dsh-kiro
Kiro provider for DeepSeek Harness with multi-method login, usage, model controls, and token/profile refresh 适合使用Kiro服务并需要在DSH中直接管理其模型和额度的用户。
Other repositories with this package name
Install
$
npx -p @deepseek-ai/dsh dsh plugin --profile web add github:liujiesmart/dsh-kiroREADME
Read the full README ↗Configuration
Put machine-level configuration in $DSH_HOME/settings.yaml (normally ~/.dsh/settings.yaml):
llm-kiro:
proxyUrl: http://proxy.example:1082
reasoningEffort: medium
Every field is optional:
| Field | Default | Meaning |
|---|---|---|
proxyUrl | direct | HTTP/HTTPS proxy for Kiro and OIDC requests; credentials in the URL are supported. |
region | signed-in token region | Selects the q..amazonaws.com endpoint. |
profileArn | account default | CodeWhisperer profile used for requests and model discovery. |
thinking | enabled | disabled suppresses reasoning controls and native effort fields. |
reasoningEffort | model's live default | Optional override: none, off, low, medium, high, xhigh, or max; unsupported values are rejected for that model. |
defaultContextWindow | 200000 | Fallback capacity when discovery reports no exact limit. |
models | bundled fallback | Advisory fallback catalog; live account discovery normally replaces it. |
streamIdleTimeoutMs | 300000 | Maximum idle time while a provider read is outstanding. |
tokenExpiryBufferMs | 300000 | Refresh an access token this long before expiration. |
retryPolicy | bounded normal | Provider retry policy executed by dsh-llm-retry. |
refusalRecovery | narrow | narrow retries a silent refusal with a progressively smaller request; off reports the first refusal unchanged. |
refusalRecoveryAttempts | 3 | How many narrowing retries a refusal may spend (0–3). |
To pin values to one profile instead, patch the bundle row by id in $DSH_HOME/profiles//cordis.patch.yml:
- id: llm-kiro
config:
proxyUrl: http://proxy.example:1082
Do not wrap this override in insert:; the bundle already inserts llm-kiro.