refactor(agents): one pane-identity resolver behind six thin adapters (tranche 0) (#18243)

* feat(agents): pane-identity canonical adapter, comparison telemetry, inventory ratchet phase 1

* fix(agents): preserve canonical coverage provenance

* refactor(agents): unify pane identity adapters for tranche 0

* fix(agents): keep title resolver cache-free after rebase

* Fix ladder tranche zero review findings

* fix(agents): restore title classifier memoization

* fix(agents): fence unknown canonical evidence sources

* docs: drop the ladder plan and decision table from the PR

Design docs stay out of the shipped tree; the code carries its own comments
and the decision table lives in the test fixture.

---------

Co-authored-by: Merge Sim <sim@local>
This commit is contained in:
Brennan Benson
2026-09-03 18:07:56 -07:00
committed by GitHub
co-authored by Merge Sim
parent 8463dcb7b9
commit 90780acb85
13 changed files with 1520 additions and 84 deletions
@@ -0,0 +1,198 @@
import { writeFileSync } from 'node:fs'
import { tmpdir } from 'node:os'
import { join } from 'node:path'
import { describe, expect, it } from 'vitest'
import {
resolveCanonicalPaneAgentIdentity,
type CanonicalPaneAgentIdentity
} from '../../../shared/pane-agent-identity-adapter'
import { resolveTabAgentFromSignals } from './tab-agent-from-signals'
import type { TuiAgent } from '../../../shared/tui-agent'
const AGENTS: readonly TuiAgent[] = ['claude', 'codex']
const SLOT_COUNT = 7
const SHAPE_COUNT = 3 ** SLOT_COUNT * 4 * 2
const TITLES: readonly string[] = ['', 'zsh', 'Task - claude', 'Task - codex']
type Breakdown = Record<
'launch' | 'completed-hook' | 'sleeping-session' | 'process' | 'sibling' | 'title',
number
>
function slotValues(mask: number): (TuiAgent | null)[] {
let remaining = mask
return Array.from({ length: SLOT_COUNT }, () => {
const value = remaining % 3
remaining = Math.floor(remaining / 3)
return value === 0 ? null : AGENTS[value - 1]
})
}
function canonicalResult(
values: readonly (TuiAgent | null)[],
title: string,
withProof: boolean
): CanonicalPaneAgentIdentity {
const [hook, siblingHook, completed, siblingCompleted, process, sleeping, launch] = values
return resolveCanonicalPaneAgentIdentity({
hookAgent: hook,
hookIsLive: hook !== null,
completedHookAgent: completed,
launchAgent: launch,
foregroundAgent: process,
processProof:
withProof && process
? {
agent: process,
processIncarnation: 'fixture-process',
authorityId: 'fixture-authority',
capturedAgeMs: 10,
validForMs: 1_000
}
: undefined,
sleepingSessionAgent: sleeping,
siblingAgents: [siblingHook, siblingCompleted].filter(
(agent): agent is TuiAgent => agent !== null
),
allowSibling: true,
title
})
}
function realResult(values: readonly (TuiAgent | null)[], title: string, remote: boolean) {
const [hook, siblingHook, completed, siblingCompleted, process, sleeping, launch] = values
// The seven slots model steady-state observations; this runtime memory bit is intentionally
// held true instead of adding an eighth dimension to the approved 17,496-shape table.
return resolveTabAgentFromSignals({
hasObservedAgentSignal: true,
isRemote: remote,
title,
hookAgent: hook,
siblingHookAgent: siblingHook,
focusedCompletedHookAgent: completed,
siblingCompletedHookAgent: siblingCompleted,
processAgent: process,
processShellForeground: false,
sleepingSessionAgent: sleeping,
launchAgent: launch ?? undefined
})
}
function runDecisionTable(withProof: boolean) {
let disagreements = 0
let flipped = 0
const breakdown: Breakdown = {
launch: 0,
'completed-hook': 0,
'sleeping-session': 0,
process: 0,
sibling: 0,
title: 0
}
for (let mask = 0; mask < 3 ** SLOT_COUNT; mask += 1) {
const values = slotValues(mask)
for (const title of TITLES) {
for (const remote of [false, true]) {
const real = realResult(values, title, remote)
const canonical = canonicalResult(values, title, withProof)
if (real !== canonical.agent) {
