fix: distinct signal exit codes, mid-stream API error surfacing, require model name with base URL by min0625 · Pull Request #54 · min0625/mint

min0625 · 2026-07-02T06:28:53Z

Summary

Map SIGINT/SIGTERM to conventional 128+N exit codes (130/143) instead of always returning 130, by tracking the signal via context.WithCancelCause instead of signal.NotifyContext.
Surface mid-stream error events/objects from all three providers (Anthropic, OpenAI, Google GenAI) instead of silently returning empty/partial output — streaming responses return HTTP 200 before an error can appear later in the SSE stream.
Detect Anthropic max_tokens truncation via stop_reason and return an error instead of printing incomplete output with exit 0.
Require MINT_MODEL_NAME when MINT_BASE_URL is set, since a custom endpoint has no meaningful default model (e.g. Ollama would silently get sent a cloud provider's default model name).
Add LC_MESSAGES to the locale fallback chain (LC_ALL > LC_MESSAGES > LANG), matching POSIX priority.
Strip leading v from the locally-built version string so mint --version matches goreleaser's {{.Version}} for both local and released binaries.
Check out the release tag (not main HEAD) in publish-npm.yml/publish-pypi.yml so packaging scripts/README match the release being published.
Handle unset $SHELL in install.sh and signal-killed child exit status in the npm wrapper script.
Docs (AGENTS.md, all README variants) updated to match.

Test plan

make check (lint + go test -race -failfast) passes
New unit tests cover: SIGINT vs SIGTERM exit codes, LC_MESSAGES priority, mid-stream error events for all three providers, Anthropic truncation detection, config validation requiring model name with base URL

qodo-code-review · 2026-07-02T06:29:01Z

Qodo is busy working

Check back in a few minutes. Qodo's code review agents are on it.

codecov · 2026-07-02T06:29:25Z

Codecov Report

❌ Patch coverage is 85.93750% with 9 lines in your changes missing coverage. Please review.

Files with missing lines	Patch %	Lines
internal/provider/googlegenai/google_genai.go	76.47%	2 Missing and 2 partials ⚠️
internal/provider/anthropic/anthropic.go	85.71%	1 Missing and 1 partial ⚠️
internal/provider/openai/openai.go	84.61%	1 Missing and 1 partial ⚠️
cmd/mint/main.go	94.44%	1 Missing ⚠️

📢 Thoughts on this report? Let us know!

qodo-code-review · 2026-07-02T06:30:19Z

PR Summary by Qodo

Fix signal exit codes, stream error surfacing, and base URL model validation

🐞 Bug fix 🧪 Tests 📝 Documentation ⚙️ Configuration changes 🕐 40+ Minutes

AI Description

• Map SIGINT/SIGTERM to distinct conventional exit codes (130/143).
• Surface mid-stream SSE errors and Anthropic truncation as non-zero failures.
• Require model name for custom base URLs; align locale/version and release packaging.

Diagram

graph TD
  A["mint CLI"] --> B["Signal-aware context"] --> C["Root command exec"] --> D["Config + locale resolve"] --> E["Provider completer"] --> F{{"LLM SSE API"}}
  E --> G["Stream output + usage"]

High-Level Assessment

The following are alternative approaches to this PR:

1. Centralize SSE decoding/error surfacing in a shared stream layer

➕ Eliminates duplicated per-provider checks for mid-stream error objects/events
➕ Makes future provider additions more consistent and less error-prone
➕ Easier to unit test stream semantics once
➖ Requires designing a common event model across providers with different schemas
➖ Potentially more refactor churn than the current targeted fixes

2. Return partial streamed output but annotate and exit non-zero on truncation/errors

➕ Preserves currently-streamed output for interactive users
➕ Keeps failure semantics explicit for scripting
➖ Still risks consumers treating partial output as valid if they ignore exit codes
➖ Adds UX decisions (where/how to annotate) that vary by output mode

3. Keep signal.NotifyContext and separately track last signal via Notify channel

➕ Less custom context wiring in main()
➕ Retains standard library convenience
➖ Racy/ambiguous if multiple signals arrive or context cancels for other reasons
➖ Harder to guarantee the exit code matches the actual cancel cause

Recommendation: The PR’s approach is a good pragmatic balance: it fixes correctness issues (exit codes and silent stream failures) with minimal surface-area changes to each provider. If streaming logic grows further, consider consolidating SSE parsing/error extraction into a shared component to reduce duplication and ensure consistent semantics.

