Skip to content

fix(kiro): offer reasoning effort for Claude Haiku 5.5 - #30935

Open
juliusmarminge wants to merge 3 commits into
mainfrom
fix/issue-30920
Open

juliusmarminge wants to merge 3 commits into
mainfrom
fix/issue-30920

Conversation

@juliusmarminge

@juliusmarminge juliusmarminge commented Nov 4, 2026 •

Copy link
Copy Markdown
Member

Problem

The Kiro provider shows no Reasoning option for Claude Haiku 5.5. kiro-cli 2.37.1 advertises effortLevel for claude-haiku-5.5 (low, medium, high, xhigh, max; default medium). In T3 Code, picking Kiro and Claude Haiku 5.5 in a new thread shows "No effort levels" in the composer picker, and Settings → Providers → Kiro labels the model "No reasoning". Sonnet 5.5 and Opus 5.5 on the same instance do show Reasoning.

Change

Kiro's per-model reasoning levels come from the KIRO_MODEL_EFFORT table in packages/provider-kiro/src/server/status.ts, because kiro-cli chat --list-models names none. That table had no claude-haiku-5.5 entry. This adds it with the Claude levels and medium as the default, and notes in the table's comment which Kiro version this entry was read from. Claude Haiku 4.5 stays out: Kiro advertises no effort for it. The adapter still sends only a level the live session advertises, so it needs no change.

Scope and approval

Fixes #30920

Verification

  • vp test run apps/server/src/provider/KiroProvider.test.ts: 12/12 pass. offers each model only the reasoning levels Kiro takes for it now includes claude-haiku-5.5. On main (9e85538) it fails: Haiku 5.5 gets no option descriptors instead of the Reasoning select.
  • vp run check:changed: format, lint, knip, typecheck (10 packages) and tests for the 2 changed files pass.
  • Runtime check on a fresh dev server (no seeded data), with a synthetic kiro-cli stand-in that reports 2.37.1 and lists auto, claude-sonnet-5.5, claude-haiku-5.5 and claude-haiku-4.5. Same steps on both sides: new thread → Kiro → claude-haiku-5.5, then Settings → Providers → Kiro.
Surface Before (9e85538) After (726843a)
Composer model chip claude-haiku-5.5 claude-haiku-5.5 Medium
Composer picker side pane "Effort: No effort levels" "Reasoning": Low, Medium (checked by default), High, Extra High, Max
Settings → Providers → Kiro models claude-haiku-5.5 and claude-haiku-4.5 both labelled "No reasoning" only claude-haiku-4.5 labelled "No reasoning"
claude-sonnet-5.5 / claude-haiku-4.5 Reasoning (High) / no effort unchanged

Captures (web, 1280×800 viewport, light theme, cropped to the changed region):

Before (9e85538) After (726843a)
Before: Kiro claude-haiku-5.5 picker shows No effort levels After: Kiro claude-haiku-5.5 picker shows Reasoning levels with Medium default
Before: Settings marks claude-haiku-5.5 No reasoning After: Settings marks only claude-haiku-4.5 No reasoning

Not checked: a real kiro-cli 2.37.1 session with a signed-in account. The level actually reaching Kiro as effortLevel goes through the unchanged adapter path covered by the existing Kiro adapter tests. Not checked on mobile, which reads the same server capabilities.

AI-assisted; verified with the tests and tools listed above.

juliusmarminge and others added 2 commits November 4, 2026 15:32
Kiro lists claude-haiku-5.5 and its sessions advertise effort levels for it,
but the picker gives it no Reasoning select. Extend the per-model reasoning
test to expect low through max with medium as the default.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Kiro advertises effort levels low through max for claude-haiku-5.5, with
medium as the current value, but the per-model effort table had no entry for
it, so the picker showed no Reasoning select. Add the entry alongside the
other Claude models. Haiku 4.5 still takes no effort, as Kiro offers none.

Fixes #30920

AI-assisted; verified with the tests and tools listed above.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
@github-actions github-actions Bot added vouch:trusted Author is trusted by repo permissions or the VOUCHED list. size:XS 0-9 changed lines (additions + deletions). labels Nov 4, 2026
@macroscopeapp

macroscopeapp Bot commented Nov 4, 2026 •

Copy link
Copy Markdown
Contributor

Approvability

Verdict: Needs Human Approval

Macroscope's review found this PR requires human approval — Claude Haiku 5.5 gains reasoning-effort choices with medium marked as the default. Human confirmation is required for that product-default choice; existing live-session validation limits which levels can be sent.

No code changes detected at 1cb66a3. Prior analysis still applies.

You can add or adjust custom eligibility rules. Learn more.

@coderabbitai

coderabbitai Bot commented Nov 4, 2026 •

Copy link
Copy Markdown
Contributor

Explain Change →

No actionable comments were generated in the recent review.

Walkthrough

The Kiro effort mapping now includes Claude Haiku 5.5 with five effort levels and a medium default. Provider tests verify the model’s reasoning-level options.

