fix: preserve assistant editor input and extend release PTY gates

README.md

@@ -20,7 +20,7 @@ The release and Nix flake support `x86_64-linux` only. With
 [Nix](https://nixos.org/download) and flakes enabled:
 
 ```sh
-nix run github:nerima-lisp/nshell/v0.6.3
+nix run github:nerima-lisp/nshell/v0.6.4
 ```
 
 At the prompt, enter `string upper hello`. Commands and paths colorize live, and a
@@ -62,16 +62,16 @@ and session settings.
 ## Install
 
 ```sh
-nix profile install github:nerima-lisp/nshell/v0.6.3
+nix profile install github:nerima-lisp/nshell/v0.6.4
 ```
 
 Pin a release tag rather than following the default branch.
 
 To install without Nix, use the `x86_64-linux` release bundle. It removes Nix store references, carries its
 ELF runtime library closure, and is checked for required files and dependency
 metadata. CI checks an extracted archive and exercises its executable with a
-real PTY, including editing, signals, job control, history restart, and terminal
-restoration. See [Getting
+real PTY, including multiline editing and paste, completion with spaces,
+signals, job control, resize, history restart, and terminal restoration. See [Getting
 started](https://nerima-lisp.github.io/nshell/getting-started/) for the bundle
 verification and installation procedure.
 

docs/src/getting-started.md

@@ -9,13 +9,13 @@ Install [Nix](https://nixos.org/download) and enable its `nix-command` and
 `flakes` experimental features before running:
 
 ```sh
-nix run github:nerima-lisp/nshell/v0.6.3
+nix run github:nerima-lisp/nshell/v0.6.4
 ```
 
 ## Install
 
 ```sh
-nix profile install github:nerima-lisp/nshell/v0.6.3
+nix profile install github:nerima-lisp/nshell/v0.6.4
 man nshell   # the manual page is installed alongside the binary
 nshell
 ```
@@ -26,34 +26,34 @@ leave the interactive shell.
 ### Prebuilt Linux bundle
 
 Without Nix, download both assets from the
-[v0.6.3 release](https://github.com/nerima-lisp/nshell/releases/tag/v0.6.3),
+[v0.6.4 release](https://github.com/nerima-lisp/nshell/releases/tag/v0.6.4),
 verify the checksum, and extract the bundle. On `x86_64-linux`, with `curl`,
 `sha256sum`, and `tar` available, run these commands in an empty directory:
 
 ```sh
-curl --fail --location --remote-name https://github.com/nerima-lisp/nshell/releases/download/v0.6.3/nshell-v0.6.3-x86_64-linux.tar.gz
-curl --fail --location --remote-name https://github.com/nerima-lisp/nshell/releases/download/v0.6.3/nshell-v0.6.3-x86_64-linux.tar.gz.sha256
-sha256sum --check nshell-v0.6.3-x86_64-linux.tar.gz.sha256
+curl --fail --location --remote-name https://github.com/nerima-lisp/nshell/releases/download/v0.6.4/nshell-v0.6.4-x86_64-linux.tar.gz
+curl --fail --location --remote-name https://github.com/nerima-lisp/nshell/releases/download/v0.6.4/nshell-v0.6.4-x86_64-linux.tar.gz.sha256
+sha256sum --check nshell-v0.6.4-x86_64-linux.tar.gz.sha256
 ```
 
 Proceed only if the checksum command exits successfully and prints
-`nshell-v0.6.3-x86_64-linux.tar.gz: OK`:
+`nshell-v0.6.4-x86_64-linux.tar.gz: OK`:
 
 ```sh
-tar -xzf nshell-v0.6.3-x86_64-linux.tar.gz
-./nshell-v0.6.3-x86_64-linux/bin/nshell --version
-man ./nshell-v0.6.3-x86_64-linux/share/man/man1/nshell.1
-./nshell-v0.6.3-x86_64-linux/bin/nshell
+tar -xzf nshell-v0.6.4-x86_64-linux.tar.gz
+./nshell-v0.6.4-x86_64-linux/bin/nshell --version
+man ./nshell-v0.6.4-x86_64-linux/share/man/man1/nshell.1
+./nshell-v0.6.4-x86_64-linux/bin/nshell
 ```
 
