fix(alquimia): render /speckit-<name> invocations for the skills-only Alquimia agent

[jawwad-ali]

Problem

alquimia is the only one of the 19 SkillsIntegration subclasses absent from every set in _invocation_style.py (ALWAYS_SLASH_AGENTS / CONDITIONAL_SLASH_AGENTS / DOLLAR_SKILLS_AGENTS / SKILL_COLON_AGENTS).

I checked that programmatically rather than by eye:

SkillsIntegration subclasses NOT in any invocation set: ['alquimia']
total in sets: 18

It installs .alquimia/skills/speckit-<name>/SKILL.md, and its own inherited build_command_invocation() already returns /speckit-plan. But is_slash_skills_agent("alquimia", True) returned False, so the two callers that consult the helper fell through to the dotted form Alquimia never registers:

• HookExecutor._render_hook_invocation (extensions/__init__.py) → /speckit.plan
• specify init's Next Steps panel (commands/init.py) → /speckit.plan

So the integration's own renderer and the shared helper disagreed for the same on-disk layout.

Reproduction on current main (bf88c9f)

Real specify init --here --integration <agent> --script sh --ignore-agent-tools:

BEFORE
  alquimia -> 2.1 /speckit.constitution      <-- wrong
  droid    -> 2.1 /speckit-constitution

AFTER
  alquimia -> 2.1 /speckit-constitution
  droid    -> 2.1 /speckit-constitution

Both agents produce an identical .../skills/speckit-*/SKILL.md layout, so there is no basis for the difference.

Which set — a judgment call, stated plainly

I put it in CONDITIONAL_SLASH_AGENTS, not ALWAYS_SLASH_AGENTS, and I want to be explicit that this is a choice rather than something the code forces:

• Every ALWAYS_SLASH member has commands_subdir: "skills" — but so does claude, which is CONDITIONAL, so that field is not the discriminator.
• Neither alquimia nor droid nor claude declares an options() mode toggle, so that isn't either.
• specify init writes ai_skills: true for alquimia (verified), so CONDITIONAL fixes the real path.

CONDITIONAL is the conservative option: it renders the hyphenated form only when skills are actually enabled, so it stays correct if alquimia ever gains a commands mode. If you'd rather it be ALWAYS alongside the other skills-only agents, that's a one-word change and I'm happy to make it.

Note on scope

The third consumer, _resolve_command_ref_tokens, already falls back to integration.build_command_invocation() and was therefore always correct — this changes nothing there.

Verification

• Fail-before / pass-after: 2 new-vs-baseline failures with the source reverted → 38 passed with the fix.
• After the change, no SkillsIntegration subclass is unmapped: SkillsIntegration subclasses still missing: NONE.
• Scoped regression over tests/integrations: no new failures vs a clean-main baseline captured on bf88c9f9 (18 pre-existing in scope, all Windows symlink-privilege).
• uvx ruff@0.15.0 check src tests → clean

Tests mirror test_integration_droid.py::TestDroidIntegration::test_is_slash_skills_agent, including the disabled case that distinguishes conditional from always.

---

Written with assistance from Claude Code. Bug found, reproduced, and verified by me on current main.