ethagent 3.1.2 → 3.2.0

package/README.md

@@ -67,10 +67,10 @@ Each agent's continuity directory holds a small set of files. Private files are
 | --- | --- | --- |
 | `SOUL.md` | Private | Soul, boundaries, standing instructions, and identity framing. |
 | `MEMORY.md` | Private | Durable preferences, project context, decisions, and operating notes. |
-| `skills/` | Mixed | Skill folders. Each skill is private, discoverable, or public; new skills default to discoverable. |
+| `skills/` | Private | Skill folders. The SKILL.md body never leaves your machine. The visibility flag only controls whether the skill's name and description get indexed in `skills.json`. New skills default to public. |
 | `skills.json` | Public | Machine-readable capabilities derived from public skills. |
 
-`SOUL.md`, `MEMORY.md`, and each `SKILL.md` are plain Markdown you edit through the Identity Hub under Continuity. Skills carry extra metadata: the frontmatter at the top of each `SKILL.md` (name, description, when_to_use, visibility, tags) tells the agent when to load it. Visibility is `private` (local-only, never shared), `discoverable` (indexed in `skills.json` so other agents can find it), or `public` (indexed and surfaced on the Agent Card).
+`SOUL.md`, `MEMORY.md`, and each `SKILL.md` are plain Markdown you edit through the Identity Hub under Continuity. Skill frontmatter (name, description, when_to_use, visibility, tags) tells the agent when to load it. The body stays local; `visibility: public` indexes the name and description in `skills.json` and the Agent Card.
 
 - **Save Snapshot Now** encrypts the private files, pins them to IPFS, and rotates the onchain pointer to the new CID.
 - **Refetch Latest** reads the pointer back, signs the decrypt challenge with your wallet, and overwrites local files from the snapshot.

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "ethagent",
-  "version": "3.1.2",
+  "version": "3.2.0",
   "description": "A privacy-first AI agent with a portable Ethereum identity",
   "type": "module",
   "main": "bin/ethagent.js",

package/src/chat/ChatScreen.tsx

@@ -44,7 +44,9 @@ import {
   shouldConfirmContextUsage,
   type ContextUsage,
 } from '../runtime/compaction.js'
-import { saveConfig } from '../storage/config.js'
+import { fetchLlamaCppContextSize, onLlamaCppContextSizeChange, setCachedLlamaCppContextSize } from '../models/llamacpp.js'
+import { llamaCppServerHostFromBaseUrl } from '../models/llamacppPreflight.js'
+import { localProviderBaseUrlFor, saveConfig } from '../storage/config.js'
 import { getCwd as getRuntimeCwd, setCwd as setRuntimeCwd, syncCwdFromProcess } from '../runtime/cwd.js'
 import { executeToolWithPermissions } from '../runtime/toolExecution.js'
 import { nextSessionMode, sessionModeLabel, type PermissionMode, type SessionMode } from '../runtime/sessionMode.js'
@@ -309,7 +311,7 @@ export const ChatScreen: React.FC<ChatScreenProps> = ({ config: initialConfig, o
       {
         role: 'progress',
         id: progressRowId,
-        title: kind === 'plan' ? 'summarizing plan context' : 'compacting conversation',
+        title: kind === 'plan' ? 'summarizing plan context' : 'Compacting conversation',
         progress: 0,
         status: state.stage,
         suffix: 'esc to cancel',
@@ -449,6 +451,22 @@ export const ChatScreen: React.FC<ChatScreenProps> = ({ config: initialConfig, o
     [],
   )
 