 Keep the extracted directory intact: the launcher resolves libraries and
-helpers relative to itself. The version output must identify `v0.6.3`.
+helpers relative to itself. The version output must identify `v0.6.4`.
 Type `exit` to return to your existing shell. To use `nshell` by name in
 the examples below, add the extracted `bin` directory to that shell's
 `PATH`. For Bash or Zsh, from the download directory:
 
 ```sh
-export PATH="$PWD/nshell-v0.6.3-x86_64-linux/bin:$PATH"
+export PATH="$PWD/nshell-v0.6.4-x86_64-linux/bin:$PATH"
 ```
 
 This changes only the current terminal session. Alternatively, replace
@@ -95,14 +95,14 @@ nshell -c 'string upper hello'
 ### Run a script
 
 ```sh
-curl --fail --location --output greet.nsh https://raw.githubusercontent.com/nerima-lisp/nshell/v0.6.3/examples/greet.nsh
+curl --fail --location --output greet.nsh https://raw.githubusercontent.com/nerima-lisp/nshell/v0.6.4/examples/greet.nsh
 nshell greet.nsh World
 ```
 
 Script files support multiline blocks (functions, `if`/`for`/`while`/`switch`),
 comments, and a `#!` shebang; arguments after the script name are available as
 `$argv`. See
-[`examples/`](https://github.com/nerima-lisp/nshell/tree/v0.6.3/examples) for a
+[`examples/`](https://github.com/nerima-lisp/nshell/tree/v0.6.4/examples) for a
 runnable sample.
 
 ### Command line
@@ -131,7 +131,7 @@ The Nix build supplies [SBCL](http://www.sbcl.org/), ASDF, and the dependencies.
 On `x86_64-linux`, with Git and flake-enabled Nix installed:
 
 ```sh
-git clone --branch v0.6.3 https://github.com/nerima-lisp/nshell
+git clone --branch v0.6.4 https://github.com/nerima-lisp/nshell
 cd nshell
 nix build
 ./result/bin/nshell

docs/src/guide/recipes.md

@@ -2,19 +2,16 @@
 
 ## Draw a pipeline without running it
 
-`pipeline-graph` renders a typed pipeline as a validated Graphviz DOT graph — or
-a Mermaid flowchart with `--mermaid` — without executing it. Quote the pipeline
+`pipeline-graph` renders a pipeline as a Graphviz DOT graph, or a Mermaid
+flowchart with `--mermaid`, without executing it. Quote the pipeline
 so the shell passes it as a single argument rather than running it:
 
 ```sh
 pipeline-graph 'cat access.log | grep 404 | wc -l'
 pipeline-graph --mermaid 'cat access.log | grep 404 | wc -l'
 ```
 
-It is a diagnostic built on the
-[cl-dataflow-kit](https://github.com/nerima-lisp/cl-dataflow-kit) computation-graph
-toolkit, and it validates the pipeline as it builds the graph, so a malformed
-redirect shows up as an error rather than a picture.
+Malformed redirects are reported as errors instead of rendered as graphs.
 
 ## Write a script
 
@@ -27,13 +24,17 @@ end
 greet $argv[1]
 ```
 
-Run it with `nshell greet.nsh World`. Multiline blocks (`function`, `if`, `for`,
+Save this as `greet.nsh` and run `nshell greet.nsh World`. Multiline blocks (`function`, `if`, `for`,
 `while`, `switch`, `begin`/`end`), comments, and a `#!` shebang all work;
 arguments after the script name arrive as `$argv`.
 
