[leon1418]

Record the Sonnet 5.5 release as a separate model while keeping the existing Sonnet 5 catalog identities, defaults, and price rows consistent. The guidance distinguishes Amazon Bedrock IDs from first-party IDs, calls out the 5.5 thinking/tool/history requirements and Bedrock computer-use exception, and reaches both GCP and Azure through synchronized vendored copies.

Sonnet 5.5 selection remains a separate integration requiring its own model/path, pricing, lifecycle, and compatibility evidence. Regression assertions pin the name and version associated with the currently selected Sonnet 5 IDs.

Validation: 323 Agent Advisor tests passed; the two added identity assertions failed before the fix. Repository manifest, plugin-parity, specification and Markdown checks passed; both shared-guide copies are byte-identical. No live model calls or deployment were performed.

Sources: Sonnet 5.5 model page and migration guide.

[leon1418]

[🤖 AI review 🤖] Reviewed 42399b2 with OCR-assisted repository review and one fresh independent reviewer.

The revision preserves Sonnet 5 names, versions and invocation IDs, documents Sonnet 5.5 as a separate model with the correct Bedrock identifier and migration requirements, and synchronizes both vendored guides. No verified findings remain in the final patch.

Validation: 323 Agent Advisor tests passed; the identity regression checks failed before the fix. Repository Markdown, manifest, plugin-parity and specification checks passed, and both shared-guide copies match byte-for-byte. Live model invocations and deployment were not performed. GitHub CI and required human approval remain separate merge conditions.

[herosjourney]

Reviewed: head 42399b2 (base 8164a8f), 10 files (+27/−3) — agent-advisor, gcp-to-aws, azure-to-aws, skills/shared. One plugin tree.
Ran: byte compare of skills/shared/ai/ai-openai-to-bedrock.md with the GCP and Azure vendored copies (identical). Grep of the Sonnet 5.5 paragraphs and of Azure Clarify. merge-tree against current main: conflicts. merge-tree against #367: conflicts. Fetched the Sonnet 5.5 model card and the migration guide on 11 Oct 2026.
Did not run: the Agent Advisor suite locally. GitHub build passed on this head. No live model calls.

Cross-plugin parity: The shared guide and both vendored ai-openai-to-bedrock.md copies are byte-identical. The GCP Gemini design ref carries the same release paragraph.
Outside skills/: the July catalog disclaimer and two identity assertions.

Change narrative: Sonnet 5.5 shipped as its own model, and the existing Sonnet 5 names, IDs, and $2/$10 rows must stay Sonnet 5. The PR records that, pins the current recommendation identity, and leaves a real 5.5 selection for a later change that has prices, lifecycle, and request handling. Body vs diff: matches. llm-to-bedrock is untouched, which the body states.

The model card lists Bedrock ID anthropic.claude-sonnet-5-5, launch 28 Sep 2026, and still lists tool computer_20251124. The migration guide uses first-party ID claude-sonnet-5-5, says thinking: {"type": "between_tools"} is accepted at low, medium, and high effort, and says tool_choice types any and tool are rejected.

Axis	Stance	One line
Parsimony	clear	A disclaimer, two assertions, and the same release paragraph on the guides
Completeness	finding	The "do not reuse Sonnet 5 rates" line is only in the Azure price cache
Security	clear	No credentials, IAM, or gates. CI scanners passed on this head
Alternatives	advisory	Keep this as the release fence; full selection stays a follow-up
Contradictions	clear	The new sentences agree with the model card and the migration guide
Overlap	clear	The Gemini design ref repeats the shared guide so that path sees the fence
Cross-plugin parity	clear	Shared guide and both vendored copies match
Propagation & data	finding	GCP pricing-cache.md still has the Sonnet 5 $2/$10 row and no 5.5 fence

Blocking

None.

Should-fix

1. [Propagation & data] The rate fence is only in the Azure cache — plugins/aws-startup-advisor/skills/azure-to-aws/references/shared/pricing-cache.md:33

   Sonnet 5.5 (anthropic.claude-sonnet-5-5 on Bedrock) is a distinct model and has no row in this cache yet. Keep the current Sonnet 5 default and rate rows until 5.5 has a separately verified entry.
plugins/aws-startup-advisor/skills/gcp-to-aws/references/shared/pricing-cache.md still has Claude Sonnet 5 | anthropic.claude-sonnet-5 | 2.00 | 10.00 and no 5.5 sentence. GCP estimate-ai.md prices from that cache.
   Consequence: a Google Cloud estimate asked for Sonnet 5.5 can attach the Sonnet 5 $2/$10 row. The Azure cache was just given a sentence that forbids that.
   Fix: add the same paragraph to the GCP cache, next to the Sonnet 5 row. An explicit 5.5 estimate stays pricing_source: "unavailable" until a verified 5.5 row exists.