disagreements += 1
if (canonical.source !== null) {
breakdown[canonical.source] += 1
}
}
if (!withProof) {
const proven = canonicalResult(values, title, true)
if (
canonical.agent !== proven.agent &&
proven.source === 'process' &&
(canonical.source === 'launch' ||
canonical.source === 'completed-hook' ||
canonical.source === 'sleeping-session')
) {
flipped += 1
}
}
}
}
}
return { disagreements, flipped, breakdown }
}
describe('renderer ladder decision table', () => {
it('replays the real shipping ladder and records all rung disagreements', () => {
const proofFree = runDecisionTable(false)
const freshProof = runDecisionTable(true)
const result = {
shapes: SHAPE_COUNT,
proofOmitted: proofFree,
freshProof,
flippedByAddingProof: proofFree.flipped
}
writeFileSync(
join(tmpdir(), 'orca-pane-agent-identity-decision-table-real.json'),
`${JSON.stringify(result, null, 2)}\n`
)
// Re-derived against resolveTabAgentFromSignals (not a hand-written model). These differ from
// the approved 2,520/648 totals and 396/144/72/36 breakdown; see the PR comment.
expect(proofFree).toEqual({
disagreements: 2_622,
flipped: 1_872,
breakdown: {
launch: 1_884,
'completed-hook': 478,
'sleeping-session': 144,
process: 0,
sibling: 54,
title: 6
}
})
expect(freshProof).toEqual({
disagreements: 658,
flipped: 0,
breakdown: {
launch: 588,
'completed-hook': 46,
'sleeping-session': 0,
process: 0,
sibling: 6,
title: 2
}
})
expect(proofFree.flipped).toBe(1_872)
})
it('requires both freshness fields before process evidence can change the no-proof result', () => {
const values = [null, null, null, null, 'codex', null, 'claude'] as const
expect(canonicalResult(values, '', false)).toMatchObject({
agent: 'claude',
source: 'launch'
})
expect(
resolveCanonicalPaneAgentIdentity({
foregroundAgent: 'codex',
processProof: {
agent: 'codex',
processIncarnation: 'fixture-process',
authorityId: 'fixture-authority',
capturedAgeMs: undefined as unknown as number,
validForMs: 1_000
},
launchAgent: 'claude'
})
).toMatchObject({ agent: 'claude', source: 'launch' })
expect(
resolveCanonicalPaneAgentIdentity({
foregroundAgent: 'codex',
processProof: {
agent: 'codex',
processIncarnation: 'fixture-process',
authorityId: 'fixture-authority',
capturedAgeMs: 10,
validForMs: undefined as unknown as number
},
launchAgent: 'claude'
})
).toMatchObject({ agent: 'claude', source: 'launch' })
})
})
+8 -1
View File
@@ -4,6 +4,8 @@ import {
type AgentStatusState,
type AgentType
} from './agent-status-types'
import { resolveCanonicalPaneAgentIdentity } from './pane-agent-identity-adapter'
import type { TuiAgent } from './tui-agent'
type ExistingAgentIdentity = {
agentType?: AgentType
@@ -63,6 +65,11 @@ export function resolveAgentStatusIdentity(args: {
inheritedFromActivePane: false
}
}
const canonical = resolveCanonicalPaneAgentIdentity({
hookAgent: incomingAgentType as TuiAgent,
hookIsLive: true,
completedHookAgent: args.existing.state === 'done' ? (existingAgentType as TuiAgent) : undefined
})
if (isActiveExistingIdentity(args.existing, args.now, staleAfterMs)) {
return {
// Why: child agent CLIs inherit ORCA_PANE_KEY from their parent terminal.
@@ -74,7 +81,7 @@ export function resolveAgentStatusIdentity(args: {
}
return {
agentType: incomingAgentType,
agentType: canonical.agent ?? incomingAgentType,
inheritedFromActivePane: false
}
}
+19
View File
@@ -0,0 +1,19 @@
/** Evidence classes in canonical strength order, strongest first. */
export const PANE_AGENT_EVIDENCE_SOURCES = [
/** A live provider hook for a turn in progress. The agent is running and said so. */
'live-hook',
/** The pane's foreground process, as read on the execution host. */
'process',
/** Orca launched, resumed, or accepted a command for this agent. A fact Orca owns. */
'launch',
/** A provider hook from a turn that finished. Still authoritative about identity. */
'completed-hook',
/** A sleeping session record restored for this pane. */