 ## Use the assistant
 
-Start an interactive assistant task from a terminal:
+The assistant requires an authenticated `claude` CLI on PATH. Set
+`NSHELL_AI_COMMAND` to use another executable implementing the same streaming
+protocol. `NSHELL_AI_DISABLE` disables the assistant.
+
+Start an interactive assistant task from an nshell terminal:
 
 ```sh
 agent "inspect the current job state"
@@ -45,22 +46,46 @@ commands are not executed. Approval applies to one proposal, not the rest of
 the task. Review each command, especially in a directory containing sensitive
 data.
 
-The assistant refuses to start if its sidecar reports tools or MCP servers
-outside the permitted initialization contract, even if a later initialization
-event reports an acceptable configuration.
+The safety label comes from static command analysis, not a security sandbox.
+For example, recursive deletion of the filesystem root is blocked. A safe label
+does not replace your review of the command, its arguments, and its working
+directory.
+
+Before command approval, the assistant receives your task and context: the last
+command, its exit status and duration, the working directory, Git branch and
+dirty state, a bounded tail of command output, and environment-variable names.
+Known sensitive paths, commands, and environment values are redacted, but this
+does not guarantee that task text or command output contains no secrets. Do not
+submit confidential information unless the configured provider is permitted to
+receive it.
+
+At the review panel, press `e` to edit, `s` to skip the proposal, or `q` to cancel
+the task. While editing, `q` and `s` insert text; Ctrl-C cancels the task. Enter
+reclassifies the edited command and approves it for execution if it is not
+blocked.
+
+The sidecar may report only the `StructuredOutput` tool (or no tools) and no MCP
+servers. Both fields must be present in its initialization event. Other
+configurations are rejected even if a later event reports an acceptable one.
 
 Inspect assistant state and configure the session with `ai`:
 
 ```sh
 ai status
 ai set effort medium
-ai set model 'model-name'
 ai set max-steps 20
-ai set budget 1.50
 ai log 10
-ai reset
 ```
 
+`ai log 10` shows the last ten audit entries. `ai reset` resets the conversation
+and usage counters. `ai set model 'model-name'` selects a model identifier
+supported by your CLI; without a setting, the CLI chooses its default model.
+The default maximum is ten proposals per task. Skipped and blocked proposals
+also count toward `max-steps`; editing a proposal does not add a step.
+
+`ai set budget 1.50` sets a US-dollar reference amount for the usage display.
+It does **not** enforce a spending limit or stop provider requests.
+
 `ai set` changes the current session. The corresponding environment variables
 are `NSHELL_AI_EFFORT`, `NSHELL_AI_MODEL`, `NSHELL_AI_MAX_STEPS`, and
 `NSHELL_AI_BUDGET_USD`.
@@ -85,7 +110,7 @@ perl -e '$SIG{ALRM}=sub { exit 124 }; alarm 300; exec @ARGV' nix run .#test
 perl -e '$SIG{ALRM}=sub { exit 124 }; alarm 300; exec @ARGV' sbcl --script run-tests.lisp
 ```
 
-Just the focused completion suite (`nshell/weave`):
+Just the focused completion suite (`nshell/weave`), inside `nix develop`:
 
 ```sh
 perl -e '$SIG{ALRM}=sub { exit 124 }; alarm 300; exec @ARGV' sbcl --script scripts/weave.lisp

docs/src/project/releasing.md

@@ -52,7 +52,7 @@ tools on the Linux verification host.
   manifest, store-reference hygiene, platform library closure, and smoke
   startup. Do not replace this with a check of the unbundled default package.
 - `./result/bin/nshell --version` reports the intended version.
-- `./result/bin/nshell --help` and `man ./man/nshell.1` match the documentation
+- `./result/bin/nshell --help` and `man ./result/share/man/man1/nshell.1` match the documentation
   and shipped behaviour.
 - Package the built bundle using GNU tar and gzip, as the release workflow
   does. Replace `vX.Y.Z` with the intended tag:
@@ -80,8 +80,9 @@ tools on the Linux verification host.
   perl scripts/test-release-pty.pl "$extracted/$name"
   ```
 
