mirror of
https://github.com/NousResearch/hermes-agent.git
synced 2026-04-26 01:01:40 +00:00
Revisits #13729. That PR buffered each `tool.complete`'s inline_diff and merged them into the final assistant message body as a fenced ```diff block. The merge-at-end placement reads as "the agent wrote this after the summary", even when the edit fired mid-turn — which is both misleading and (per blitz feedback) feels like noise tacked onto the end of every task. Segment-anchored placement instead: - On tool.complete with inline_diff, `pushInlineDiffSegment` calls `flushStreamingSegment` first (so any in-progress narration lands as its own segment), then pushes the ```diff block as its own segment into segmentMessages. The diff is now anchored BETWEEN the narration that preceded the edit and whatever the agent streams afterwards, which is where the edit actually happened. - `recordMessageComplete` no longer merges buffered diffs. The only remaining dedupe is "drop diff-only segments whose body the final assistant text narrates verbatim (or whose diff fence the final text already contains)" — same tradeoff as before, kept so an agent that narrates its own diff doesn't render two stacked copies. - Drops `pendingInlineDiffs` and `queueInlineDiff` — buffer + end- merge machinery is gone; segmentMessages is now the only source of truth. Side benefit: Ctrl+C interrupt (`interruptTurn`) iterates segmentMessages, so diff segments are now preserved in the transcript when the user cancels after an edit. Previously the pending buffer was silently dropped on interrupt. Reported by Teknium during blitz usage: "no diffs are ever at the end because it didn't make this file edit after the final message".
578 lines
18 KiB
TypeScript
578 lines
18 KiB
TypeScript
import { REASONING_PULSE_MS, STREAM_BATCH_MS } from '../config/timing.js'
|
|
import type { SessionInterruptResponse, SubagentEventPayload } from '../gatewayTypes.js'
|
|
import { hasReasoningTag, splitReasoning } from '../lib/reasoning.js'
|
|
import {
|
|
buildToolTrailLine,
|
|
estimateTokensRough,
|
|
isTransientTrailLine,
|
|
sameToolTrailGroup,
|
|
toolTrailLabel
|
|
} from '../lib/text.js'
|
|
import type { ActiveTool, ActivityItem, Msg, SubagentProgress } from '../types.js'
|
|
|
|
import { resetFlowOverlays } from './overlayStore.js'
|
|
import { pushSnapshot } from './spawnHistoryStore.js'
|
|
import { getTurnState, patchTurnState, resetTurnState } from './turnStore.js'
|
|
import { getUiState, patchUiState } from './uiStore.js'
|
|
|
|
const INTERRUPT_COOLDOWN_MS = 1500
|
|
const ACTIVITY_LIMIT = 8
|
|
const TRAIL_LIMIT = 8
|
|
|
|
// Matches segments produced by pushInlineDiffSegment — a bare ```diff fence
|
|
// wrapping the raw patch, no surrounding prose. Used at message.complete to
|
|
// dedupe against final assistant text that narrates the same patch.