'sleeping-session',
/** Another pane in the same tab. Tab-level surfaces only; never pane-scoped routing. */
'sibling',
/** Parsed from the terminal title. A decoration channel; anyone can type an agent's name. */
'title'
] as const
export type PaneAgentEvidenceSource = (typeof PANE_AGENT_EVIDENCE_SOURCES)[number]
@@ -0,0 +1,278 @@
import { describe, expect, it } from 'vitest'
import {
buildPaneAgentIdentityEvidenceWire,
isForegroundProcessProofFresh,
resolveCanonicalPaneAgentIdentity,
type ForegroundProcessProof
} from './pane-agent-identity-adapter'
const freshProof: ForegroundProcessProof = {
agent: 'codex',
processIncarnation: 'opaque-pid-token',
authorityId: 'main:test',
capturedAgeMs: 50,
validForMs: 5_000
}
describe('per-pane coverage gate', () => {
it('covers a pane from hook, launch, or sleeping-session evidence alone', () => {
expect(
resolveCanonicalPaneAgentIdentity({ hookAgent: 'claude', hookIsLive: true }).coverage
).toBe('covered')
expect(resolveCanonicalPaneAgentIdentity({ completedHookAgent: 'claude' }).coverage).toBe(
'covered'
)
expect(resolveCanonicalPaneAgentIdentity({ launchAgent: 'codex' }).coverage).toBe('covered')
expect(resolveCanonicalPaneAgentIdentity({ sleepingSessionAgent: 'gemini' }).coverage).toBe(
'covered'
)
})
it('never covers a pane from a title, a sibling, or a bare foreground name', () => {
expect(resolveCanonicalPaneAgentIdentity({ title: 'claude' }).coverage).toBe('uncovered')
expect(
resolveCanonicalPaneAgentIdentity({ siblingAgent: 'claude', allowSibling: true }).coverage
).toBe('uncovered')
expect(resolveCanonicalPaneAgentIdentity({ foregroundAgent: 'codex' }).coverage).toBe(
'uncovered'
)
})
it('is computed from evidence, never from a platform or remote flag', () => {
// The input deliberately has no platform/isRemote field to branch on; this pins that a
// hook-covered pane resolves identically regardless of any caller-side host knowledge.
const identity = resolveCanonicalPaneAgentIdentity({ hookAgent: 'claude', hookIsLive: true })
expect(identity).toMatchObject({ agent: 'claude', source: 'live-hook', coverage: 'covered' })
})
})
describe('process rung requires a host-stamped proof', () => {
it('rejects a stale or malformed proof and accepts a fresh one', () => {
expect(isForegroundProcessProofFresh(freshProof)).toBe(true)
expect(isForegroundProcessProofFresh({ ...freshProof, capturedAgeMs: 6_000 })).toBe(false)
expect(isForegroundProcessProofFresh({ ...freshProof, capturedAgeMs: -1 })).toBe(false)
expect(isForegroundProcessProofFresh({ ...freshProof, validForMs: 0 })).toBe(false)
expect(isForegroundProcessProofFresh({ ...freshProof, capturedAgeMs: Number.NaN })).toBe(false)
})
it('a bare foreground name cannot outrank launch; a proven process can', () => {
const unproven = resolveCanonicalPaneAgentIdentity({
launchAgent: 'claude',
foregroundAgent: 'codex'
})
expect(unproven).toMatchObject({ agent: 'claude', source: 'launch' })
const proven = resolveCanonicalPaneAgentIdentity({
launchAgent: 'claude',
foregroundAgent: 'codex',
processProof: freshProof
})
expect(proven).toMatchObject({ agent: 'codex', source: 'process', coverage: 'covered' })
})
it('an expired proof and a name-mismatched proof both drop the process rung', () => {
const expired = resolveCanonicalPaneAgentIdentity({
launchAgent: 'claude',
foregroundAgent: 'codex',
processProof: { ...freshProof, capturedAgeMs: 10_000 }
})
expect(expired).toMatchObject({ agent: 'claude', source: 'launch' })
const mismatched = resolveCanonicalPaneAgentIdentity({
launchAgent: 'claude',
foregroundAgent: 'gemini',
processProof: freshProof
})
expect(mismatched).toMatchObject({ agent: 'claude', source: 'launch' })
})
})
describe('uncovered compatibility lane', () => {
it('preserves the caller-provided legacy result verbatim', () => {
const identity = resolveCanonicalPaneAgentIdentity({
title: 'Fix the parser - grok',
