feat(config): add Claude 5 family (Fable 5, Sonnet 5) to the Model enum

[arcaputo3]

Summary

• Adds Model.Fable5 (claude-fable-5) and Model.Sonnet5 (claude-sonnet-5) as first-class enum cases with fromId mappings.
• New Model.fable convenience alias.
• ⚠️ Repoints Model.sonnet and Model.default from Sonnet4_6 to Sonnet5 — the aliases are documented as "point to current generation", but default affects any caller that never sets a model. Flagging for explicit review; easy to revert to Sonnet4_6 if you want the default pinned.
• SupportsEffort: adds XHigh + Max compile-time evidence for both new models (Sonnet 5 is the first Sonnet tier with xhigh). Negative cases (e.g. Sonnet4_6 + XHigh) unchanged.

Motivation

TJC-1550 moves the tjc-agents fleet default to Claude Fable 5. tjc-agents currently carries a Models.fable = Model.fromId("claude-fable-5") shim (lands as Custom); once this ships in a release, the shim collapses to Model.fable and the pin bumps.

Validation

• ModelSpec: 32/32 on JVM and JS (id mapping, fromId round-trip, JSON codec round-trip, alias assertions).
• AgentOptionsSpec (JS): 105/105 — typeChecks positives for Fable5/Sonnet5 × XHigh/Max, negatives still rejected.
• __.compile 389/389; scalafmt applied.
• Live check: claude-sonnet-5 responds 200 with the TJC org key. claude-fable-5 is currently gated org-side (400 model_not_available — data retention must be enabled ≥30d), which is a Console setting, not an SDK concern.

🤖 Generated with Claude Code

[claude]

Review — feat(config): add Claude 5 family (Fable 5, Sonnet 5)

Clean, focused change. The additions to the Model enum, fromId mappings, SupportsEffort evidence, and matching test coverage all line up. The self-flagged default repoint in the description is exactly the right thing to surface. A few notes below, mostly minor.

🟡 Stale docs in Effort.scala (should be updated in this PR)

This PR adds XHigh + Max evidence for Fable5 and Sonnet5, but the Effort enum docstrings still only mention the Opus/Sonnet-4.6 tiers:

• Effort.scala:21 — /** Deeper than high. Supported by Opus 4.8 and Opus 4.7. */ → should now also list Fable 5 and Sonnet 5.
• Effort.scala:24 — /** Maximum effort. Supported by Opus 4.8, Opus 4.7, Opus 4.6, and Sonnet 4.6. */ → same.

These are the source of truth developers read for which models accept xhigh/max, so it's worth keeping them in sync with the new SupportsEffort givens.

🟡 Behavioral change: Model.default / Model.sonnet → Sonnet5

You already called this out (👍), just documenting the blast radius for reviewers: every caller that never sets a model, plus the ~18 examples using .withModel(Model.sonnet) (A2A, Session, Subagent, Hook, Permission, MacroTool, etc.), now resolve to claude-sonnet-5 instead of claude-sonnet-4-6. Given the live check shows claude-sonnet-5 responds 200 with the org key, this seems safe — but note claude-fable-5 is still org-gated (400 model_not_available), so leaving default = Sonnet5 (rather than Fable5) is the correct conservative choice here, even though the TJC-1550 fleet default moves to Fable 5. Worth a one-liner in the PR/commit that scalagent's default intentionally stays on Sonnet 5 until Fable 5 is generally available.

🟢 Minor: Model.opus under the "current generation" comment

fable/sonnet now point at the 5 family while opus remains Opus4_8. That's correct (there's no Opus 5), but the shared // Convenience aliases (point to current generation) comment reads a touch inconsistently. Optional: a short note that opus tracks the 4.8 tier as the current Opus.

✅ Strengths

• fromId round-trip, JSON codec round-trip, and alias assertions all extended for both new models — the default is Sonnet5 test was correctly updated rather than left asserting the old value.
• SupportsEffort positives added in AgentOptionsSpec typeChecks and negatives left intact — good that the compile-time-evidence contract is exercised for the new tiers.
• No behavioral surprises in fromId: unknown IDs still fall through to Custom.

Nothing blocking. Recommend folding the Effort.scala docstring update into this PR since it's the same conceptual change; the rest are optional polish.

[arcaputo3]

Review follow-ups addressed in d-latest:

• 🟡 Effort.scala docstrings — XHigh and Max now list Fable 5 + Sonnet 5 alongside the 4.x tiers, in sync with the SupportsEffort givens.
• 🟡 Default rationale — documented directly on Model.default: it intentionally stays Sonnet 5 because Fable 5 requires org-level data retention (still gated for us); Fable is opt-in via Model.fable. Durable in code rather than only in the PR body.
• 🟢 Alias comment — now reads "track each tier's current generation" with an explicit note that opus stays on the 4.8 family since there is no Opus-tier Claude 5.

🤖 Generated with Claude Code

[claude]

Review — Claude 5 family (Fable 5, Sonnet 5) 🤖

Reviewed the diff plus surrounding code. Tight, well-scoped change with solid internal consistency. Approving in spirit; one behavior change deserves an explicit owner sign-off.

What's good

• Consistency across the three source files. The SupportsEffort givens line up exactly with the updated Effort docstrings — XHigh: Fable5 / Sonnet5 / Opus4_8 / Opus4_7; Max: those plus Opus4_6 / Sonnet4_6. Good that the doc was updated in the follow-up commit rather than left to drift.
• fromId round-trips. Both new IDs are mapped and the round-trip test list was extended so the property holds.
• Model IDs match the published identifiers (claude-fable-5, claude-sonnet-5).
• Good judgment on the default. Keeping Model.default = Sonnet5 (not Fable5) because Fable is org-gated on data retention is the right call, and documenting why directly on the val is exactly where a future reader will look.
• No exhaustive match on Model exists elsewhere in src/, so the new enum cases can't silently break a downstream case. fromId correctly still routes unknowns to Custom.

Worth a second look

• The default/sonnet repoint Sonnet4_6 → Sonnet5 is a runtime behavior change, not just an additive enum bump. Every caller that never sets a model — and every example in examples/ using Model.sonnet (18 call sites: A2AExample, MacroToolExample, PermissionExample, etc.) — now resolves to claude-sonnet-5 instead of claude-sonnet-4-6. Consistent with the documented "aliases track the current generation" contract, and you flagged it in the PR body, so this is just asking for an explicit sign-off from whoever owns the default rather than a change request. The pin is trivially revertable if Sonnet 5 GA/pricing status isn't settled for all consumers.

Minor / optional

• Test redundancy in ModelSpec. The new "Model enum has correct IDs for Claude 5 family" + "fromId parses known Claude 5 models" tests already assert exactly what the later "Fable 5 maps to claude-fable-5" / "Sonnet 5 maps to claude-sonnet-5" tests assert. Harmless, and it mirrors the existing per-model style for Opus 4.8/4.7, so keeping it for symmetry is defensible — just noting the duplication.
• No new negative type-check case. Existing negatives (e.g. Sonnet4_6 + XHigh) are unchanged, which is fine. A typeChecks-should-fail assertion for something like Model.Haiku4_5 + Effort.XHigh would lock in that the new givens did not accidentally widen the surface — but the givens are per-.type, so this is genuinely low risk.

Nothing blocking. Validation numbers in the description (ModelSpec 32/32, AgentOptionsSpec 105/105, compile 389/389) cover the surface area well.