|
|
const DIFF_SEGMENT_RE = /^```diff\n([\s\S]*?)\n```$/
|
|
|
|
const diffSegmentBody = (msg: Msg): null | string => {
|
|
if (msg.role !== 'assistant' || msg.tools?.length) {
|
|
return null
|
|
}
|
|
|
|
const m = msg.text.match(DIFF_SEGMENT_RE)
|
|
|
|
return m ? m[1]! : null
|
|
}
|
|
|
|
export interface InterruptDeps {
|
|
appendMessage: (msg: Msg) => void
|
|
gw: { request: <T = unknown>(method: string, params?: Record<string, unknown>) => Promise<T> }
|
|
sid: string
|
|
sys: (text: string) => void
|
|
}
|
|
|
|
type Timer = null | ReturnType<typeof setTimeout>
|
|
|
|
const clear = (t: Timer): null => {
|
|
if (t) {
|
|
clearTimeout(t)
|
|
}
|
|
|
|
return null
|
|
}
|
|
|
|
class TurnController {
|
|
bufRef = ''
|
|
interrupted = false
|
|
lastStatusNote = ''
|
|
persistedToolLabels = new Set<string>()
|
|
persistSpawnTree?: (subagents: SubagentProgress[], sessionId: null | string) => Promise<void>
|
|
protocolWarned = false
|
|
reasoningText = ''
|
|
segmentMessages: Msg[] = []
|
|
pendingSegmentTools: string[] = []
|
|
statusTimer: Timer = null
|
|
toolTokenAcc = 0
|
|
turnTools: string[] = []
|
|
|
|
private activeTools: ActiveTool[] = []
|
|
private activityId = 0
|
|
private reasoningStreamingTimer: Timer = null
|
|
private reasoningTimer: Timer = null
|
|
private streamTimer: Timer = null
|
|
private toolProgressTimer: Timer = null
|
|
|
|
clearReasoning() {
|
|
this.reasoningTimer = clear(this.reasoningTimer)
|
|
this.reasoningText = ''
|
|
this.toolTokenAcc = 0
|
|
patchTurnState({ reasoning: '', reasoningTokens: 0, toolTokens: 0 })
|
|
}
|
|
|
|
clearStatusTimer() {
|
|
this.statusTimer = clear(this.statusTimer)
|
|
}
|
|
|
|
endReasoningPhase() {
|
|
this.reasoningStreamingTimer = clear(this.reasoningStreamingTimer)
|
|
patchTurnState({ reasoningActive: false, reasoningStreaming: false })
|
|
}
|
|
|
|
idle() {
|
|
this.endReasoningPhase()
|
|
this.activeTools = []
|
|
this.streamTimer = clear(this.streamTimer)
|
|
this.bufRef = ''
|
|
this.pendingSegmentTools = []
|
|
this.segmentMessages = []
|
|
|
|
patchTurnState({
|
|
streamPendingTools: [],
|
|
streamSegments: [],
|
|
streaming: '',
|
|
subagents: [],
|
|
tools: [],
|
|
turnTrail: []
|
|
})
|
|
patchUiState({ busy: false })
|
|
resetFlowOverlays()
|
|
}
|
|
|
|
interruptTurn({ appendMessage, gw, sid, sys }: InterruptDeps) {
|
|
this.interrupted = true
|
|
gw.request<SessionInterruptResponse>('session.interrupt', { session_id: sid }).catch(() => {})
|
|
|
|
const segments = this.segmentMessages
|
|
const partial = this.bufRef.trimStart()
|
|
const tools = this.pendingSegmentTools
|
|
|
|
// Drain streaming/segment state off the nanostore before writing the
|
|
// preserved snapshot to the transcript — otherwise each flushed segment
|
|
// appears in both `turn.streamSegments` and the transcript for one frame.
|
|
this.idle()
|
|
this.clearReasoning()
|
|
this.turnTools = []
|
|
patchTurnState({ activity: [], outcome: '' })
|
|
|
|
for (const msg of segments) {
|
|
appendMessage(msg)
|
|
}
|
|
|
|
// Always surface an interruption indicator — if there's an in-flight
|
|
// `partial` or pending tools, fold them into a single assistant message;
|
|
// otherwise emit a sys note so the transcript always records that the
|
|
// turn was cancelled, even when only prior `segments` were preserved.