+  useEffect(() => {
+    if (config.provider !== 'llamacpp') return
+    const host = llamaCppServerHostFromBaseUrl(localProviderBaseUrlFor('llamacpp', config.baseUrl))
+    void fetchLlamaCppContextSize(host)
+    const unsubscribe = onLlamaCppContextSizeChange(() => {
+      refreshVisibleStats(
+        sessionMessagesRef.current,
+        providerRef.current.supportsTools,
+        cwdRef.current,
+        configRef.current,
+        modeRef.current,
+      )
+    })
+    return unsubscribe
+  }, [config.provider, config.baseUrl, refreshVisibleStats])
+
   const warnIfContextPressure = useCallback(
     (usage: ContextUsage, configForUsage: EthagentConfig) => {
       if (!shouldConfirmContextUsage(usage, CONTEXT_CONFIRM_PERCENT)) return
@@ -880,6 +898,11 @@ export const ChatScreen: React.FC<ChatScreenProps> = ({ config: initialConfig, o
           pendingPlanRef.current = planCandidate
           setPendingPlan(planCandidate)
         },
+        onContextExceeded: ({ contextLimit }) => {
+          setCachedLlamaCppContextSize(contextLimit)
+          pushNote('Context full. Compacting transcript. Re-send your message once compaction finishes.', 'dim')
+          void runCompaction()
+        },
         pendingAssistantTextRef,
         pendingThinkingTextRef,
         streamFlushTimerRef,

package/src/chat/chatTurnOrchestrator.ts

@@ -56,6 +56,7 @@ export type TurnOrchestratorContext = {
   applySessionRule: (rule?: SessionPermissionRule, persistRule?: boolean) => Promise<void>
   preflightProvider?: () => Promise<{ ok: true } | { ok: false; message: string }>
   onPlanReady?: (plan: string) => void
+  onContextExceeded?: (info: { contextLimit: number }) => void
   pendingAssistantTextRef: MutableRef<string | null>
   pendingThinkingTextRef: MutableRef<string | null>
   streamFlushTimerRef: MutableRef<ReturnType<typeof setTimeout> | null>
@@ -89,6 +90,7 @@ export async function runStreamingTurn(
     applySessionRule,
     preflightProvider,
     onPlanReady,
+    onContextExceeded,
     pendingAssistantTextRef,
     pendingThinkingTextRef,
     streamFlushTimerRef,
@@ -311,6 +313,7 @@ export async function runStreamingTurn(
         nowIso,
         mode,
         onPlanReady,
+        onContextExceeded,
         turnId: activeCheckpoint.turnId,
         model: getConfig().model,
         onFinishedNormally: () => { finishedNormally = true },
@@ -356,6 +359,7 @@ type EventHandlerContext = {
   nowIso: () => string
   mode: SessionMode
   onPlanReady?: (plan: string) => void
+  onContextExceeded?: (info: { contextLimit: number }) => void
   turnId: string
   model: string
   onFinishedNormally: () => void
@@ -365,6 +369,13 @@ function isCancelledEvent(ev: TurnEvent): boolean {
   return ev.type === 'cancelled'
 }
 
+export function parseContextExceededLimit(message: string): number | null {
+  const match = /exceeds the available context size \((\d+)\s*tokens?\)/i.exec(message)
+  if (!match) return null
+  const limit = Number.parseInt(match[1]!, 10)
+  return Number.isFinite(limit) && limit > 0 ? limit : null
+}
+
 async function handleEvent(ev: TurnEvent, ctx: EventHandlerContext): Promise<void> {
   switch (ev.type) {
     case 'iteration_start': {
@@ -453,6 +464,12 @@ async function handleEvent(ev: TurnEvent, ctx: EventHandlerContext): Promise<voi
       return
     }
     case 'error': {
+      const contextLimit = parseContextExceededLimit(ev.message)
+      if (contextLimit !== null && ctx.onContextExceeded) {
+        ctx.discardStreamingRows()
+        ctx.onContextExceeded({ contextLimit })
+        return
+      }
       ctx.pushNote(ev.message, 'error')
       if (ev.discardAssistant) {
         ctx.discardStreamingRows()

package/src/identity/continuity/publicSkills.ts

@@ -103,7 +103,7 @@ export function appendPublicSkillEntries(
   const appended: PublicSkill[] = []
   const usedIds = new Set(baselineIds)
   for (const entry of entries) {
-    if (entry.visibility !== 'public' && entry.visibility !== 'discoverable') continue
+    if (entry.visibility !== 'public') continue
     const id = uniqueSkillId(entry.name, usedIds)
     usedIds.add(id)
     appended.push({

package/src/identity/continuity/skills/frontmatter.ts

@@ -14,7 +14,8 @@ const SUPPORTED_KEYS = new Set([
   'visibility',
 ])
 