uncoveredFallback: { agent: 'grok', titleOnly: true }
})
expect(identity).toMatchObject({
agent: 'grok',
source: 'title',
coverage: 'uncovered',
titleOnly: true
})
})
it('answers from title evidence marked title-only when no fallback is supplied', () => {
const identity = resolveCanonicalPaneAgentIdentity({
title: 'STA-4011 Linux Antigravity Commit Messages - grok'
})
expect(identity).toMatchObject({
agent: 'grok',
source: 'title',
coverage: 'uncovered',
titleOnly: true
})
})
it('a legacy null stays null rather than re-deriving from the title', () => {
const identity = resolveCanonicalPaneAgentIdentity({
title: 'anything - grok',
uncoveredFallback: { agent: null }
})
expect(identity).toMatchObject({ agent: null, source: null, coverage: 'uncovered' })
})
it('does not let a legacy title fallback bypass the ambiguity fence', () => {
expect(
resolveCanonicalPaneAgentIdentity({
title: 'OC | something - grok',
uncoveredFallback: { agent: 'opencode', titleOnly: true }
})
).toMatchObject({ agent: null, source: null, ambiguousAt: 'title' })
expect(
resolveCanonicalPaneAgentIdentity({
title: 'compare codex with grok',
uncoveredFallback: { agent: 'codex', titleOnly: true }
})
).toMatchObject({ agent: null, source: null, coverage: 'uncovered' })
})
it('does not label a foreground-only compatibility answer as title-only', () => {
const identity = resolveCanonicalPaneAgentIdentity({
foregroundAgent: 'codex',
uncoveredFallback: { agent: 'codex' }
})
expect(identity).toMatchObject({
agent: 'codex',
source: null,
coverage: 'uncovered',
titleOnly: false
})
})
})
describe('canonical ladder inside the covered lane', () => {
it('keeps title last: a covered launch beats a parsed title', () => {
const identity = resolveCanonicalPaneAgentIdentity({
launchAgent: 'claude',
title: 'STA-4011 Linux Antigravity Commit Messages - grok'
})
expect(identity).toMatchObject({ agent: 'claude', source: 'launch', titleOnly: false })
})
it('sibling evidence needs the explicit tab-scope opt-in', () => {
const withoutOptIn = resolveCanonicalPaneAgentIdentity({
launchAgent: 'claude',
siblingAgent: 'codex'
})
expect(withoutOptIn.agent).toBe('claude')
const optedIn = resolveCanonicalPaneAgentIdentity({
hookAgent: 'claude',
hookIsLive: true,
siblingAgent: 'codex',
allowSibling: true
})
expect(optedIn).toMatchObject({ agent: 'claude', source: 'live-hook' })
})
it('surfaces ambiguity instead of picking by array order', () => {
const identity = resolveCanonicalPaneAgentIdentity({
hookAgent: 'claude',
hookIsLive: false,
completedHookAgent: 'codex'
})
expect(identity).toMatchObject({ agent: null, ambiguousAt: 'completed-hook' })
})
})
describe('reclaim-versus-stale-hook discriminator (run keys, not title text)', () => {
const run1 = { authorityId: 'main:a', incarnation: 1 }
const run2 = { authorityId: 'main:a', incarnation: 2 }
const otherAuthority = { authorityId: 'renderer:b', incarnation: 9 }
it('bug shape: hook and pane share the current run, so the completed hook wins over the title', () => {
const identity = resolveCanonicalPaneAgentIdentity({
completedHookAgent: 'claude',
completedHookRun: run1,
currentRun: run1,
title: 'STA-4011 Linux Antigravity Commit Messages - grok'
})
expect(identity).toMatchObject({ agent: 'claude', source: 'completed-hook' })
})
it('reclaim shape: a superseded hook is ineligible and the current title evidence answers', () => {
const identity = resolveCanonicalPaneAgentIdentity({
completedHookAgent: 'claude',
completedHookRun: run1,
currentRun: run2,
title: 'STA-4011 Linux Antigravity Commit Messages - grok'
})
expect(identity).toMatchObject({
agent: 'grok',
source: 'title',
coverage: 'uncovered',
titleOnly: true
})
expect(identity.supersededSources).toEqual(['completed-hook'])
})
it('cross-authority runs are incomparable, so the hook stays eligible', () => {
const identity = resolveCanonicalPaneAgentIdentity({
completedHookAgent: 'claude',