|
|
if (partial || tools.length) {
|
|
appendMessage({
|
|
role: 'assistant',
|
|
text: partial ? `${partial}\n\n*[interrupted]*` : '*[interrupted]*',
|
|
...(tools.length && { tools })
|
|
})
|
|
} else {
|
|
sys('interrupted')
|
|
}
|
|
|
|
patchUiState({ status: 'interrupted' })
|
|
this.clearStatusTimer()
|
|
|
|
this.statusTimer = setTimeout(() => {
|
|
this.statusTimer = null
|
|
patchUiState({ status: 'ready' })
|
|
}, INTERRUPT_COOLDOWN_MS)
|
|
}
|
|
|
|
pruneTransient() {
|
|
this.turnTools = this.turnTools.filter(line => !isTransientTrailLine(line))
|
|
patchTurnState(state => {
|
|
const next = state.turnTrail.filter(line => !isTransientTrailLine(line))
|
|
|
|
return next.length === state.turnTrail.length ? state : { ...state, turnTrail: next }
|
|
})
|
|
}
|
|
|
|
flushStreamingSegment() {
|
|
const raw = this.bufRef.trimStart()
|
|
|
|
if (!raw) {
|
|
return
|
|
}
|
|
|
|
const split = hasReasoningTag(raw) ? splitReasoning(raw) : { reasoning: '', text: raw }
|
|
|
|
if (split.reasoning && !this.reasoningText.trim()) {
|
|
this.reasoningText = split.reasoning
|
|
patchTurnState({ reasoning: this.reasoningText, reasoningTokens: estimateTokensRough(this.reasoningText) })
|
|
}
|
|
|
|
const text = split.text
|
|
|
|
this.streamTimer = clear(this.streamTimer)
|
|
|
|
if (text) {
|
|
const tools = this.pendingSegmentTools
|
|
|
|
this.segmentMessages = [...this.segmentMessages, { role: 'assistant', text, ...(tools.length && { tools }) }]
|
|
this.pendingSegmentTools = []
|
|
}
|
|
|
|
this.bufRef = ''
|
|
patchTurnState({ streamPendingTools: [], streamSegments: this.segmentMessages, streaming: '' })
|
|
}
|
|
|
|
pulseReasoningStreaming() {
|
|
this.reasoningStreamingTimer = clear(this.reasoningStreamingTimer)
|
|
patchTurnState({ reasoningActive: true, reasoningStreaming: true })
|
|
|
|
this.reasoningStreamingTimer = setTimeout(() => {
|
|
this.reasoningStreamingTimer = null
|
|
patchTurnState({ reasoningStreaming: false })
|
|
}, REASONING_PULSE_MS)
|
|
}
|
|
|
|
pushInlineDiffSegment(diffText: string) {
|
|
// Strip CLI chrome the gateway emits before the unified diff (e.g. a
|
|
// leading "┊ review diff" header written by `_emit_inline_diff` for the
|
|
// terminal printer). That header only makes sense as stdout dressing,
|
|
// not inside a markdown ```diff block.
|
|
const stripped = diffText.replace(/^\s*┊[^\n]*\n?/, '').trim()
|
|
|
|
if (!stripped) {
|
|
return
|
|
}
|
|
|
|
// Flush any in-progress streaming text as its own segment first, so the
|
|
// diff lands BETWEEN the assistant narration that preceded the edit and
|
|
// whatever the agent streams afterwards — not glued onto the final
|
|
// message. This is the whole point of segment-anchored diffs: the diff
|
|
// renders where the edit actually happened.
|
|
this.flushStreamingSegment()
|
|
|
|
const block = `\`\`\`diff\n${stripped}\n\`\`\``
|
|
|
|
// Skip consecutive duplicates (same tool firing tool.complete twice, or
|
|
// two edits producing the same patch). Keeping this cheap — deeper
|
|
// dedupe against the final assistant text happens at message.complete.
|
|
if (this.segmentMessages.at(-1)?.text === block) {
|
|
return
|
|
}
|
|
|
|
this.segmentMessages = [...this.segmentMessages, { role: 'assistant', text: block }]
|
|
patchTurnState({ streamSegments: this.segmentMessages })
|
|
}
|
|
|
|
pushActivity(text: string, tone: ActivityItem['tone'] = 'info', replaceLabel?: string) {
|
|
patchTurnState(state => {
|
|
const base = replaceLabel
|
|
? state.activity.filter(item => !sameToolTrailGroup(replaceLabel, item.text))
|
|
: state.activity
|
|
|
|
const tail = base.at(-1)
|
|
|
|
if (tail?.text === text && tail.tone === tone) {
|
|
return state
|
|
}
|
|
|
|
return { ...state, activity: [...base, { id: ++this.activityId, text, tone }].slice(-ACTIVITY_LIMIT) }
|
|
})
|
|
}
|
|
|
|
pushTrail(line: string) {
|
|
patchTurnState(state => {
|
|
if (state.turnTrail.at(-1) === line) {
|
|
return state
|
|
}
|
|
|
|
const next = [...state.turnTrail.filter(item => !isTransientTrailLine(item)), line].slice(-TRAIL_LIMIT)
|
|
|
|
this.turnTools = next
|
|
|
|
return { ...state, turnTrail: next }
|
|
})
|
|
}
|
|
|
|
recordError() {
|
|
this.idle()
|
|
this.clearReasoning()
|
|
this.clearStatusTimer()
|
|
this.pendingSegmentTools = []
|
|
this.segmentMessages = []
|
|
this.turnTools = []
|
|
this.persistedToolLabels.clear()
|
|
}
|
|
|
|
recordMessageComplete(payload: { rendered?: string; reasoning?: string; text?: string }) {
|
|
const rawText = (payload.rendered ?? payload.text ?? this.bufRef).trimStart()
|
|
const split = splitReasoning(rawText)
|
|
const finalText = split.text
|
|
const existingReasoning = this.reasoningText.trim() || String(payload.reasoning ?? '').trim()
|
|
const savedReasoning = [existingReasoning, existingReasoning ? '' : split.reasoning].filter(Boolean).join('\n\n')
|
|
const savedReasoningTokens = savedReasoning ? estimateTokensRough(savedReasoning) : 0
|
|
const savedToolTokens = this.toolTokenAcc
|
|
const tools = this.pendingSegmentTools
|
|
|
|
// Drop diff-only segments the agent is about to narrate in the final
|
|
// reply. Without this, a closing "here's the diff …" message would
|
|
// render two stacked copies of the same patch. Only touches segments
|
|
// whose entire body is a ```diff``` fence emitted by pushInlineDiff-
|
|
// Segment — real assistant narration stays put.