-const VISIBILITY_VALUES: SkillVisibility[] = ['private', 'public', 'discoverable']
+const VISIBILITY_VALUES: SkillVisibility[] = ['private', 'public']
+const LEGACY_VISIBILITY_TO_PRIVATE = new Set(['discoverable'])
 
 export type ParsedSkillFile = {
   frontmatter: SkillFrontmatter
@@ -104,6 +105,8 @@ function assignKey(out: SkillFrontmatter, key: keyof SkillFrontmatter, rawValue:
     const literal = parseScalar(stripped).toLowerCase()
     if ((VISIBILITY_VALUES as string[]).includes(literal)) {
       out.visibility = literal as SkillVisibility
+    } else if (LEGACY_VISIBILITY_TO_PRIVATE.has(literal)) {
+      out.visibility = 'private'
     }
     return
   }

package/src/identity/continuity/skills/loadSkills.ts

@@ -215,8 +215,11 @@ export async function loadSkillsTree(identity: EthagentIdentity): Promise<Contin
       const rel = `${skillEnt.name}/${file.relativePath}`
       if (!isValidSkillFilePath(rel)) continue
       if (file.sizeBytes > MAX_SKILL_FILE_BYTES) continue
-      const content = await fs.readFile(file.absolutePath, 'utf8').catch(() => null)
-      if (content === null) continue
+      const rawContent = await fs.readFile(file.absolutePath, 'utf8').catch(() => null)
+      if (rawContent === null) continue
+      const content = file.relativePath === SKILL_FILE_NAME
+        ? await ensureSkillVisibilityWritten(file.absolutePath, rawContent)
+        : rawContent
       tree[rel] = content
       totalFiles++
     }
@@ -359,7 +362,12 @@ export async function migrateLegacySkillFiles(skillsRoot: string): Promise<void>
     return
   }
   for (const topEnt of topDirents) {
-    if (!topEnt.isDirectory() || topEnt.isSymbolicLink()) continue
+    if (topEnt.isSymbolicLink()) continue
+    if (topEnt.isFile() && /\.md$/i.test(topEnt.name)) {
+      await adoptBareSkillFile(skillsRoot, topEnt.name)
+      continue
+    }
+    if (!topEnt.isDirectory()) continue
     if (!isValidSegment(topEnt.name)) continue
     const topDir = path.join(skillsRoot, topEnt.name)
     let children: import('node:fs').Dirent[]
@@ -403,6 +411,39 @@ export async function migrateLegacySkillFiles(skillsRoot: string): Promise<void>
   }
 }
 
+async function adoptBareSkillFile(skillsRoot: string, fileName: string): Promise<void> {
+  const sourcePath = path.join(skillsRoot, fileName)
+  let baseName: string
+  if (/^SKILL\.md$/i.test(fileName)) {
+    let parsedName: string | undefined
+    try {
+      const raw = await fs.readFile(sourcePath, 'utf8')
+      const parsed = parseSkillFile(raw)
+      const fmName = parsed.frontmatter.name?.trim()
+      if (fmName && isValidSegment(fmName)) parsedName = fmName
+    } catch {
+    }
+    baseName = parsedName ?? 'imported-skill'
+  } else {
+    const slug = fileName.replace(/\.md$/i, '')
+    if (!isValidSegment(slug)) return
+    baseName = slug
+  }
+  let target: string
+  try {
+    target = await chooseFlatTarget(skillsRoot, baseName)
+  } catch {
+    return
+  }
+  const targetDir = path.join(skillsRoot, target)
+  const targetFile = path.join(targetDir, SKILL_FILE_NAME)
+  try {
+    await fs.mkdir(targetDir, { recursive: true, mode: 0o700 })
+    await fs.rename(sourcePath, targetFile)
+  } catch {
+  }
+}
+
 async function chooseFlatTarget(skillsRoot: string, base: string): Promise<string> {
   let candidate = base
   let suffix = 2
@@ -476,6 +517,32 @@ async function pathExists(file: string): Promise<boolean> {
   }
 }
 
