Milestone: v0.5 beta — contracts locked, release workflow proven

[tig]

Tracks the third milestone in specs/clet-spec.md §7. This is the lock-in milestone: anything that ships in v1.0 is determined here, and contracts published past this point cannot break without a schemaVersion bump per §4.3.1.

Exit criteria — locks

• Naming locked. "clet" is the name. Aliases for the 14 input clets and md are final.
• JSON schema locked (§4.3, §4.3.1, §4.3.2). schemaVersion: 1 is the contract for all of clet 1.x. No type field on envelopes. Cancel envelope is {"schemaVersion":1,"status":"cancelled"}. Per-clet value shapes per the §4.3.2 table.
• Exit-code table locked (§4.8).
• v1.0 input list (14) and viewer list (1: md) locked. Notable absence: password. Additional viewers (json, log, diff) explicitly deferred to v2.

Exit criteria — md viewer

• md viewer clet implemented per §4.5 viewer pattern; uses TG's Markdown View end-to-end.
• Link safety verified: default SurfaceOnly policy per Appendix A; --allow-link-open opts in.
• clet --help and clet help render via the same Markdown View pipeline (§4.7), replacing the v0.11 plain-text help.

Exit criteria — distribution

• dotnet tool install -g Terminal.Gui.clet packs and installs locally (mdv pattern, see D-019).
• Continuous-release loop wired up and proven on the develop channel (D-020). Real tg-develop-published dispatch from TG fires the workflow; Terminal.Gui.clet 2.0.2-develop.35 shipped to nuget.org autonomously; dotnet tool install -g Terminal.Gui.clet --prerelease resolves it from the public feed.

Exit criteria — verification

• Four-terminal manual matrix passes per §6.5 (macOS Terminal, iTerm2, Windows Terminal, GNOME Terminal × clet text, clet pick-file, clet md, theme switch, mouse, inline restore). Captured in release-checklist issue #23.
• Inline rendering confirmed inline (cursor restore, no scrollback dump) on all four.

Exit criteria — pipeline + threat model

• Threat model published per Appendix A: untrusted-input boundaries, terminal-escape sanitization, --title sanitization, Markdown link policy, file-access scope, plugin-loading exclusion. (docs/threat-model.md)
• #5 Open Question: clet md content source (file arg, stdin, or both) decided and shipped. (D-015)
• Upstream PR/FAQ at Terminal.Gui#5155 updated to match the spec (no longer references Terminal.Gui.Clets as a separate assembly).

Moved to v0.9 RC (#5)

The following criteria were originally tracked here but are gated on a real TG release tag and don't block v0.5 lock-in. They live in #5 now:

• Release workflow proven against a real TG release cut. Develop-channel half is done (Terminal.Gui.clet 2.0.2-develop.35 shipped end-to-end via tg-develop-published); release-tag half is captured under "One real release cycle end-to-end" in #5.
• Homebrew tap + WinGet manifest in working draft form. Currently placeholder echoes per D-012 (build-from-source acceptable for first release). Tracked under "Operational" / "Homebrew formula + WinGet manifest in working-draft form" in #5.

Terminal.Gui dependency

The original criterion read "By v0.5 the Terminal.Gui dependency must be a release tag, not a *-develop.* preview NuGet." That has been superseded by D-020 — clet now uses a <TerminalGuiVersion> MSBuild variable in src/Clet/Clet.csproj, set per-build from the dispatch payload. There is no longer a "pin to remove" — clet builds against whatever TG version the dispatch names. The §8 risk row tracking the pin is resolved in the spec.

• #5157 and #5158 shipped on TG develop. (Release-tag promotion no longer required to close v0.5; the develop pin is gone per D-020.)
• #5156 (Markdown View golden-file rendering tests) — resolved.
• src/Clet/Clet.csproj Terminal.Gui PackageReference updated from *-develop.* to a TG 2.x.y release version — superseded by D-020. csproj uses <PackageReference Include="Terminal.Gui" Version="$(TerminalGuiVersion)" />; the workflow overrides per dispatch.

Kill criteria

Per the press release: we walk away from v1.0 if at v0.9 any of the following is unresolved:

• Inline rendering doesn't actually feel inline across the four target terminals.
• AI-agent JSON contract churn (consumers stop trusting it).
• Native-installer pipeline that breaks on every release.
• The Markdown View losing ground to glow on common Markdown variants.

v0.5 is where these become measurable, not vibes.