|
|
const finalHasOwnDiffFence = /```(?:diff|patch)\b/i.test(finalText)
|
|
|
|
const segments = this.segmentMessages.filter(msg => {
|
|
const body = diffSegmentBody(msg)
|
|
|
|
if (body === null) {
|
|
return true
|
|
}
|
|
|
|
return !finalHasOwnDiffFence && !finalText.includes(body)
|
|
})
|
|
|
|
const finalMessages = [...segments]
|
|
|
|
if (finalText) {
|
|
finalMessages.push({
|
|
role: 'assistant',
|
|
text: finalText,
|
|
thinking: savedReasoning || undefined,
|
|
thinkingTokens: savedReasoning ? savedReasoningTokens : undefined,
|
|
toolTokens: savedToolTokens || undefined,
|
|
...(tools.length && { tools })
|
|
})
|
|
}
|
|
|
|
const wasInterrupted = this.interrupted
|
|
|
|
// Archive the turn's spawn tree to history BEFORE idle() drops subagents
|
|
// from turnState. Lets /replay and the overlay's history nav pull up
|
|
// finished fan-outs without a round-trip to disk.
|
|
const finishedSubagents = getTurnState().subagents
|
|
const sessionId = getUiState().sid
|
|
|
|
if (finishedSubagents.length > 0) {
|
|
pushSnapshot(finishedSubagents, { sessionId, startedAt: null })
|
|
// Fire-and-forget disk persistence so /replay survives process restarts.
|
|
// The same snapshot lives in memory via spawnHistoryStore for immediate
|
|
// recall — disk is the long-term archive.
|
|
void this.persistSpawnTree?.(finishedSubagents, sessionId)
|
|
}
|
|
|
|
this.idle()
|
|
this.clearReasoning()
|
|
this.turnTools = []
|
|
this.persistedToolLabels.clear()
|
|
this.bufRef = ''
|
|
patchTurnState({ activity: [], outcome: '' })
|
|
|
|
return { finalMessages, finalText, wasInterrupted }
|
|
}
|
|
|
|
recordMessageDelta({ rendered, text }: { rendered?: string; text?: string }) {
|
|
this.pruneTransient()
|
|
this.endReasoningPhase()
|
|
|
|
if (!text || this.interrupted) {
|
|
return
|
|
}
|
|
|
|
this.bufRef = rendered ?? this.bufRef + text
|
|
|
|
if (getUiState().streaming) {
|
|
this.scheduleStreaming()
|
|
}
|
|
}
|
|
|
|
recordReasoningAvailable(text: string) {
|
|
if (!getUiState().showReasoning) {
|
|
return
|
|
}
|
|
|
|
const incoming = text.trim()
|
|
|
|
if (!incoming || this.reasoningText.trim()) {
|
|
return
|
|
}
|
|
|
|
this.reasoningText = incoming
|
|
this.scheduleReasoning()
|
|
this.pulseReasoningStreaming()
|
|
}
|
|
|
|
recordReasoningDelta(text: string) {
|
|
if (!getUiState().showReasoning) {
|
|
return
|
|
}
|
|
|
|
this.reasoningText += text
|
|
this.scheduleReasoning()
|
|
this.pulseReasoningStreaming()
|
|
}
|
|
|
|
recordToolComplete(toolId: string, fallbackName?: string, error?: string, summary?: string) {
|
|
const done = this.activeTools.find(tool => tool.id === toolId)
|
|
const name = done?.name ?? fallbackName ?? 'tool'
|
|
const label = toolTrailLabel(name)
|
|
const line = buildToolTrailLine(name, done?.context || '', Boolean(error), error || summary || '')
|
|
|
|
this.activeTools = this.activeTools.filter(tool => tool.id !== toolId)
|
|
this.pendingSegmentTools = [...this.pendingSegmentTools, line]
|
|
|
|
const next = this.turnTools.filter(item => !sameToolTrailGroup(label, item))
|
|
|
|
if (!this.activeTools.length) {