-  The PTY check exercises editing, Ctrl-C, Ctrl-Z/`bg`/`fg`, history across
-  restart, and terminal restoration. A source-loaded suite is not a substitute.
+  The PTY check exercises multiline editing and paste, completion with spaces,
+  Ctrl-C, Ctrl-Z/`bg`/`fg`, resize during editing, history across restart, and
+  terminal restoration. A source-loaded suite is not a substitute.
 - Release notes are drafted. Review changes since the previous tag and describe
   user-visible features, fixes, and any required migration. After the
   workflow goes green, paste them in and publish the draft:

man/nshell.1

@@ -1,4 +1,4 @@
-.TH NSHELL 1 "2026-09-30" "nshell 0.6.3" "User Commands"
+.TH NSHELL 1 "2026-09-30" "nshell 0.6.4" "User Commands"
 .SH NAME
 nshell \- a fish-inspired interactive shell written in Common Lisp
 .SH SYNOPSIS

nshell.asd

@@ -19,7 +19,7 @@
   :license "MIT"
   ;; Single source of truth for the version. flake.nix reads this form
   ;; line-by-line, and release.yml refuses a tag that disagrees with it.
-  :version "0.6.3"
+  :version "0.6.4"
   :homepage "https://github.com/nerima-lisp/nshell"
   :bug-tracker "https://github.com/nerima-lisp/nshell/issues"
   :source-control (:git "https://github.com/nerima-lisp/nshell.git")
@@ -403,7 +403,7 @@
   :in-order-to ((test-op (test-op "nshell/test"))))
 
 (asdf:defsystem "nshell/test"
-  :version "0.6.3"
+  :version "0.6.4"
   :author "takeokunn <bararararatty@gmail.com>"
   :maintainer "takeokunn <bararararatty@gmail.com>"
   :license "MIT"
@@ -590,7 +590,7 @@
                 (error "cl-weave tests failed"))))
 
 (asdf:defsystem "nshell/weave"
-  :version "0.6.3"
+  :version "0.6.4"
   :author "takeokunn <bararararatty@gmail.com>"
   :maintainer "takeokunn <bararararatty@gmail.com>"
   :license "MIT"
@@ -620,7 +620,7 @@ primary suite in nshell/test."
                               :reporter :spec)
                (error "cl-weave suite failed"))))
 (asdf:defsystem "nshell/benchmark"
-  :version "0.6.3"
+  :version "0.6.4"
   :author "takeokunn <bararararatty@gmail.com>"
   :maintainer "takeokunn <bararararatty@gmail.com>"
   :license "MIT"

scripts/test-release-pty.pl

@@ -188,7 +188,7 @@
         expected = ('rel-%s:<%s>' % (label, value)).encode()
         start = len(self.output)
         self.send(command)
-        match, end = self.expect(re.escape(expected), start, command)
+        match, end = self.expect(re.escape(expected).replace(re.escape(b'\n'), rb'\r?\n'), start, command)
         self.expect(re.escape(prompt), end)
         return command
 
