mirror of
https://github.com/stablyai/orca.git
synced 2026-11-02 10:08:13 +00:00
Brennan Bensonandbeduldul
47018d4844
* fix(source-control): read the Kimi answer from stream-json Kimi Code's text mode prefixes thinking and hook output with a bullet, so the commit message arrived mixed with narration. The recipe now asks for --output-format stream-json and the answer is taken from the last assistant record, which is the commit message (#33556). A record that requests a tool is narration rather than an answer, so it clears the candidate: a stream that ends on a tool call yields no answer, which finalizeFromAgentOutput already treats as a failed run. Only a body whose every non-empty line is a record counts as stream output, so a plain answer that happens to contain a JSON line is left alone. * fix(source-control): decode Kimi output only for the selected protocol * fix(source-control): satisfy generation hook checks * Preserve Kimi generation failures and normalize options linearly * Fix Kimi argument failure messages * Bound Kimi option failure recognition * Fix Kimi generation test fixture types --------- Co-authored-by: beduldul <alkaysan07@gmail.com>
296 lines
9.1 KiB
TypeScript
296 lines
9.1 KiB
TypeScript
import type { TuiAgent } from './tui-agent'
|
||
import type {
|
||
CommitMessageAgentSpec,
|
||
CommitMessageModel,
|
||
ThinkingLevel
|
||
} from './commit-message-agent-spec'
|
||
|
||
type SecondaryAgentSpecDeps = {
|
||
BASIC_THINKING_LEVELS: ThinkingLevel[]
|
||
OPENAI_THINKING_LEVELS: ThinkingLevel[]
|
||
parseCursorModels: (stdout: string) => CommitMessageModel[]
|
||
parseAntigravityModels: (stdout: string) => CommitMessageModel[]
|
||
parseLineModels: (stdout: string) => CommitMessageModel[]
|
||
}
|
||
|
||
export function buildSecondaryCommitMessageAgentSpecs({
|
||
BASIC_THINKING_LEVELS,
|
||
OPENAI_THINKING_LEVELS,
|
||
parseCursorModels,
|
||
parseAntigravityModels,
|
||
parseLineModels
|
||
}: SecondaryAgentSpecDeps): Partial<Record<TuiAgent, CommitMessageAgentSpec>> {
|
||
return {
|
||
amp: {
|
||
id: 'amp',
|
||
label: 'Amp',
|
||
binary: 'amp',
|
||
promptDelivery: 'stdin',
|
||
buildArgs: ({ model, thinkingLevel }) => [
|
||
'--execute',
|
||
'--no-notifications',
|
||
'--no-ide',
|
||
'--no-jetbrains',
|
||
'--mode',
|
||
model,
|
||
...(thinkingLevel ? ['--effort', thinkingLevel] : [])
|
||
],
|
||
// Amp selects the model with `--mode`, not `--model`.
|
||
singletonOptions: [['--mode']],
|
||
modelSource: 'static',
|
||
models: [
|
||
{ id: 'smart', label: 'Smart' },
|
||
{ id: 'rush', label: 'Rush' },
|
||
{
|
||
id: 'large',
|
||
label: 'Large',
|
||
thinkingLevels: BASIC_THINKING_LEVELS,
|
||
defaultThinkingLevel: 'low'
|
||
},
|
||
{
|
||
id: 'deep',
|
||
label: 'Deep',
|
||
thinkingLevels: BASIC_THINKING_LEVELS,
|
||
defaultThinkingLevel: 'low'
|
||
}
|
||
],
|
||
defaultModelId: 'smart'
|
||
},
|
||
cursor: {
|
||
id: 'cursor',
|
||
label: 'Cursor',
|
||
binary: 'cursor-agent',
|
||
promptDelivery: 'argv',
|
||
buildArgs: ({ prompt, model }) => [
|
||
'--print',
|
||
'--mode',
|
||
'ask',
|
||
'--trust',
|
||
'--output-format',
|
||
'text',
|
||
'--model',
|
||
model,
|
||
prompt
|
||
],
|
||
modelSource: 'dynamic',
|
||
modelDiscovery: { binary: 'cursor-agent', args: ['--list-models'], parse: parseCursorModels },
|
||
models: [{ id: 'auto', label: 'Auto' }],
|
||
defaultModelId: 'auto'
|
||
},
|
||
kimi: {
|
||
id: 'kimi',
|
||
label: 'Kimi',
|
||
binary: 'kimi',
|
||
// kimi-code requires --prompt/-p text on argv; Claude-style --print is rejected (#11669).