Nits

None.

Heads-up (out of diff, pre-existing — not blocking)

• merge-tree against current main conflicts in pricing-cache.md, both vendored guides, the shared guide, ai-gemini-to-bedrock.md, and both GCP Clarify files. Keep the 5.5 paragraphs when resolving. merge-tree against #367 also conflicts; land the Opus 5.5 pricing and lifecycle text and this fence together so neither overwrites the other.
• This pull request does not make Sonnet 5.5 selectable, priced, or generated. The body already leaves that to a later change with its own model path, lifecycle, prices, and compatibility evidence.

Alternatives (max 2)

1. Add the same one-paragraph fence to the GCP price cache — wins: both estimate paths refuse to reuse $2/$10; costs: one more conflict hunk with main. Recommendation: switch.
2. Wait and add Sonnet 5.5 as a full catalog row in a follow-up — wins: one change carries ID, price, and lifecycle; costs: until then a GCP estimate can still quote the Sonnet 5 row. Recommendation: do that follow-up after this fence is on both caches.

Verdict

Comment — the guide copies match and the Bedrock ID is right, and the Google Cloud price cache still has no fence against reusing the Sonnet 5 rate. Who can merge: a member of @aws/aws-startup-advisor or @aws/agent-toolkit-admins, after main is merged and the 5.5 paragraphs are kept.

[herosjourney]

Reviewed: head b26d33d (base 769533d), 13 files (+33/−3) — agent-advisor, gcp-to-aws, azure-to-aws, skills/shared. One plugin tree.
Ran: byte compare of skills/shared/ai/bedrock-pricing-cache.md and ai-openai-to-bedrock.md with the agent-advisor, GCP, and Azure vendored copies (identical). Grep for the Sonnet 5.5 fence and for Sonnet 5 rate rows. merge-tree against current main: clean. merge-tree against #367 and #369: conflicts. Fetched the Sonnet 5.5 model card and the migration guide on 19 Oct 2026. GitHub build and startup-advisor passed on this head.
Did not run: the Agent Advisor suite locally. No live model calls.

Cross-plugin parity: The shared Bedrock cache and guide match the agent-advisor, GCP, and Azure vendored copies.
Outside skills/: the July catalog disclaimer and the two identity assertions.

Change narrative: Sonnet 5.5 is its own model, and the Sonnet 5 names, IDs, and $2/$10 rows stay Sonnet 5. The PR records that fence and leaves a real 5.5 selection for a later change. Body vs diff: matches.

The model card, read today, lists Bedrock ID anthropic.claude-sonnet-5-5, launch 28 Sep 2026, tool computer_20251124, and tool_choice types auto and none only. The migration guide says thinking: {"type": "between_tools"} is accepted at low, medium, and high effort, rejects tool_choice types any and tool, and limits the computer_20251124 removal to the Claude API and Google Cloud.

Axis	Stance	One line
Parsimony	clear	One fence paragraph, copied through the existing vendored Bedrock cache
Completeness	clear	GCP and Azure estimate paths both read that cache
Security	clear	No credentials, IAM, or gates. CI scanners passed on this head
Alternatives	clear	None worth raising
Contradictions	clear	The new sentences agree with the model card and the migration guide
Overlap	clear	The Gemini design ref repeats the shared guide so that path sees the fence
Cross-plugin parity	clear	Shared cache and guide match every vendored copy
Propagation & data	clear	The fence is on the shared Bedrock cache and all three vendored copies

Blocking

None.

Should-fix

None.

Nits

None.

Heads-up (out of diff, pre-existing — not blocking)

merge-tree against current main is clean. merge-tree against #367 still conflicts in clarify-ai.md and clarify-ai-only.md; the Bedrock cache auto-merges. merge-tree against #369 conflicts in the shared guide, the Bedrock cache, and both GCP Clarify files. Keep the 5.5 paragraphs when resolving either.

This pull request does not make Sonnet 5.5 selectable, priced, or generated. The body already leaves that to a later change.

Alternatives (max 2)

None worth raising.

Verdict

Approve — the Google Cloud Bedrock cache now has the same Sonnet 5.5 rate fence as Azure, and the Sonnet 5 $2/$10 row is unchanged. Who can merge: a member of @aws/aws-startup-advisor or @aws/agent-toolkit-admins.

