[bpamiri]

Pins the Wheels CLI runtime to the published wheels-dev/LuCLI v0.6.2.3 (release, tag 350ded024).

Changes

• tools/lucli.json → 0.6.2.3, with the published release's sha256s (they match its checksums_sha256.txt):
  • jar d554249d1d1af2fa8068f3ca945c4261b9741887410e1c9ffc62a7230f70b20f
  • linux/macos 418eebf3843caacacc3a405d51e32a157db8213c1ed1b87ef5b24ebf0148c6f4
• tools/lucli-tested.json:
  • adds the 4.2 line entry ["0.6.2.3"];
  • adds a _notes."4.1" entry: 4.1 stays at max 0.6.2.2. LuCLI 0.6.2.3 passes a runtime-owned __lucliMcpCall argument to every MCP tools/call. 4.1.x has no handling for it (fix(cli): refuse generate type=app when called over MCP, not only drop it from the schema #3980 is 4.2-only), so on 0.6.2.3 every 4.1.x MCP tool call fails with Unknown argument '--__lucliMcpCall'.
• changelog.d/lucli-0-6-2-3.changed.md: the user-visible changes.

What 0.6.2.3 brings (fork PRs #19, #21-#24, #26-#28)

• --env / --envfile reach wheels <command>.
• wheels start --dry-run writes nothing and previews --name.
• A root option before a root subcommand works (wheels --timing cfml …).
• REPL history and settings stay in the CLI's home.
• run <file>.cfm no longer prints the setup.
• Hints say wheels.
• The MCP marker that lets develop refuse generate type=app over MCP.

Tested

All runs used the published jar against develop, with an isolated LuCLI home and user.home, and no writes to the real ~/.wheels, ~/.lucli or ~/.lucee:

Suite	0.6.2.2 (current pin)	0.6.2.3
core, Lucee 7 + SQLite (tools/test-local.sh)	7217/7217	7217/7217
CLI suite (tools/test-cli-local.sh)	2437 pass, 0 fail	2437 pass, 0 fail
onboarding harness	36/12/2	36/12/2: the same 12 fresh-home failures (sqlite-jdbc placed after the app server's first boot; a harness gap, tracked separately)
stdio MCP probe	generate type=app scaffolds over MCP	generate type=app refused over MCP (intended)

The fork's install-validation passed on the published assets on Linux, macOS and Windows.

The Homebrew / Scoop / apt / yum formulas follow separately via release-manager, and 4.1.x formulas stay on 0.6.2.2.

🤖 Generated with Claude Code

[bpamiri]

Merge ordering (release-manager, 2026-10-08): homebrew-wheels' stable auto-update currently reads develop's tools/lucli.json, so merging this as-is would make the bot propose LuCLI 0.6.2.3 for the stable (4.1.x) formula. 4.1.x must stay at max 0.6.2.2 (see _notes.4.1 in tools/lucli-tested.json).

• homebrew-wheels' stable bot uses tools/lucli-tested.json per release line (release-manager) — merge this PR only after that lands

🤖 Generated with Claude Code

[bpamiri]

rev1-r3 review @ 92249ed: APPROVE (merge held until homebrew-wheels#1100 lands; see the last line)

The pinned SHA-256s are verified two ways:

• They match the v0.6.2.3 release's checksums_sha256.txt: lucli-0.6.2.3-{linux,macos} = 418eebf3843c…, lucli-0.6.2.3.jar = d554249d1d1a….
• I downloaded lucli-0.6.2.3-linux, lucli-0.6.2.3-macos and lucli-0.6.2.3.jar and hashed them myself, with the same results.

Provenance:

• The v0.6.2.3 tag points at 350ded0, the merge commit of build: prepare fork release 0.6.2.3 LuCLI#29 (the 0.6.2.3 release prep I reviewed).
• The release is not a draft or prerelease, and it is releases/latest.
• The jar's lucli/version.properties reads lucli.version=0.6.2.3 and lucli.build.commit=350ded0.

tools/lucli-tested.json:

• The new "4.2": ["0.6.2.3"] and the _notes key don't change how the install smoke reads the file. Its jq reads only .lines[$line].
• I simulated the resolve and check step with this file:
  • 4.1.2 accepts 0.6.2.1 and 0.6.2.2, and rejects 0.6.2.3;
  • 4.2.0 accepts 0.6.2.3 and rejects 0.6.2.2;
  • an unlisted line accepts its tag pin only.
• The 4.1 note (max 0.6.2.2, because fix(cli): refuse generate type=app when called over MCP, not only drop it from the schema #3980's __lucliMcpCall handling is 4.2-only) is accurate: v4.1.2 has no handling under cli/lucli.

CI on this head: 17 pass, 2 skipping, 0 fail. The CLI jobs set up the runtime from this pin; the log shows LuCLI Version: 0.6.2.3 (sha256 418eebf3843c…). The full CLI and core suites against the published bytes were run outside CI, per the PR (CLI 2437/0/0, core 7217/7217). MERGEABLE.

INFO: the v4-2-0 installation guide's sample wheels version output still shows Wheels 4.0.3 / LuCLI 0.6.2.2. It is labelled "something like" and was already stale before this PR, so it can be refreshed with the 4.2 docs.

Merge order: this must not merge before wheels-dev/homebrew-wheels#1100 (approved at 18b38a917). Without it, the stable formula's auto-update would take develop's new 0.6.2.3 pin and move the 4.1.2 formula onto a runtime 4.1.x can't run MCP calls on.

[github-actions]

Wheels Test Results

36 925 tests   36 190 ✅  12m 55s ⏱️
 2 960 suites     735 💤
     5 files         0 ❌

Results for commit 92249ed.