+const DEFAULT_PASTED_VISIBILITY: SkillVisibility = 'public'
+const LEGACY_DISCOVERABLE_RE = /^\s*visibility\s*:\s*['"]?discoverable['"]?\s*$/im
+
+async function ensureSkillVisibilityWritten(skillFile: string, raw: string): Promise<string> {
+  let parsed: { frontmatter: import('./types.js').SkillFrontmatter; body: string }
+  try {
+    parsed = parseSkillFile(raw)
+  } catch {
+    return raw
+  }
+  let target: SkillVisibility | null = null
+  if (LEGACY_DISCOVERABLE_RE.test(raw)) {
+    target = 'private'
+  } else if (parsed.frontmatter.visibility === undefined) {
+    target = DEFAULT_PASTED_VISIBILITY
+  }
+  if (target === null) return raw
+  const next = rewriteVisibility(raw, target)
+  if (next === raw) return raw
+  try {
+    await atomicWriteText(skillFile, next, { mode: 0o600 })
+  } catch {
+  }
+  return next
+}
+
 async function collectSkillEntries(root: string): Promise<SkillIndexEntry[]> {
   const out: SkillIndexEntry[] = []
   let topDirents: import('node:fs').Dirent[]
@@ -493,7 +560,8 @@ async function collectSkillEntries(root: string): Promise<SkillIndexEntry[]> {
       const stat = await fs.stat(skillFile)
       if (!stat.isFile()) continue
       if (stat.size > MAX_SKILL_FILE_BYTES) continue
-      const raw = await fs.readFile(skillFile, 'utf8')
+      const rawInitial = await fs.readFile(skillFile, 'utf8')
+      const raw = await ensureSkillVisibilityWritten(skillFile, rawInitial)
       const parsed = parseSkillFile(raw)
       const relativePath = `${skillEnt.name}/${SKILL_FILE_NAME}`
       out.push(buildIndexEntry({
@@ -519,7 +587,7 @@ function buildIndexEntry(args: {
   const derivedName = folder || segments.join('/')
   const fm = args.parsed.frontmatter
   const description = pickDescription(fm.description, args.parsed.body)
-  const visibility: SkillVisibility = fm.visibility ?? 'discoverable'
+  const visibility: SkillVisibility = fm.visibility ?? DEFAULT_PASTED_VISIBILITY
   return {
     name: derivedName,
     ...(fm.name ? { displayName: fm.name } : {}),

package/src/identity/continuity/skills/publicSkillsSync.ts

@@ -11,7 +11,7 @@ import type { SkillIndexEntry } from './types.js'
 
 export async function derivePublicSkillEntries(identity: EthagentIdentity): Promise<SkillIndexEntry[]> {
   const entries = await listSkills(identity)
-  return entries.filter(entry => entry.visibility === 'public' || entry.visibility === 'discoverable')
+  return entries.filter(entry => entry.visibility === 'public')
 }
 
 export async function renderPublicSkillsJsonForIdentity(identity: EthagentIdentity): Promise<string> {

package/src/identity/continuity/skills/scaffold.ts

@@ -5,7 +5,7 @@ export type SkillScaffoldArgs = {
   visibility?: SkillVisibility
 }
 
