feat(terminal): send text to the Claude pane (:ClaudeCodeSendText) (#197)

[ThomasK33]

Summary

Adds a way to send arbitrary text to the currently-open Claude Code terminal pane, as if typed at the prompt — the feature requested in #197 (for scripting and keymaps).

• :ClaudeCodeSendText {text} — types {text} into the open Claude terminal and submits it. :ClaudeCodeSendText! (bang) inserts without submitting.
• require("claudecode.terminal").send_to_terminal(text, opts) — the programmatic primitive (opts.submit default true, opts.focus default false).

It writes to the terminal's job channel with vim.fn.chansend; a trailing carriage return (\r) is what submits at Claude's prompt. No window focus is required, which is the point for scripting.

Closes #197.

How it works

chansend into a Neovim :terminal writes to the same PTY master Neovim uses when forwarding a real keypress, so chansend(chan, "\r") is byte-identical to pressing Enter and submits. (This is not affected by anthropics/claude-code#15553, which is about VS Code's terminal.sendText — a different layer.)

Implementation notes:

• Channel resolved from the plugin's own get_active_terminal_bufnr() → b:terminal_job_id, with a bo.channel fallback for a recovered terminal whose module-level job id was lost.
• Gated on nvim_buf_is_valid + a resolvable, non-zero channel; the chansend write result is honored (a closed channel → warn + false), so the boolean return is truthful.
• Multi-line text is wrapped in one bracketed-paste block (ESC[200~ … ESC[201~) and line endings are normalized, so interior newlines/CRs don't fire premature submits; the submit \r is sent after the closing marker. (chansend bypasses vim.paste, so the fix_streamed_paste shim is irrelevant here.)
• In-editor providers only (native/snacks). external/none run Claude outside Neovim — there's no pane to write to, so the call warns and returns false (same structural limit as focus_after_send; see the ClaudeCodeSendComplete event for focusing an out-of-editor session).

Verification

Verified end-to-end against the real Claude CLI (2.1.169) driving the plugin's native terminal in an isolated PTY (cursor never left a separate buffer — no focus needed):

Scenario	Result
:ClaudeCodeSendText What is 7 times 6? …	submitted → Claude replied 42
:ClaudeCodeSendText! this is a draft …	inserted into the prompt, not submitted
send_to_terminal("…\nWhat is 9 plus 10?")	sent as one bracketed-paste block → submitted once → 19

Tests: 22 new unit tests (tests/unit/terminal/send_text_spec.lua, tests/unit/claudecode_send_text_command_spec.lua) covering single/multi-line, submit/bang, CRLF + lone-CR normalization, the terminal_job_id→bo.channel fallback (incl. job id 0), nil/stale channel, focus, and the external/none warnings. mise run all is green (608 tests, lint + format clean).

Relates to #234 (not closed here)

#234 ("send prompt with selection") builds directly on this primitive, and the content composition works — live testing confirmed that broadcasting the selection's at_mention (Claude inserts @file#Lx-y) and then chansend-ing the prompt produces the correct @file#Lx-y <prompt> and Claude reads the file. However, an @-mention in Claude's input box (whether MCP-inserted or typed) absorbs the immediately-following \r, so reliable auto-submit needs a separate, timing-gapped \r after the prompt is inserted. That extra timing dependency (against an async TUI, plus a dependency on undocumented CLI mention-insertion behavior) is fragile enough that I've kept it out of this PR rather than bundle a flaky auto-submit. This PR provides the clean primitive #234 needs; a focused follow-up can layer the selection+prompt UX (e.g. :ClaudeCodeSendWithPrompt via vim.ui.input) on top, with the 2-stage submit and a configurable delay.

Notes

• :ClaudeCodeSendText sends to the currently-open pane; it does not auto-launch Claude (open it with :ClaudeCode first). This keeps behavior predictable — a deferred send into a just-spawned CLI that isn't input-ready yet would silently drop.

🤖 Generated with Claude Code

[ThomasK33]

@codex review

@claude review

[chatgpt-codex-connector]

Codex Review: Didn't find any major issues. Breezy!

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".