|
|
next.push('analyzing tool output…')
|
|
}
|
|
|
|
this.turnTools = next.slice(-TRAIL_LIMIT)
|
|
patchTurnState({
|
|
streamPendingTools: this.pendingSegmentTools,
|
|
tools: this.activeTools,
|
|
turnTrail: this.turnTools
|
|
})
|
|
}
|
|
|
|
recordToolProgress(toolName: string, preview: string) {
|
|
const index = this.activeTools.findIndex(tool => tool.name === toolName)
|
|
|
|
if (index < 0) {
|
|
return
|
|
}
|
|
|
|
this.activeTools = this.activeTools.map((tool, i) => (i === index ? { ...tool, context: preview } : tool))
|
|
|
|
if (this.toolProgressTimer) {
|
|
return
|
|
}
|
|
|
|
this.toolProgressTimer = setTimeout(() => {
|
|
this.toolProgressTimer = null
|
|
patchTurnState({ tools: [...this.activeTools] })
|
|
}, STREAM_BATCH_MS)
|
|
}
|
|
|
|
recordToolStart(toolId: string, name: string, context: string) {
|
|
this.flushStreamingSegment()
|
|
this.pruneTransient()
|
|
this.endReasoningPhase()
|
|
|
|
const sample = `${name} ${context}`.trim()
|
|
|
|
this.toolTokenAcc += sample ? estimateTokensRough(sample) : 0
|
|
this.activeTools = [...this.activeTools, { context, id: toolId, name, startedAt: Date.now() }]
|
|
|
|
patchTurnState({ toolTokens: this.toolTokenAcc, tools: this.activeTools })
|
|
}
|
|
|
|
reset() {
|
|
this.clearReasoning()
|
|
this.clearStatusTimer()
|
|
this.idle()
|
|
this.bufRef = ''
|
|
this.interrupted = false
|
|
this.lastStatusNote = ''
|
|
this.pendingSegmentTools = []
|
|
this.protocolWarned = false
|
|
this.segmentMessages = []
|
|
this.turnTools = []
|
|
this.toolTokenAcc = 0
|
|
this.persistedToolLabels.clear()
|
|
patchTurnState({ activity: [], outcome: '' })
|
|
}
|
|
|
|
fullReset() {
|
|
this.reset()
|
|
resetTurnState()
|
|
}
|
|
|
|
scheduleReasoning() {
|
|
if (this.reasoningTimer) {
|
|
return
|
|
}
|
|
|
|
this.reasoningTimer = setTimeout(() => {
|
|
this.reasoningTimer = null
|
|
patchTurnState({
|
|
reasoning: this.reasoningText,
|
|
reasoningTokens: estimateTokensRough(this.reasoningText)
|
|
})
|
|
}, STREAM_BATCH_MS)
|
|
}
|
|
|
|
scheduleStreaming() {
|
|
if (this.streamTimer) {
|
|
return
|
|
}
|
|
|
|
this.streamTimer = setTimeout(() => {
|
|
this.streamTimer = null
|
|
const raw = this.bufRef.trimStart()
|
|
const visible = hasReasoningTag(raw) ? splitReasoning(raw).text : raw
|
|
patchTurnState({ streaming: visible })
|
|
}, STREAM_BATCH_MS)
|
|
}
|
|
|
|
startMessage() {
|
|
this.endReasoningPhase()
|
|
this.clearReasoning()
|
|
this.activeTools = []
|
|
this.turnTools = []
|
|
this.toolTokenAcc = 0
|
|
this.persistedToolLabels.clear()
|
|
patchUiState({ busy: true })
|
|
patchTurnState({ activity: [], outcome: '', subagents: [], toolTokens: 0, tools: [], turnTrail: [] })
|
|
}
|
|
|
|
upsertSubagent(
|
|
p: SubagentEventPayload,
|
|
patch: (current: SubagentProgress) => Partial<SubagentProgress>,
|
|
opts: { createIfMissing?: boolean } = { createIfMissing: true }
|
|
) {
|
|
// Stable id: prefer the server-issued subagent_id (survives nested
|
|
// grandchildren + cross-tree joins). Fall back to the composite key
|
|
// for older gateways that omit the field — those produce a flat list.