-export function defaultSkillScaffold({ name, visibility = 'discoverable' }: SkillScaffoldArgs): string {
+export function defaultSkillScaffold({ name, visibility = 'public' }: SkillScaffoldArgs): string {
   return [
     '---',
     `name: ${name}`,

package/src/identity/continuity/skills/types.ts

@@ -1,4 +1,4 @@
-export type SkillVisibility = 'private' | 'public' | 'discoverable'
+export type SkillVisibility = 'private' | 'public'
 
 export type SkillFrontmatter = {
   name?: string

package/src/identity/hub/continuity/skills/NewSkillVisibilityScreen.tsx

@@ -22,7 +22,7 @@ export const NewSkillVisibilityScreen: React.FC<NewSkillVisibilityScreenProps> =
 }) => (
   <Surface
     title={`Visibility · ${name}`}
-    subtitle="Discoverable is the default. You can change it later from Change Visibility."
+    subtitle="Public is the default. You can change it later from Change Visibility."
     footer={footer}
   >
     {error && (
@@ -34,8 +34,7 @@ export const NewSkillVisibilityScreen: React.FC<NewSkillVisibilityScreenProps> =
       <Select<SkillVisibility | 'back'>
         options={[
           { value: 'private', label: 'Private', hint: 'Local-only. Not in skills.json.' },
-          { value: 'discoverable', label: 'Discoverable', hint: 'Default. Indexed in skills.json with description.' },
-          { value: 'public', label: 'Public', hint: 'Indexed in skills.json and Agent Card.' },
+          { value: 'public', label: 'Public', hint: 'Default. Indexed in skills.json and Agent Card.' },
           { value: 'back', role: 'section', label: 'Navigation' },
           { value: 'back', label: 'Back', hint: 'Return to the name step', role: 'utility' },
         ]}

package/src/identity/hub/continuity/skills/SkillActionsScreen.tsx

@@ -74,7 +74,6 @@ export const SkillActionsScreen: React.FC<SkillActionsScreenProps> = ({
 
   options.push({ value: noop, role: 'section', label: 'Visibility' })
   options.push(visibilityOption('private', visibility))
-  options.push(visibilityOption('discoverable', visibility))
   options.push(visibilityOption('public', visibility))
 
   options.push({ value: noop, role: 'section', label: 'Manage' })
@@ -83,6 +82,7 @@ export const SkillActionsScreen: React.FC<SkillActionsScreenProps> = ({
     label: 'Delete',
     hint: 'Remove this skill folder and its supporting files',
   })
+  options.push({ value: noop, role: 'section', label: 'Return' })
   options.push({
     value: { kind: 'back' },
     label: 'Back',
@@ -141,8 +141,7 @@ function visibilityOption(level: SkillVisibility, current?: SkillVisibility): Se
 
 function visibilityHint(level: SkillVisibility): string {
   if (level === 'private') return 'Local-only. Not in skills.json.'
-  if (level === 'discoverable') return 'Default. Indexed with description.'
-  return 'Indexed with description and Agent Card link.'
+  return 'Default. Indexed with description and Agent Card link.'
 }
 
 function capitalize(value: string): string {

package/src/identity/hub/continuity/skills/SkillsTreeScreen.tsx

@@ -73,7 +73,7 @@ export const SkillsTreeScreen: React.FC<SkillsTreeScreenProps> = ({
     }
   }, [identity, editorOpened])
 
-  const subtitle = notice ?? 'Open a skill, create one, or remove one.'
+  const subtitle = notice ?? 'Select a skill to open, change visibility, or delete.'
   const isLoading = tree === null
   const skills = tree?.skills ?? []
   const supportingCounts = tree?.supportingCounts ?? {}
@@ -145,6 +145,7 @@ function buildOptions(
       const supportCount = supportingCounts[skill.name] ?? 0
       const meta = [capitalize(skill.visibility)]
       if (supportCount > 0) meta.push(`${supportCount + 1} files`)
+      meta.push('enter for actions')
       rows.push({
         value: { kind: 'skill', relativePath: skill.relativePath },
         label: `${branch}${skill.name}/SKILL.md`,

package/src/models/llamacpp.ts

@@ -256,12 +256,68 @@ async function fetchServedModels(host: string = DEFAULT_LLAMA_HOST, timeoutMs =
   }
 }
 