Changes: Kiro Haiku 5.5 effort support

Layer / File(s) Summary
Effort table and regression coverage
packages/provider-kiro/src/server/status.ts, apps/server/src/provider/KiroProvider.test.ts
The effort table adds claude-haiku-5.5 with Claude’s effort levels and a medium default. The comment identifies Kiro 2.37.1 as the support version for Haiku 5.5. Tests verify all five levels, with medium as the default and current value.

Estimated code review effort: 1 (Trivial) | ~5 minutes

Change: Bug fix · Severity of issue fixed: Medium

Possibly related PRs

Suggested reviewers: saphid

Merge Risk: Minimal · up to 1cb66

Haiku 5.5 now has the requested reasoning options and medium default; no actionable merge-blocking issue is established.

Pre-merge checks: Passed 4

Check name Status Explanation
Title check Passed The title clearly and concisely describes the change: adding reasoning effort options for Claude Haiku 5.5 in Kiro.
Description check Passed The description covers the problem, change, scope, and verification. It includes focused tests, runtime comparison details, and limitations. The linked issue supports the stated scope.
Linked Issues check Passed Issue [#30920] requires the Haiku 5.5 Reasoning levels and a medium default. status.ts adds claude-haiku-5.5 with the Claude levels and that default. The test adds the expected levels and curren…
Out of Scope Changes check Passed Both changed files support issue [#30920]. The table entry implements the missing model capability. The test verifies its levels and default. The comment records the Kiro version for the capability da…
Full details: Linked Issues check

Explanation

Issue [#30920] requires the Haiku 5.5 Reasoning levels and a medium default. status.ts adds claude-haiku-5.5 with the Claude levels and that default. The test adds the expected levels and current value. The existing applyKiroEffort path sends the selected value as effortLevel only when the live session advertises it. Haiku 4.5 remains unlisted.

Full details: Out of Scope Changes check

Explanation

Both changed files support issue [#30920]. The table entry implements the missing model capability. The test verifies its levels and default. The comment records the Kiro version for the capability data. No unrelated changes appear in the whole-PR diff.

Finishing Touches

Generate docstrings

  • Commit to this branch
  • Create a new PR
Generate unit tests (beta)

  • Commit to this branch
  • Create a new PR
  • Autofix · Keep fixing CodeRabbit findings and required CI, and resolving merge conflicts

Usage-based review receipt

Note

This review was completed with usage-based billing: files reviewed beyond your plan's included limits are billed at $0.25/file. View usage-based billing.

Recent review info

Run configuration

Commits

Reviewing files that changed from the base of the PR and between 726843a and 1cb66a3.

Files selected for processing: 1

  • packages/provider-kiro/src/server/status.ts

Limit details: You’ve used all 10 included reviews currently available.

Copy link
Copy Markdown
Member Author

Before/after captures for the Verification table, from a fresh dev server with a synthetic kiro-cli stand-in (reports 2.37.1; lists auto, claude-sonnet-5.5, claude-haiku-5.5, claude-haiku-4.5). Web, 1280×800 viewport, light theme, cropped to the changed region.

  1. Composer picker, Kiro → claude-haiku-5.5, before (main 9e85538): "No effort levels", chip without a level.
  2. Same, after (726843a): Reasoning pane with Low, Medium (Default), High, Extra High, Max; chip "claude-haiku-5.5 Medium".
  3. Settings → Providers → Kiro models, before: claude-haiku-5.5 and claude-haiku-4.5 both "No reasoning".
  4. Same, after: only claude-haiku-4.5 "No reasoning".

Before: Kiro claude-haiku-5.5 picker shows No effort levels

After: Kiro claude-haiku-5.5 picker shows Reasoning levels with Medium default

Before: Settings marks claude-haiku-5.5 No reasoning

After: Settings marks only claude-haiku-4.5 No reasoning

Copy link
Copy Markdown
Member Author

The medium default is the level kiro-cli 2.37.1 itself reports for claude-haiku-5.5: in the issue, session/set_config_option over kiro-cli acp --agent-engine=v3 --auth-method=cli returns effortLevel with levels low/medium/high/xhigh/max and current value medium. The table mirrors that, as it does for claude-opus-5.5 (also medium). The adapter still sends only a level the live session advertises.

@github-actions

github-actions Bot commented Nov 4, 2026 •

Copy link
Copy Markdown
Contributor

Thread transfer impact

✅ Thread transfer remains within every enforced ceiling.