|
||
promptDelivery: 'argv',
|
||
singletonOptions: [['--model', '-m'], ['--output-format']],
|
||
// Why: Kimi Code has no --quiet/--thinking/--no-thinking; -p alone is headless (#14771).
|
||
buildArgs: ({ prompt, model }) => [
|
||
'--prompt',
|
||
prompt,
|
||
// Text mode gives hooks and assistant replies the same presentation bullet.
|
||
'--output-format',
|
||
'stream-json',
|
||
...(model && model !== 'default' ? ['--model', model] : [])
|
||
],
|
||
modelSource: 'static',
|
||
models: [
|
||
{ id: 'default', label: 'Config default' },
|
||
{
|
||
// Kimi requires provider/model IDs; bare names fail with LLM not set.
|
||
id: 'kimi-code/kimi-for-coding',
|
||
label: 'Kimi K2.6'
|
||
}
|
||
],
|
||
defaultModelId: 'default'
|
||
},
|
||
muse: {
|
||
id: 'muse',
|
||
label: 'Muse',
|
||
binary: 'muse',
|
||
// Muse exec takes a positional prompt; keep generation one-shot and read-only.
|
||
promptDelivery: 'argv',
|
||
buildArgs: ({ prompt, model, thinkingLevel }) => [
|
||
'exec',
|
||
'--no-session-log',
|
||
'--approval-mode',
|
||
'never',
|
||
'--disable-sandbox',
|
||
'--disable-shell',
|
||
'--disable-write',
|
||
'--disable-web-tools',
|
||
...(model && model !== 'default' ? ['--model', model] : []),
|
||
...(thinkingLevel ? ['--reasoning-effort', thinkingLevel] : []),
|
||
'--',
|
||
prompt
|
||
],
|
||
singletonOptions: [['--model'], ['--reasoning-effort']],
|
||
modelSource: 'static',
|
||
models: [{ id: 'default', label: 'Config default' }],
|
||
defaultModelId: 'default'
|
||
},
|
||
dsh: {
|
||
id: 'dsh',
|
||
label: 'DeepSeek Harness',
|
||
binary: 'dsh',
|
||
// DSH headless runs one persisted session, prints and exits. Use stdin to hide diffs from process listings and avoid argv limits. 0.1.5-rc.1 requires explicit - even with piped stdin.
|
||
promptDelivery: 'stdin',
|
||
buildArgs: () => ['--profile', 'headless', '-'],
|
||
// Why: the launcher owns `--profile`; a second one would boot a different profile.
|
||
singletonOptions: [['--profile']],
|
||
modelSource: 'static',
|
||
// Headless has no --model; its llm-deepseek profile row owns selection, so offer only configured default.
|
||
models: [{ id: 'default', label: 'Config default' }],
|
||
defaultModelId: 'default'
|
||
},
|
||
copilot: {
|
||
id: 'copilot',
|
||
label: 'GitHub Copilot',
|
||
binary: 'copilot',
|
||
promptDelivery: 'argv',
|
||
buildArgs: ({ prompt, model, thinkingLevel }) => [
|
||
'--prompt',
|
||
prompt,
|
||
'--silent',
|
||
'--stream',
|
||
'off',
|
||
'--no-custom-instructions',
|
||
'--model',
|
||
model,
|
||
...(thinkingLevel ? ['--effort', thinkingLevel] : [])
|
||
],
|
||
modelSource: 'static',
|
||
// Copilot filters the picker by account/org policy; retain the full hosted catalog for permitted selections.
