Interactive Tab completion never suggests option names (--flags)

[autocarl]

Description

The interactive REPL's Tab completion does not suggest option names (--xxx), even though docs/interactive-loop.md documents that it should:

As the user types, the autocompletion system suggests available commands, contexts, and option names visible from the current scope.

Repro steps

In an interactive session:

> [nothing typed] + Tab
Matches: catalog, list, install, update, uninstall (+5)

> i + Tab
Command: install

> install + Tab
Param skillName: The skill to install

> install msaf-architect -- + Tab
[nothing useful is suggested]

Command-name completion and positional-parameter completion both work correctly. Only option/flag completion (--) produces no usable suggestion.

Expected behavior

Tab should list the options available on the resolved route (for example --help and route-specific options), the same way the external shell completion bridge (completion __complete, used for bash/zsh/fish/PowerShell/Nu) already does.

Observed behavior

No option names are suggested and no live hint is shown for -- or partial option prefixes.

A local probe on current origin/main (de6de5b) produced only a non-selectable invalid placeholder for the interactive path:

interactive candidates for: install msaf-architect --
--|Invalid|selectable=False
interactive live hint: <null>

Shell completion validation

The external shell completion path does handle option candidates.

Existing regression tests pass:

dotnet test src/Repl.IntegrationTests/Repl.IntegrationTests.csproj -c Release --filter 'FullyQualifiedName~Given_Completions'
# total: 6, failed: 0, succeeded: 6, skipped: 0

A local route-option probe also returned route and global option candidates from completion __complete:

shell candidates for: repl install msaf-architect --
--force
--help
--human
--interactive
--json
--markdown
--no-interactive
--no-logo
--output:
--output:human
--output:json
--output:markdown
--output:xml
--output:yaml
--xml
--yaml
--yml

So this appears scoped to the interactive autocomplete engine, not the shell bridge.

Technical diagnosis

The repo currently has two separate completion engines:

• src/Repl.Core/ShellCompletion/ShellCompletionEngine.cs drives the external shell bridge described in docs/shell-completion.md. It implements option suggestions:
  • ResolveShellCompletionCandidates detects option tokens and calls AddShellOptionCandidates.
  • AddGlobalShellOptionCandidates and AddRouteShellOptionCandidates use static globals, output aliases/formats, custom globals, and route.OptionSchema.KnownTokens.
  • Existing coverage includes src/Repl.IntegrationTests/Given_Completions.cs for custom global option completion.
• src/Repl.Core/Autocomplete/AutocompleteEngine.cs drives the interactive loop (CoreReplApp.Autocomplete.cs, Session/InteractiveSession.cs, Console/ConsoleLineReader.Autocomplete.cs). Its CollectAutocompleteSuggestionsAsync currently assembles command, context, dynamic, ambient, and ambient-continuation candidates, but has no equivalent branch that generates ---prefixed option candidates.

AutocompleteEngine.IsGlobalOptionToken exists, but today it is used for token classification/live-hint behavior, not for collecting selectable option suggestions.

There also appears to be no existing interactive autocomplete regression for option/flag candidates: Given_ConsoleLineReader_Autocomplete.cs and Given_InteractiveAutocomplete_LiveHint.cs contain no -- / option-name coverage.

Suggested direction

Share or port the option-candidate logic from ShellCompletionEngine into AutocompleteEngine, so interactive Tab completion uses the same sources of truth:

• static global options (--help, --interactive, etc.);
• output aliases and output formats;
• custom global options;
• route-specific options from route.OptionSchema.KnownTokens.

Add regression tests for at least:

1. install msaf-architect -- suggests route/global options interactively;
2. partial option prefixes (for example --fo) filter correctly;
3. shell completion remains green, to keep parity between both completion paths.

Context

Reported by an external user following the shell-completion documentation who expected -- completion to work the same way inside the interactive REPL.