+let cachedLlamaCppContextSize: number | null = null
+const llamaCppContextSizeListeners = new Set<(size: number) => void>()
+
+export async function fetchLlamaCppContextSize(
+  host: string = DEFAULT_LLAMA_HOST,
+  timeoutMs = 1500,
+): Promise<number | null> {
+  const response = await fetchWithTimeout(`${host.replace(/\/+$/, '')}/props`, timeoutMs)
+  if (!response || !response.ok) return null
+  try {
+    const data = await response.json() as {
+      n_ctx?: unknown
+      default_generation_settings?: { n_ctx?: unknown }
+    }
+    const raw = typeof data.n_ctx === 'number'
+      ? data.n_ctx
+      : typeof data.default_generation_settings?.n_ctx === 'number'
+        ? data.default_generation_settings.n_ctx
+        : null
+    if (typeof raw === 'number' && raw > 0) {
+      const changed = cachedLlamaCppContextSize !== raw
+      cachedLlamaCppContextSize = raw
+      if (changed) {
+        for (const listener of llamaCppContextSizeListeners) {
+          try { listener(raw) } catch { void 0 }
+        }
+      }
+      return raw
+    }
+    return null
+  } catch {
+    return null
+  }
+}
+
+export function getCachedLlamaCppContextSize(): number | null {
+  return cachedLlamaCppContextSize
+}
+
+export function setCachedLlamaCppContextSize(size: number): void {
+  if (!(size > 0)) return
+  const changed = cachedLlamaCppContextSize !== size
+  cachedLlamaCppContextSize = size
+  if (changed) {
+    for (const listener of llamaCppContextSizeListeners) {
+      try { listener(size) } catch { void 0 }
+    }
+  }
+}
+
+export function onLlamaCppContextSizeChange(listener: (size: number) => void): () => void {
+  llamaCppContextSizeListeners.add(listener)
+  return () => { llamaCppContextSizeListeners.delete(listener) }
+}
+
 export async function detectLlamaCpp(host: string = DEFAULT_LLAMA_HOST): Promise<LlamaCppStatus> {
   const [binary, serverUp] = await Promise.all([
     detectLlamaCppServerBinary(),
     isLlamaCppServerUp(host),
   ])
   const servedModels = serverUp ? await listServedModels(host) : []
+  if (serverUp) void fetchLlamaCppContextSize(host)
   return {
     binaryPresent: binary.path !== null,
     binaryPath: binary.path,
@@ -298,6 +354,7 @@ export async function startLlamaCppServer(args: {
     }
   }
   if (initialStatus.state === 'ready') {
+    void fetchLlamaCppContextSize(host)
     return { ok: true, alreadyRunning: true }
   }
   if (initialStatus.state === 'different') {
@@ -377,7 +434,10 @@ export async function startLlamaCppServer(args: {
     pollMs: args.pollMs ?? 500,
     childFailure: () => childFailure,
   })
-  if (ready.ok) return { ok: true, alreadyRunning: false }
+  if (ready.ok) {
+    void fetchLlamaCppContextSize(host)
+    return { ok: true, alreadyRunning: false }
+  }
   if (ready.code === 'readiness-timeout') {
     return startFailure('readiness-timeout', { detail: capture() })
   }

package/src/models/llamacppPreflight.ts

@@ -1,4 +1,5 @@
 import {
+  fetchLlamaCppContextSize,
   startLlamaCppServer,
   stopLlamaCppServer,
   type LlamaCppStartFailureCode,
@@ -64,7 +65,10 @@ export async function ensureLlamaCppRunnerReady(
         servedModels: probe.models,
       }
     }
-    if (!local.mmprojPath) return { ok: true, alreadyRunning: true }
+    if (!local.mmprojPath) {
+      void fetchLlamaCppContextSize(llamaCppServerHostFromBaseUrl(baseUrl))
+      return { ok: true, alreadyRunning: true }
+    }
     await (deps.stopServer ?? stopLlamaCppServer)().catch(() => null)
   }
 