Files changed (19) +299 / -46

Bug fix (7) +118 / -15

main.goTrack signal cancel causes for correct exit codes; add LC_MESSAGES fallback +41/-8

Track signal cancel causes for correct exit codes; add LC_MESSAGES fallback

• Replaces signal.NotifyContext with context.WithCancelCause plus a signal channel to preserve which signal triggered cancellation, enabling conventional 128+N exit codes. Updates locale detection to follow LC_ALL > LC_MESSAGES > LANG.

cmd/mint/main.go

anthropic.goSurface Anthropic stream errors and max_tokens truncation +34/-4

Surface Anthropic stream errors and max_tokens truncation
• Parses mid-stream error events and returns them as errors even when HTTP status is 200. Detects output truncation via stop_reason=max_tokens and returns an explicit truncation error after streaming.
internal/provider/anthropic/anthropic.go

config.goRequire model name when a custom base URL is set +7/-0

Require model name when a custom base URL is set
• Adds validation rejecting configs that set MINT_BASE_URL without also providing MINT_MODEL_NAME to avoid meaningless provider defaults against custom endpoints.
internal/provider/config.go

google_genai.goTreat Google GenAI stream error objects as failures +13/-0

Treat Google GenAI stream error objects as failures
• Extends stream response parsing to detect an embedded error object and return it as an error, addressing cases where HTTP status is 200 but the stream reports failure later.
internal/provider/googlegenai/google_genai.go

openai.goTreat OpenAI-compatible stream error objects as failures +13/-0

Treat OpenAI-compatible stream error objects as failures
• Adds support for detecting a mid-stream error object in the SSE payload and returning an error instead of silently producing empty/partial output.
internal/provider/openai/openai.go

mintPropagate signal-killed child processes as 128+N exit codes +6/-0

Propagate signal-killed child processes as 128+N exit codes
• Detects when the wrapped binary exited due to a signal and exits with the conventional 128+N code instead of masking it as success (status null).
npm/mint-ai/scripts/mint

install.shHandle unset $SHELL under set -u during install +4/-3

Handle unset $SHELL under set -u during install
• Uses ${SHELL:-} in shell detection so the installer doesn't abort in minimal environments where SHELL is unset.
script/install.sh

Tests (5) +161 / -25

main_test.goAdd tests for LC_MESSAGES priority and distinct signal exit codes +33/-24

Add tests for LC_MESSAGES priority and distinct signal exit codes
• Extends getSystemLanguage coverage to validate LC_MESSAGES precedence and fallback behavior. Updates run() interruption test to assert SIGINT=130 and SIGTERM=143.
cmd/mint/main_test.go

anthropic_test.goTest Anthropic truncation detection and mid-stream error events +65/-0

Test Anthropic truncation detection and mid-stream error events
• Adds SSE-based tests ensuring max_tokens truncation becomes an error while still delivering already-streamed output and usage. Adds coverage for error events arriving mid-stream.
internal/provider/anthropic/anthropic_test.go

config_test.goAdd config validation tests for base URL + model name requirement +11/-1

Add config validation tests for base URL + model name requirement
• Updates existing base-URL test cases to include an explicit model. Adds new cases ensuring missing model name is rejected even when an API key is present.
internal/provider/config_test.go

google_genai_test.goTest Google GenAI mid-stream error object propagation +25/-0

Test Google GenAI mid-stream error object propagation
• Adds an SSE-based unit test asserting that a chunk containing an error object results in a non-nil error with details.
internal/provider/googlegenai/google_genai_test.go

openai_test.goTest OpenAI mid-stream error object propagation +27/-0

Test OpenAI mid-stream error object propagation
• Adds an SSE-based unit test ensuring a stream error object produces a non-nil error containing the server-provided message.
internal/provider/openai/openai_test.go

Documentation (4) +6 / -4

AGENTS.mdDocument LC_MESSAGES locale priority and internal modules +3/-1

Document LC_MESSAGES locale priority and internal modules

• Updates the documented target-language fallback chain to include LC_MESSAGES per POSIX. Also refreshes internal file pointers referenced by the agent docs.