|
|
const id = p.subagent_id || `sa:${p.task_index}:${p.goal || 'subagent'}`
|
|
|
|
patchTurnState(state => {
|
|
const existing = state.subagents.find(item => item.id === id)
|
|
|
|
// Late events (subagent.complete/tool/progress arriving after message.complete
|
|
// has already fired idle()) would otherwise resurrect a finished
|
|
// subagent into turn.subagents and block the "finished" title on the
|
|
// /agents overlay. When `createIfMissing` is false we drop silently.
|
|
if (!existing && !opts.createIfMissing) {
|
|
return state
|
|
}
|
|
|
|
const base: SubagentProgress = existing ?? {
|
|
depth: p.depth ?? 0,
|
|
goal: p.goal,
|
|
id,
|
|
index: p.task_index,
|
|
model: p.model,
|
|
notes: [],
|
|
parentId: p.parent_id ?? null,
|
|
startedAt: Date.now(),
|
|
status: 'running',
|
|
taskCount: p.task_count ?? 1,
|
|
thinking: [],
|
|
toolCount: p.tool_count ?? 0,
|
|
tools: [],
|
|
toolsets: p.toolsets
|
|
}
|
|
|
|
// Map snake_case payload keys onto camelCase state. Only overwrite
|
|
// when the event actually carries the field; `??` preserves prior
|
|
// values across streaming events that emit partial payloads.
|
|
const outputTail = p.output_tail
|
|
? p.output_tail.map(e => ({
|
|
isError: Boolean(e.is_error),
|
|
preview: String(e.preview ?? ''),
|
|
tool: String(e.tool ?? 'tool')
|
|
}))
|
|
: base.outputTail
|
|
|
|
const next: SubagentProgress = {
|
|
...base,
|
|
apiCalls: p.api_calls ?? base.apiCalls,
|
|
costUsd: p.cost_usd ?? base.costUsd,
|
|
depth: p.depth ?? base.depth,
|
|
filesRead: p.files_read ?? base.filesRead,
|
|
filesWritten: p.files_written ?? base.filesWritten,
|
|
goal: p.goal || base.goal,
|
|
inputTokens: p.input_tokens ?? base.inputTokens,
|
|
iteration: p.iteration ?? base.iteration,
|
|
model: p.model ?? base.model,
|
|
outputTail,
|
|
outputTokens: p.output_tokens ?? base.outputTokens,
|
|
parentId: p.parent_id ?? base.parentId,
|
|
reasoningTokens: p.reasoning_tokens ?? base.reasoningTokens,
|
|
taskCount: p.task_count ?? base.taskCount,
|
|
toolCount: p.tool_count ?? base.toolCount,
|
|
toolsets: p.toolsets ?? base.toolsets,
|
|
...patch(base)
|
|
}
|
|
|
|
// Stable order: by spawn (depth, parent, index) rather than insert time.
|
|
// Without it, grandchildren can shuffle relative to siblings when
|
|
// events arrive out of order under high concurrency.
|
|
const subagents = existing
|
|
? state.subagents.map(item => (item.id === id ? next : item))
|
|
: [...state.subagents, next].sort((a, b) => a.depth - b.depth || a.index - b.index)
|
|
|
|
return { ...state, subagents }
|
|
})
|
|
}
|
|
}
|
|
|
|
export const turnController = new TurnController()
|
|
|
|
export type { TurnController }
|