package/src/runtime/compaction.ts

@@ -1,6 +1,7 @@
 import type { Message, Provider } from '../providers/contracts.js'
 import { approximateTokens, messageTextContent } from '../utils/messages.js'
 import type { SessionMessage } from '../storage/sessions.js'
+import { getCachedLlamaCppContextSize } from '../models/llamacpp.js'
 
 const COMPACT_SYSTEM = `Create a continuation handoff for this coding-agent conversation.
 Keep it concise but complete. Preserve the current goal, user constraints, key decisions, relevant files, tool results, pending tasks, and known failures. Do not claim unverified work was completed. No preamble.`
@@ -17,8 +18,7 @@ const CLOUD_MESSAGE_CHAR_LIMIT = 2_000
 export type CompactionStage =
   | 'preparing transcript'
   | 'compressing long context'
-  | 'summarizing with local model'
-  | 'summarizing with provider'
+  | 'summarizing transcript'
 
 export type CompactTranscriptOptions = {
   signal?: AbortSignal
@@ -60,6 +60,12 @@ export function contextWindow(model: string): number {
 export function contextWindowInfo(provider: string, model: string): ContextWindowInfo {
   const lower = model.toLowerCase()
   const providerLower = provider.toLowerCase()
+  if (providerLower === 'llamacpp') {
+    const cached = getCachedLlamaCppContextSize()
+    if (cached) {
+      return { tokens: cached, confidence: 'exact', source: 'llama.cpp /props' }
+    }
+  }
   if (lower.startsWith('qwen3:4b') || lower.startsWith('qwen3:30b') || lower.startsWith('qwen3:235b')) {
     return { tokens: 256_000, confidence: 'inferred', source: 'qwen3 long-context tag' }
   }
@@ -138,7 +144,7 @@ export async function compactTranscript(
   const signal = options.signal ?? controller!.signal
   let summary = ''
   const local = isLocalProviderId(provider.id)
-  options.onStage?.(local ? 'summarizing with local model' : 'summarizing with provider')
+  options.onStage?.('summarizing transcript')
   try {
     for await (const ev of provider.complete(prompt, signal, {
       maxTokens: options.maxOutputTokens ?? (local ? LOCAL_COMPACTION_OUTPUT_TOKENS : CLOUD_COMPACTION_OUTPUT_TOKENS),
@@ -168,7 +174,7 @@ export function buildCompactionSource(
   const nonSystem = transcript.filter(m => m.role !== 'system')
   const local = isLocalProviderId(providerId)
   const tokenBudget = options.maxInputTokens ?? (local ? LOCAL_COMPACTION_INPUT_TOKENS : CLOUD_COMPACTION_INPUT_TOKENS)
-  const charBudget = Math.max(1_000, tokenBudget * 4)
+  const charBudget = Math.max(1_000, tokenBudget * 3)
   const recentMessageCount = local ? LOCAL_RECENT_MESSAGE_COUNT : CLOUD_RECENT_MESSAGE_COUNT
   const messageCharLimit = local ? LOCAL_MESSAGE_CHAR_LIMIT : CLOUD_MESSAGE_CHAR_LIMIT
   const rawTokenEstimate = approximateTokens(nonSystem)
@@ -385,5 +391,5 @@ function limitCompactionText(text: string, charBudget: number): string {
 }
 
 function approximateTextTokens(text: string): number {
-  return Math.ceil(text.length / 4)
+  return Math.ceil(text.length / 3)
 }

package/src/utils/messages.ts

@@ -33,5 +33,5 @@ export function blocksToText(blocks: MessageContentBlock[]): string {
 export function approximateTokens(messages: Message[]): number {
   let chars = 0
   for (const m of messages) chars += messageTextContent(m).length
-  return Math.ceil(chars / 4)
+  return Math.ceil(chars / 3)
 }