|
||
models: [
|
||
{ id: 'auto', label: 'Auto' },
|
||
{
|
||
id: 'claude-haiku-4.5',
|
||
label: 'Claude Haiku 4.5'
|
||
},
|
||
{
|
||
id: 'claude-sonnet-4.5',
|
||
label: 'Claude Sonnet 4.5'
|
||
},
|
||
{
|
||
id: 'claude-sonnet-4.6',
|
||
label: 'Claude Sonnet 4.6'
|
||
},
|
||
{
|
||
id: 'claude-opus-4.5',
|
||
label: 'Claude Opus 4.5'
|
||
},
|
||
{
|
||
id: 'claude-opus-4.6',
|
||
label: 'Claude Opus 4.6'
|
||
},
|
||
{
|
||
id: 'claude-opus-4.6-fast',
|
||
label: 'Claude Opus 4.6 Fast'
|
||
},
|
||
{
|
||
id: 'claude-opus-4.7',
|
||
label: 'Claude Opus 4.7'
|
||
},
|
||
{
|
||
id: 'gpt-4.1',
|
||
label: 'GPT-4.1'
|
||
},
|
||
{
|
||
id: 'gpt-5-mini',
|
||
label: 'GPT-5 Mini',
|
||
thinkingLevels: OPENAI_THINKING_LEVELS,
|
||
defaultThinkingLevel: 'low'
|
||
},
|
||
{
|
||
id: 'gpt-5.2',
|
||
label: 'GPT-5.2',
|
||
thinkingLevels: OPENAI_THINKING_LEVELS,
|
||
defaultThinkingLevel: 'low'
|
||
},
|
||
{
|
||
id: 'gpt-5.2-codex',
|
||
label: 'GPT-5.2 Codex',
|
||
thinkingLevels: OPENAI_THINKING_LEVELS,
|
||
defaultThinkingLevel: 'low'
|
||
},
|
||
{
|
||
id: 'gpt-5.3-codex',
|
||
label: 'GPT-5.3 Codex',
|
||
thinkingLevels: OPENAI_THINKING_LEVELS,
|
||
defaultThinkingLevel: 'low'
|
||
},
|
||
{
|
||
id: 'gpt-5.4',
|
||
label: 'GPT-5.4',
|
||
thinkingLevels: OPENAI_THINKING_LEVELS,
|
||
defaultThinkingLevel: 'low'
|
||
},
|
||
{
|
||
id: 'gpt-5.4-mini',
|
||
label: 'GPT-5.4 Mini',
|
||
thinkingLevels: OPENAI_THINKING_LEVELS,
|
||
defaultThinkingLevel: 'low'
|
||
},
|
||
{
|
||
id: 'gpt-5.5',
|
||
label: 'GPT-5.5',
|
||
thinkingLevels: OPENAI_THINKING_LEVELS,
|
||
defaultThinkingLevel: 'low'
|
||
}
|
||
],
|
||
defaultModelId: 'gpt-5.4'
|
||
},
|
||
antigravity: {
|
||
id: 'antigravity',
|
||
label: 'Antigravity',
|
||
binary: 'agy',
|
||
// Empty --print selects stream input without consuming the next flag (agy >= 1.1.15).
|
||
promptDelivery: 'stdin',
|
||
buildArgs: ({ model, thinkingLevel }) => [
|
||
'--print=',
|
||
'--input-format=stream-json',
|
||
'--output-format=stream-json',
|
||
'--sandbox',
|
||
...(model && model !== 'default' ? ['--model', model] : []),
|
||
...(thinkingLevel ? ['--effort', thinkingLevel] : [])
|
||
],
|
||
singletonOptions: [['--model'], ['--effort']],
|
||
modelSource: 'dynamic',
|
||
modelDiscovery: { binary: 'agy', args: ['models'], parse: parseAntigravityModels },
|
||
models: [{ id: 'default', label: 'Config default' }],
|
||
defaultModelId: 'default'
|
||
},
|
||
jcode: {
|
||
id: 'jcode',
|
||
label: 'Jcode',
|
||
binary: 'jcode',
|
||
// jcode run supports positional argv only; large diffs remain subject to Windows argv limits.
|
||
promptDelivery: 'argv',
|
||
buildArgs: ({ prompt, model }) => [
|
||
// clap accepts these global flags only before run.
|
||
'--no-update',
|
||
'--quiet',
|
||
'--no-selfdev',
|
||
// Staged patches are attacker-influenced; none removes shell/read/write and MCP via tools.rs base_allowed_tools, matching read-only siblings.
|
||
'--tool-profile',
|
||
'none',
|
||
...(model && model !== 'default' ? ['--model', model] : []),
|
||
'run',
|
||
'--json',
|
||
prompt
|
||
],
|
||
singletonOptions: [['--model']],
|
||
modelSource: 'dynamic',
|
||
// Read bare model IDs from jcode’s authenticated-provider catalog instead of hardcoding it.
|
||
modelDiscovery: {
|
||
binary: 'jcode',
|
||
args: ['--no-update', '--quiet', 'model', 'list'],
|
||
parse: parseLineModels
|
||
},
|
||
// default omits --model so jcode’s config.toml chooses an authenticated provider; it is not a real model ID.
|
||
models: [{ id: 'default', label: 'Config default' }],
|
||
defaultModelId: 'default'
|
||
}
|
||
}
|
||
}
|