completedHookRun: otherAuthority,
currentRun: run2,
title: 'STA-4011 Linux Antigravity Commit Messages - grok'
})
expect(identity).toMatchObject({ agent: 'claude', source: 'completed-hook' })
})
it('an absent run key keeps evidence eligible (old peer), never guessed stale', () => {
const identity = resolveCanonicalPaneAgentIdentity({
completedHookAgent: 'claude',
currentRun: run2,
title: 'STA-4011 Linux Antigravity Commit Messages - grok'
})
expect(identity).toMatchObject({ agent: 'claude', source: 'completed-hook' })
})
})
describe('action floor', () => {
it("minimumSource: 'launch' refuses title and completed-hook answers outright", () => {
const identity = resolveCanonicalPaneAgentIdentity({
completedHookAgent: 'claude',
title: 'claude',
minimumSource: 'launch'
})
expect(identity).toMatchObject({ agent: null, source: null, coverage: 'covered' })
})
})
describe('wire evidence projection', () => {
it('publishes nothing for an absent identity — absence stays absence', () => {
expect(
buildPaneAgentIdentityEvidenceWire(resolveCanonicalPaneAgentIdentity({}))
).toBeUndefined()
})
it('marks the uncovered title-only route explicitly and carries the run key when known', () => {
const wire = buildPaneAgentIdentityEvidenceWire(
resolveCanonicalPaneAgentIdentity({ title: 'claude - claude' }),
{ authorityId: 'main:a', incarnation: 3 },
{ capturedAgeMs: 10, validForMs: 1_000 }
)
expect(wire).toMatchObject({
coverage: 'uncovered',
titleOnlyActionFallback: true,
authorityId: 'main:a',
incarnation: 3,
freshness: { capturedAgeMs: 10, validForMs: 1_000 }
})
})
it('a covered identity never carries the title-only action marker', () => {
const wire = buildPaneAgentIdentityEvidenceWire(
resolveCanonicalPaneAgentIdentity({ launchAgent: 'claude' })
)
expect(wire).toMatchObject({ source: 'launch', coverage: 'covered' })
expect(wire?.titleOnlyActionFallback).toBeUndefined()
})
})
+361
View File
@@ -0,0 +1,361 @@
import { collectAgentTitleEvidence } from './agent-title-evidence'
import { PANE_AGENT_EVIDENCE_SOURCES } from './pane-agent-evidence-sources'
import type {
PaneAgentEvidence,
PaneAgentIdentity,
PaneAgentIdentityInput,
PaneAgentRunKey
} from './pane-agent-identity-resolver'
import type { PaneAgentEvidenceSource } from './pane-agent-evidence-sources'
import type { TuiAgent } from './tui-agent'
/**
* Canonical pane identity ranking. All adapters, including the compatibility resolver, delegate to
* this implementation so source precedence, ambiguity, and run eligibility cannot drift.
*/
/**
* Whether the execution authority proved at least one identity-bearing source for this pane.
* Computed from evidence presence, never from `platform`, `isRemote`, or OS: a remote pane with a
* host-stamped hook is covered; a local pane with only a title is uncovered.
*/
export type PaneAgentCoverage = 'covered' | 'uncovered'
/**
* Host-stamped proof that a recognized agent process is the pane's foreground process.
*
* A process NAME is not a PID-reuse-safe identity, so a bare foreground read never enters the
* covered process rung. `processIncarnation` is an opaque token the execution host derives from
* the selected PID plus start/creation time (or an equivalent platform-native identity); the raw
* tuple never crosses the renderer/remote wire. The host emits no proof when the start identity
* is unavailable or ambiguous, so that pane reads `uncovered` rather than guessed.
*/
export type ForegroundProcessProof = {
agent: TuiAgent
/** Opaque host-derived PID+start-time token. Compared for equality only, never decoded. */
processIncarnation: string
ptyIncarnationId?: string
/** The execution authority that stamped the proof (see agent-status-observation.ts). */
authorityId: string
/** Age on the AUTHORITY's clock at capture. Replicas decay from this plus `validForMs`,
* never by subtracting a host wall clock from a local `Date.now()`. */
capturedAgeMs: number
