect.

The header now scopes its completeness claim — stating what is deliberately excluded (transient-UI keys: the rename float, oil buffers, alpha's buttons, the panel's own q/<Esc>) rather than overstating and drifting again.

which-key

<leader>r (edit config, rename symbol), <leader>o (organize imports) and <leader>p (copy file path) had real children but no group entry, so they rendered as bare unnamed prefixes. <leader>p is declared for normal mode only — in visual it is itself a mapping (paste-without-yank), not a prefix.

Dead config

Each verified against the installed plugin source, not assumed:

Change	Why
bufferline: drop hover.reveal = { "close" }	get_close_icon() (ui.lua:263-270) consults reveal, then unconditionally bails on if not options.show_buffer_close_icons then return end — false here, so there was never a close icon to reveal. hover stays on for highlighting.
nvim-tree: adaptive_size = true → width = {}	A 2023-01-15 legacy key silently rewritten by legacy.lua:73-81; with no explicit width it yields { min = nil }, i.e. {}. Verified by running the migration and comparing deep-equal.
fidget: winblend 0 → 100	100 is fidget's default and its see-through setting; the old comment claimed 0 matched "transparent floats", which is backwards (notification/window.lua:33-49).
nvim-treesitter-context: drop separator = nil	nil in a table literal omits the key, and nil is already the default — it read as a setting but did nothing.
conform: drop the config function	It re-implemented lazy.nvim's default for a spec with opts. Verified the 20 formatters_by_ft entries still apply without it.
blink.cmp: fix a comment	It claimed the snippet keys use native vim.snippet; with preset = "luasnip" they route to LuaSnip.

Harpoon

<leader>ha on an unnamed buffer stored an unnavigable empty entry and toasted a bare "Harpoon: added ". It now refuses with a warning.

Validation

Check	Result
luacheck lua init.lua	0 warnings / 0 errors, 86 files
nvim --headless -c 'qa!'	exit 0, no stderr
Cheatsheet opens	renders 90 lines; asserted every new row is present
7 changed plugins force-loaded	no errors in :messages
conform opts without explicit config	20 formatters_by_ft entries, lua -> { "stylua" }
make audit	222 pass / 0 fail

One audit check skipped (yaml parse — PyYAML not importable); environmental and pre-existing.

🤖 Generated with Claude Code

https://claude.ai/code/session_018wpwbjkGJKPFgvBFEKsFWE

[Copilot]

Pull request overview

Aligns Neovim documentation and plugin configuration with current behavior while preventing invalid Harpoon entries.

Changes:

• Expands the cheatsheet and names missing which-key prefixes.
• Removes redundant or legacy plugin configuration.
• Rejects unnamed buffers when adding Harpoon entries.

Reviewed changes

Copilot reviewed 10 out of 10 changed files in this pull request and generated 2 comments.

Show a summary per file

File	Description
CHANGELOG.md	Records Neovim changes.
nvim/lua/gerrrt/cheatsheet.lua	Adds missing binding references.
nvim/lua/gerrrt/plugins/blink-cmp.lua	Corrects snippet-engine commentary.
nvim/lua/gerrrt/plugins/bufferline-nvim.lua	Removes unreachable hover reveal configuration.
nvim/lua/gerrrt/plugins/conform.lua	Uses lazy.nvim’s default setup behavior.
nvim/lua/gerrrt/plugins/fidget-nvim.lua	Restores the default transparent window blend.
nvim/lua/gerrrt/plugins/harpoon.lua	Guards against unnamed buffers.
nvim/lua/gerrrt/plugins/nvim-tree.lua	Replaces a legacy adaptive-width option.
nvim/lua/gerrrt/plugins/treesitter-context.lua	Removes an ineffective nil option.
nvim/lua/gerrrt/plugins/which-key.lua	Names three previously unnamed prefixes.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[Gerrrt]

Both addressed in 44feda2. Both were mistakes I introduced in this PR.

1. <C-s> in the Completion card — correct, removed.

Confirmed: it's mapped at utils/lsp.lua:121, and blink's keymap block has zero occurrences of it (the block ends at <S-Tab>, as you said). It was already listed in the LSP & Code card, so the new row both duplicated it and contradicted that card's own note that every key in it is set explicitly in the blink spec.

Removed, with a comment recording why it isn't there — blink's signature window is automatic via signature.enabled and has no key of its own, which is what I'd confused it with.

The two remaining <C-s> rows are distinct and both correct: Flash in command-line search (c mode) and signature help (i mode). Verified in the rendered panel:

<C-s>  Toggle Flash in / search…
<C-s>  Signature help (i)
total C-s rows rendered = 2

2. width = {} is not "no fixed min/max" — correct, and my wording was wrong.

The migration equivalence still holds, but the explanation didn't. nvim-tree fills the absent keys with its own defaults — width.min or DEFAULT_MIN_WIDTH (30) and width.max or DEFAULT_MAX_WIDTH (-1) at view-state.lua:5-6,77-78. So the pane sizes to content but never narrower than 30 columns. Confirmed at runtime:

with width = {} -> Active.width = 30   max_width = -1

Both the inline comment and the CHANGELOG entry now say that, and point at where the defaults come from so the next reader doesn't have to take it on faith.

Worth noting the pattern: I verified the equivalence of the migration carefully and then wrote an unverified sentence about what the result means. Same failure mode as the vimade claim on #257 — checking one thing and asserting a neighbouring one.

Re-validated: luacheck 0/0 across 86 files, cheatsheet renders 90 lines, make audit 222 pass / 0 fail.