Skip to content

Public Readiness

Positioning

nshell targets interactive use. The 0.6 series is a development preview, not a POSIX/fish-compatible script interpreter.

Historical release evidence (v0.6.4)

The published v0.6.4 release contains the x86_64-linux bundle and SHA-256 checksum. The main CI run and v0.6.4 release run completed with success for commit 5969f455312025356905a0f8e2e104da317a0d6a. These workflow results do not establish production readiness, complete dependency assertion coverage, or successful live AI provider requests.

Local verification for v0.6.5

The macOS arm64 source-loaded suite passed with PTY cases enabled and no skipped cases. The pinned dependency suites and private history storage regressions also passed. The dependency gate rejects zero-assertion runs of three selected history properties and tests the pinned process library's native spawn helper separately from its Lisp suite.

A locally built executable was tested in a real PTY with Claude Code 2.1.295. The probe checked that a proposed touch command did not run before approval, Enter executed it once, cancellation left no file, a subsequent request could execute a fresh proposal, and exit reaped the provider processes. These are bounded checks with one provider on one platform, not an exhaustive provider or terminal compatibility claim. They are not part of the linked CI runs.

Historical capability matrix (v0.6.2)

The evidence and statuses below describe the published v0.6.2 tree. They are historical evidence, not results from the v0.6.4 workflows above.

Area Capability or policy Published evidence Status
Interactive editor Live syntax feedback, Emacs bindings, optional vi mode, multiline editing, undo/yank, predictable rendering input-state, rendering, and PTY integration tests CI-verified on x86_64-linux
History and suggestions Persistent history, reverse search, prefix autosuggestions, handling of multiline entries history, autosuggest, and E2E editing tests CI-verified on x86_64-linux
Completion Context-aware command/path/flag completion, candidate menu, deterministic cycling and cancellation completion domain and REPL rendering tests, hierarchical command resolution, curated external command metadata and selective help-text enrichment Broader command discovery/cache policy and subcommand coverage remain future work
Shell language Functions, control flow, command substitution, expansions, heredocs, here-strings, redirection, pipelines parser, expansion, source, pipeline, process-substitution, descriptor-duplication, tab-stripping-heredoc, and smoke tests Tested subset; full POSIX/fish parity is not claimed
Process control Foreground/background jobs, jobs/fg/bg/disown, Ctrl-C recovery, terminal restoration; foreground external commands whose standard output is a terminal have no default timeout, while file/pipe output defaults to 3600 seconds and command substitution to 300 seconds job-control and non-sandboxed PTY integration tests CI-verified on x86_64-linux
Distribution Nix-pinned build, installed man page, release binary smoke, checksummed artifacts flake build, man page, CI/release workflows, v0.6.2 release assets Prebuilt x86_64-linux binary and SHA-256 checksum published; nixpkgs/Homebrew remain future work
Security and operations Private vulnerability reporting and explicit security scope repository security policy and contribution guidelines Reporting policy exists; it does not establish a security audit of runtime behavior

The v0.6.2 CI run and release workflow validate the tagged tree. The non-sandboxed integration job runs the PTY and external-process cases skipped by the Nix sandbox.

Release Gates

These are the existing nshell gates, not a production-readiness verdict. The historical v0.6.2 release run includes dependency-suite and extracted-artifact checks, including the extracted executable's real-PTY gate.

Before a public release can claim the capabilities listed here:

  1. nix flake check --print-build-logs passes on x86_64-linux CI, the only declared hermetic check target.
  2. The non-sandboxed integration suite passes for PTY, subprocess, terminal, signal, and job-control coverage on the sole CI matrix target, x86_64-linux, without skipping those cases.
  3. The pinned dependency suites execute directly, with nonempty discovery, matching selected/executed counts, and no skipped cases. Suite-level pass counts alone do not establish assertion coverage. The private history storage regressions pass.
  4. An x86_64-linux release archive passes checksum verification before extraction. The extracted bundle passes required-file, dependency-closure, license-file, and runtime smoke checks. Its executable passes the real-PTY gate for editing, signals, job control, history restart, and terminal restoration, and ships with README.md, LICENSE, dependency licenses, and nshell(1).
  5. User-visible behavior changes are represented in README, man page, the GitHub Release notes, and completion metadata when applicable.
  6. Open roadmap gaps remain explicit instead of being implied as complete.

The v0.6.2 release workflow runs hermetic checks and a non-sandboxed integration job against the same resolved tagged commit, and only then builds the release assets in a draft. A maintainer reviews the assets and release notes before publishing the draft.

Verification outside CI

Run these checks from the checkout being verified. To reproduce the historical release evidence above, use the v0.6.2 tag. The flake declares only x86_64-linux. On another host, the Nix checks require an x86_64-linux builder. Running these Nix-based commands also requires an x86_64-linux execution environment; a remote builder alone does not supply that runtime coverage.

  • Run nix flake check --print-build-logs on an x86_64-linux builder.
  • Run the non-sandboxed suites on the same target with a temporary home and XDG directories, as the integration jobs do. Tests can access history and configuration through the normal runtime APIs; do not use personal data. Run this block from the repository root:
(
  verification_dir=$(mktemp -d)
  mkdir -p "$verification_dir/home" "$verification_dir/cache" \
    "$verification_dir/config" "$verification_dir/state"
  export HOME="$verification_dir/home"
  export XDG_CACHE_HOME="$verification_dir/cache"
  export XDG_CONFIG_HOME="$verification_dir/config"
  export XDG_STATE_HOME="$verification_dir/state"
  export NSHELL_AI_COMMAND=/nonexistent NSHELL_TEST_PTY=1
  nix develop -c sbcl --script run-tests.lisp &&
    nix develop -c bash scripts/test-dependencies.sh &&
    nix develop -c sbcl --script scripts/test-history-storage.lisp
)

Confirm nonempty test discovery and no skipped PTY or external-process cases in the test report; an exit status of zero alone does not establish coverage. - The integrated suite covers OSC 52 clipboard output, tab-stripping <<- heredocs, process substitution, ordered descriptor duplication, path-like argument completion, hierarchical command completion, and SGR mouse selection. - The dependency runner executes the pinned libraries' own suites; the history script checks private persistence. A successful library load is not a test result. - Verify the release archive against its SHA-256 checksum before extraction. Extract it and run perl scripts/verify-release-bundle.pl <extracted-bundle> followed by perl scripts/test-release-pty.pl <extracted-bundle> on x86_64-linux.

Known Limits and Future Work

  • The local v0.6.5 probe above covers explicit approval, cancellation, and subsequent requests with one provider; other providers and failure modes remain unverified.
  • Broader help-text-driven command discovery remains future work, especially for subcommand coverage and non-curated external tools.
  • Mouse selection and clipboard behavior depend on the terminal. No exhaustive terminal matrix is claimed; the editor maps SGR coordinates through captured prompt geometry and falls back to OSC 52 when host clipboard integration is unavailable.
  • The extracted executable's PTY scenarios are a separate bounded gate, not the entire source-loaded integration suite. Passing them does not establish every integration case against the portable executable.
  • Release assets include a SHA-256 checksum. GitHub artifact attestations are not configured; the checksum alone is not a provenance attestation.
  • Validate the process-isolated benchmark scenarios in CI and collect equivalent fixtures beyond the implemented minimal noninteractive literal-print case.
  • Collect privileged cold-cache, interactive, completion, and end-to-end tail-latency evidence before making any broad performance claim. See Performance evidence.