Re-review at b26d33d (previous: 42399b2). Delta: 2 commits (4ff506f, b26d33d).

Prior finding	Status at head
Rate fence only in the Azure cache	Fixed at 4ff506f, still present after the main merge — the paragraph is in skills/shared/ai/bedrock-pricing-cache.md:41 and the GCP, Azure, and agent-advisor vendored copies are byte-identical. estimate-ai.md on both providers prices Bedrock from that file.
Conflicts with main	Superseded — merge-tree against current main is clean

New in the delta: none. Sweep for a Sonnet 5.5 row or a reused $2/$10 Bedrock ID: the only Sonnet 5 Bedrock row is still anthropic.claude-sonnet-5 at $2/$10, and it sits under the fence.

Verdict: Approve — supersedes the 12 Oct comment.

[herosjourney]

Reviewed #365 against the current model-catalog conventions on main (and the in-flight #449/#450/#469/#439 set). The change is sound and conservative — it keeps Sonnet 5 as the pool default and refuses to promote 5.5 without its own model/path/pricing/lifecycle evidence, which is exactly the discipline the catalog encodes. Two conformance notes before merge, both stemming from a stale base rather than a defect in the change itself.

What conforms (verified at head eb6ccc02):

• Vendored parity is correct: the Sonnet-5.5 block added to bedrock-pricing-cache.md is byte-identical across the canonical skills/shared/ai/ copy and all three vendored copies (agent-advisor, azure-to-aws, gcp-to-aws), as is the ai-openai-to-bedrock.md block across canonical + both vendored copies. lint:vendored-parity and lint:skill-refs pass.
• Registry hygiene holds: the anthropic-bedrock-2026-07-21.json edit adds disclaimer guidance without introducing a new model ID or relabeling Sonnet 5, and model-id-lint + model-staleness --strict both pass. The restricted-model / no-silent-rate-reuse rules are respected.

1. Stale base — rebase onto current main. At this head, lint:vendored-parity reports 51 vendored files and lint:skill-refs reports 1316 references — the pre-#450 numbers. main now carries the target-only AI capability catalog (#450, 55 files / 1357 refs). Please rebase so the change lands on top of the current catalog surface before merge.

2. Shared vs adapter-local Gemini guide will diverge post-#469. This PR adds its Sonnet-5.5 note to the adapter-local skills/gcp-to-aws/references/design-refs/ai-gemini-to-bedrock.md, in the header region (above ## Competitive Reality). #469 introduces a shared skills/shared/ai/ai-gemini-to-bedrock.md (vendored into gcp/azure) plus a body-parity drift gate. Because this note sits in the header — which the gate deliberately excludes — it would not fail the gate, but the shared Gemini guide would silently lack the Sonnet-5.5 note the adapter-local one gains. If #469 merges first, please also add the same note to the shared ai-gemini-to-bedrock.md and re-sync its vendored copies so the two guides stay consistent. (If this PR merges first, #469 will reconcile it instead.)

Neither is blocking the substance. With the rebase and the shared-Gemini sync, it fully conforms to the current constructs.

_{Conformance check run locally against head eb6ccc02; vendored-parity, skill-refs, model-id, and model-staleness gates all pass.}

[herosjourney]

Reconciling my three reviews on this PR so the record is clear:

• Oct 12 (head 42399b2, Comment): raised one Should-fix — the Sonnet-5.5 rate fence was only in the Azure cache, so a GCP estimate could reuse the Sonnet 5 $2/$10 row.
• Oct 19 (head b26d33d, Approve — since auto-dismissed): confirmed that Should-fix was fixed (fence now on the shared bedrock-pricing-cache.md, byte-identical across all vendored copies) and approved. GitHub dismissed that approval automatically when the next commit landed; b26d33d is an ancestor of the current head eb6ccc02, so nothing was rewritten and the fence is still intact at the current head.
• Oct 20 (head eb6ccc02, Comment — above): the two remaining conformance notes (stale base vs current main; shared-vs-adapter-local Gemini guide divergence post-feat(aws-startup-advisor): promote google and openrouter to supported with shared mapping guides #469).

One clarification to avoid confusion between the Oct 19 and Oct 20 notes: "merge-tree against main is clean" (Oct 19) and "base is stale / pre-#450" (Oct 20) are both true — a clean three-way merge is not the same as having the newer catalog surface in the branch. The rebase ask in note 1 is about picking up #450's capability catalog, not about resolving a merge conflict.

Net: the earlier Should-fix stays resolved; the only open items are the two non-blocking conformance notes above.