@@ -335,6 +335,73 @@
     session.wait(lambda: session.foreground() == pid)
     interrupt_job(pid)
     check(stage)
+    stage = 'multiline cross-line editing'
+    label = 'multiline-' + nonce
+    line = ("printf 'rel-%%s:<%%s>\\n' %s 'Lx\nR'" % label).encode()
+    command = b'\x1b[200~' + line + b'\x1b[201~' + b'\x1b[D' * 3 + b'\x08' + b'\x1b[C' * 3 + b'\n'
+    start = len(session.output)
+    session.send(command)
+    _, end = session.expect(re.escape(('rel-%s:<L' % label).encode()) + rb'\r?\nR>', start, command)
+    session.expect(re.escape(prompt), end)
+    check(stage)
+
+    stage = 'quoted multiline paste'
+    quoted = 'a "b" \\ c\nd'
+    label = 'paste-' + nonce
+    line = ("printf 'rel-%%s:<%%s>\\n' %s '%s'" % (label, quoted)).encode()
+    command = b'\x1b[200~' + line + b'\x1b[201~\n'
+    start = len(session.output)
+    session.send(command)
+    expected = re.escape(('rel-%s:<%s>' % (label, quoted)).encode()).replace(re.escape(b'\n'), rb'\r?\n')
+    _, end = session.expect(expected, start, command)
+    session.expect(re.escape(prompt), end)
+    check(stage)
+
+    stage = 'Tab completion spaced filename'
+    filename = 'complete-' + nonce + ' space.txt'
+    expected = ('rel-complete-%s:<file-content>' % nonce).encode()
+    with open(os.path.join(isolation, filename), 'xb') as completed_file:
+        completed_file.write(expected + b'\n')
+    command = ('cat ./complete-' + nonce).encode() + b'\t\n'
+    start = len(session.output)
+    session.send(command)
+    _, end = session.expect(re.escape(expected), start, command)
+    session.expect(re.escape(prompt), end)
+    check(stage)
+
+    stage = 'Ctrl-C multiline pending input'
+    label = 'cancel-' + nonce
+    pending = ("printf 'rel-%%s:<%%s>\\n' %s 'cancel-first\ncancel-second" % label).encode()
+    start = len(session.output)
+    session.send(pending)
+    session.expect(ansi.join(re.escape(bytes([byte])) for byte in b'cancel-second'), start)
+    start = len(session.output)
+    session.send(b'\x03')
+    session.expect(re.escape(prompt), start)
+    session.marker('multiline-recovered', 'yes')
+    if ('rel-%s:<' % label).encode() in session.output[start:]:
+        raise RuntimeError('cancelled multiline input executed')
+    if session.foreground() != session.pid:
+        raise RuntimeError('shell did not retain foreground after multiline interrupt')
+    check(stage)
+
+    stage = 'resize pending wrapped editing'
+    label = 'resize-' + nonce
+    pending = ("printf 'rel-%%s:<%%s>\\n' %s okx" % label).encode()
+    start = len(session.output)
+    session.send(pending)
+    session.expect(ansi.join(re.escape(bytes([byte])) for byte in b'okx'), start)
+    size = fcntl.ioctl(session.master, termios.TIOCGWINSZ, b'\0' * 8)
+    try:
+        fcntl.ioctl(session.master, termios.TIOCSWINSZ, struct.pack('HHHH', 24, 20, 0, 0))
+        command = b'\x08\n'
+        session.send(command)
+        _, end = session.expect(re.escape(('rel-%s:<ok>' % label).encode()), start, pending + command)
+        session.expect(re.escape(prompt), end)
+    finally:
+        fcntl.ioctl(session.master, termios.TIOCSWINSZ, size)
+    check(stage)
+
     stage = 'interactive termios restoration'
     session.finish()
     check(stage)
@@ -343,15 +410,16 @@
     writer = Session(history=True)
     writer.expect(re.escape(prompt), 0)
     writer.marker('history-older', 'older')
-    writer.marker('history-newest', 'newest')
+    writer.marker('history-newest', quoted)
     writer.finish()
     reader = Session(history=True)
     reader.expect(re.escape(prompt), 0)
     if writer.pid == reader.pid:
         raise RuntimeError('history restart did not start a different process')
     start = len(reader.output)
     reader.send(b'\x1b[A\n')
-    _, end = reader.expect(re.escape(('rel-history-newest-%s:<newest>' % nonce).encode()), start, b'\x1b[A\n')
+    expected = re.escape(('rel-history-newest-%s:<%s>' % (nonce, quoted)).encode()).replace(re.escape(b'\n'), rb'\r?\n')
+    _, end = reader.expect(expected, start, b'\x1b[A\n')
     reader.expect(re.escape(prompt), end)
     reader.finish()
     check(stage)
@@ -479,10 +547,10 @@
     for error in cleanup_errors:
         print('cleanup: ' + error, file=sys.stderr)
     sys.exit(1)
-if len(passed) != 6 or len(sessions) != 3 or any(s.status != 0 for s in sessions):
+if len(passed) != 11 or len(sessions) != 3 or any(s.status != 0 for s in sessions):
     print('FAIL: incomplete PTY execution', file=sys.stderr)
     sys.exit(1)
-print('PASS: 6 PTY checks; 3 executable sessions exited zero; owned processes cleaned', flush=True)
+print('PASS: 11 PTY checks; 3 executable sessions exited zero; owned processes cleaned', flush=True)
 PYTHON
 
 my $pid = fork // die "cannot start PTY backend: $!\n";