AGENTS.md

README.ja.mdClarify model-name requirement when using custom base URL +1/-1

Clarify model-name requirement when using custom base URL
• Updates the environment variable table to state MINT_MODEL_NAME is required when MINT_BASE_URL is set.
README.ja.md

README.mdClarify model-name requirement when using custom base URL +1/-1

Clarify model-name requirement when using custom base URL
• Updates the environment variable table to state MINT_MODEL_NAME is required when MINT_BASE_URL is set.
README.md

README.zh-TW.mdClarify model-name requirement when using custom base URL +1/-1

Clarify model-name requirement when using custom base URL
• Updates the environment variable table to state MINT_MODEL_NAME is required when MINT_BASE_URL is set.
README.zh-TW.md

Other (3) +14 / -2

publish-npm.ymlCheckout release tag when publishing npm package +5/-0

Checkout release tag when publishing npm package
• Pins actions/checkout to the workflow_run tag ref so packaging scripts/README match the published release rather than main HEAD.
.github/workflows/publish-npm.yml

publish-pypi.ymlCheckout release tag and fix PyPI publish input name +6/-1

Checkout release tag and fix PyPI publish input name
• Checks out the release tag ref for publishing parity with the release content. Renames the PyPI publish input from skip_existing to the correct skip-existing key.
.github/workflows/publish-pypi.yml

MakefileNormalize local VERSION string by stripping leading 'v' +3/-1

Normalize local VERSION string by stripping leading 'v'
• Strips a leading 'v' from git tag-derived versions so local builds match goreleaser's {{.Version}} output.
Makefile

…s, require model name with base URL - Map SIGINT/SIGTERM to conventional 128+N exit codes (130/143) instead of always 130 - Surface mid-stream error events/objects from all three providers instead of silently returning empty/partial output - Detect Anthropic max_tokens truncation and return an error instead of printing incomplete output with exit 0 - Require MINT_MODEL_NAME when MINT_BASE_URL is set, since a custom endpoint has no meaningful default model - Add LC_MESSAGES to the locale fallback chain (LC_ALL > LC_MESSAGES > LANG) - Strip leading "v" from local build version so it matches goreleaser's version string - Check out the release tag (not main) in publish-npm/publish-pypi workflows - Handle unset $SHELL and killed-by-signal exit status in install.sh / npm wrapper script

min0625 force-pushed the fix/signal-exit-codes-and-stream-errors branch from e59623e to 0564c0f Compare July 2, 2026 07:34

min0625 force-pushed the fix/signal-exit-codes-and-stream-errors branch from 0564c0f to b5eba79 Compare July 2, 2026 11:56

min0625 merged commit de2ac90 into main Jul 2, 2026
6 checks passed

min0625 deleted the fix/signal-exit-codes-and-stream-errors branch July 2, 2026 12:01

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

fix: distinct signal exit codes, mid-stream API error surfacing, require model name with base URL#54

fix: distinct signal exit codes, mid-stream API error surfacing, require model name with base URL#54
min0625 merged 1 commit into
mainfrom
fix/signal-exit-codes-and-stream-errors

min0625 commented Jul 2, 2026

Uh oh!

qodo-code-review Bot commented Jul 2, 2026

Uh oh!

codecov Bot commented Jul 2, 2026 •

edited

Loading

Uh oh!

qodo-code-review Bot commented Jul 2, 2026

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

1 participant

Conversation

min0625 commented Jul 2, 2026

Summary

Test plan

Uh oh!

qodo-code-review Bot commented Jul 2, 2026

Qodo is busy working

Uh oh!

codecov Bot commented Jul 2, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Codecov Report

Uh oh!

qodo-code-review Bot commented Jul 2, 2026

PR Summary by Qodo

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

1 participant

codecov Bot commented Jul 2, 2026 •

edited

Loading