Provider Metric Main baseline This PR Impact PR ceiling
Codex Total thread wire 5.0 KiB 5.0 KiB 0 B (0.0%) 6.8 KiB ✅
Codex Thread snapshot wire 3.8 KiB 3.8 KiB 0 B (0.0%) 4.9 KiB ✅
Codex Live turn WebSocket wire 1.2 KiB 1.2 KiB 0 B (0.0%) 2.0 KiB ✅
Codex Live turn WebSocket decoded 21.5 KiB 21.5 KiB 0 B (0.0%) 24.4 KiB ✅
Codex Live turn messages 1 1 0 (0.0%) 8 ✅
Claude Total thread wire 5.0 KiB 5.0 KiB 0 B (0.0%) 6.8 KiB ✅
Claude Thread snapshot wire 3.8 KiB 3.8 KiB 0 B (0.0%) 4.9 KiB ✅
Claude Live turn WebSocket wire 1.2 KiB 1.2 KiB 0 B (0.0%) 2.0 KiB ✅
Claude Live turn WebSocket decoded 21.9 KiB 21.9 KiB 0 B (0.0%) 24.4 KiB ✅
Claude Live turn messages 1 1 0 (0.0%) 8 ✅

Baseline: 17a6a01 · PR result: 1cb66a3 · Source CI: success

Scenario and decoded snapshot size

10 historical turns, 5 command tools per turn, 878.9 KiB retained MCP result per historical turn, and a 1.05 MiB retained result in the measured turn.

  • Codex decoded thread snapshot: 108.5 KiB
  • Claude decoded thread snapshot: 108.8 KiB

Updated in place by a trusted workflow. PR artifacts are strictly validated and never executed.

@juliusmarminge juliusmarminge added bugfix Fix PR opened by an issue lead; enters the PR loop pr:gate-failed A land gate failed labels Nov 4, 2026

Copy link
Copy Markdown
Contributor

Multiple models have missing levels like Astra, GPT 6 Luna, GPT 6.1 Sol.

Copy link
Copy Markdown

Would it make sense to follow up with dynamic effort discovery from the live Kiro session? We could keep this table as a fallback for new users, then use and store the levels/defaults Kiro advertises, updating them as we discover changes. That way they’d still be available when the CLI isn’t running.

Looks like this was deferred in the original Kiro PR. Is there still something blocking it, or would a separate follow-up PR be welcome?

Copy link
Copy Markdown
Member Author

The automated pipeline stopped on this pull request and will not merge it:

the proof has no capture anyone can open:

  • Base composer and Settings behavior: kiro-cli was absent, and no Kiro account state was found in the checked standard local locations.
  • Head effort levels, Medium default, Settings labels, unchanged Haiku 4.5 behavior, and live-session validation: the required signed-in Kiro account flow could not be run.
  • Freshness of the base ref: origin fetch failed with HTTP 503; baseSha uses the existing local origin/main ref.
  • No captures or driver scripts were produced or uploaded. ghdrop upload was not attempted.

It stays open for a person to finish or close.

@vinaytiparadi

Copy link
Copy Markdown

Still reproduces with kiro-cli 3.2.0 on the 2026-11-11 nightly: Haiku 5.5 has no Reasoning option.

Grok 4.7 has the same gap. Kiro advertises low, medium, high, xhigh (default high) for it, but T3 shows no Reasoning option.

On the dynamic-discovery follow-up: chat --list-models doesn't include effort, but ACP session/new does. Each entry in the model config option carries it in _meta.kiro:

{"value": "claude-haiku-5.5", "_meta": {"kiro": {"hasEffort": true, "effortLevels": ["low","medium","high","xhigh","max"], "defaultEffortLevel": "medium"}}}

Reading that would remove the hardcoded table. Kiro's own default for Haiku 5.5 is medium, which matches what this PR picked. I compared every public model on 3.2.0: except Haiku 5.5 and Grok 4.7, T3's table matches Kiro. auto reports hasEffort: false, so having no Reasoning option there is correct.

@willgosnold

Copy link
Copy Markdown

Note

🤖 Claude Opus 5.5 responding on behalf of @willgosnold

@juliusmarminge verified against a real signed-in Kiro account, which the pipeline could not run: this PR fixes #30920. Linux, kiro-cli 2.37.1, T3 dev server from source.

  • main at 2d8a843: Settings → Providers → Kiro labels claude-haiku-5.5 "No reasoning", and the composer picker offers only "Agent".
  • This PR at 1cb66a3: the picker shows Reasoning with Low / Medium (default) / High / Extra High / Max, and the chip reads claude-haiku-5.5 Medium. claude-haiku-4.5 stays "No reasoning".
  • A real turn on High completed normally. KiroAdapterV2.applyKiroEffort and session/set_config_option both succeeded in the server trace. The trace does not record the value sent, so I did not confirm high itself on the wire.
main 2d8a843 PR 1cb66a3
Composer picker main: claude-haiku-5.5 picker shows only the Agent option PR: claude-haiku-5.5 picker shows Reasoning levels with Medium selected
Settings models main: claude-haiku-5.5 labelled No reasoning PR: claude-haiku-5.5 no longer labelled No reasoning

+1 to the dynamic effort discovery follow-up, since other Kiro models show the same gap.

This branch has not been deployed

No deployments
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

bugfix Fix PR opened by an issue lead; enters the PR loop pr:gate-failed A land gate failed size:XS 0-9 changed lines (additions + deletions). vouch:trusted Author is trusted by repo permissions or the VOUCHED list.

Projects

None yet

Development

Successfully merging this pull request may close these issues.

[Bug]: Kiro provider has no Reasoning option for Claude Haiku 5.5, though kiro-cli advertises effortLevel for it

5 participants