validForMs: number
}
/**
* Positive evidence that a pane/process was REPLACED, required before any consumer may advance a
* pane incarnation. A retired-pane `restart` disposition, an accepted send, an ordinary provider
* turn boundary, a title change, a transport loss, or a renderer-only foreground change is never
* one of these. Defined here so the rebind-gate wave has a contract to be correct against; no
* sequencer call site consumes it yet.
*/
export type PaneReplacementProof =
| { kind: 'accepted-launch'; launchToken: string; ptyIncarnationId: string }
| { kind: 'process-replacement'; processIncarnation: string; authorityId: string }
| { kind: 'provider-session-attach'; providerSessionId: string }
/**
* The optional wire object a host will publish alongside `agentIdentity` after capability
* negotiation (host-publisher wave, not now). All fields bounded and JSON-safe; old peers ignore
* it. Never inferred from the bare `agentIdentity` string.
*/
export type PaneAgentIdentityEvidenceWire = {
source: PaneAgentEvidenceSource
coverage: PaneAgentCoverage
authorityId?: string
incarnation?: number
freshness?: { capturedAgeMs: number; validForMs: number }
/** Marks the scoped host-published title-only best-effort route (hand-started WSL panes).
* Counted separately, `unverifiable` for liveness, and never relabeled as covered proof. */
titleOnlyActionFallback?: true
}
export type CanonicalPaneAgentIdentityInput = {
hookAgent?: TuiAgent | null
hookIsLive?: boolean
hookRun?: PaneAgentRunKey
/** A distinct completed-hook signal for callers that hold live and completed rows separately
* (the tab ladder does); `hookAgent` + `hookIsLive: false` remains the single-slot spelling. */
completedHookAgent?: TuiAgent | null
completedHookRun?: PaneAgentRunKey
launchAgent?: TuiAgent | null
launchRun?: PaneAgentRunKey
/**
* Foreground process NAME as currently read. Without a fresh `processProof` this is a weak
* hint: it neither enters the covered process rung nor makes the pane covered.
*/
foregroundAgent?: TuiAgent | null
processProof?: ForegroundProcessProof | null
sleepingSessionAgent?: TuiAgent | null
sleepingRun?: PaneAgentRunKey
/** Tab-level display fallback only; ignored unless `allowSibling` opts in. */
siblingAgent?: TuiAgent | null
/** Additional tab-level sibling observations retained for ambiguity checking. */
siblingAgents?: readonly TuiAgent[]
allowSibling?: boolean
title?: string | null
currentRun?: PaneAgentRunKey
minimumSource?: PaneAgentEvidenceSource
/**
* The caller's CURRENT ladder result, preserved verbatim while the pane is uncovered. The
* uncovered lane is a temporary compatibility lane, not a new host-specific ranking; absent a
* fallback, an uncovered pane answers from title evidence alone, marked title-only.
*/
uncoveredFallback?: { agent: TuiAgent | null; titleOnly?: boolean }
}
export type CanonicalPaneAgentIdentity = {
agent: TuiAgent | null
source: PaneAgentEvidenceSource | null
coverage: PaneAgentCoverage
/** True when the answer was derived from a parsed title (the uncovered/title-only marking). */
titleOnly: boolean
ambiguousAt?: PaneAgentEvidenceSource
supersededSources: readonly PaneAgentEvidenceSource[]
}
/** Authority order, strongest first. This is the only place precedence is expressed. */
const SOURCE_RANK: readonly PaneAgentEvidenceSource[] = PANE_AGENT_EVIDENCE_SOURCES
/** Exported for the source/rank drift ratchet; the rank is the canonical source list itself. */
export const PANE_AGENT_SOURCE_RANK = SOURCE_RANK
/** Reject an unrecognised source instead of silently dropping it from the ranking loop. */
function sourceRankIndex(source: PaneAgentEvidenceSource): number {
const index = SOURCE_RANK.indexOf(source)
if (index === -1) {
throw new Error(`Unknown pane-agent evidence source: ${String(source)}`)
}
return index
}
/** Run keys only supersede evidence from the same authority; unknown authorities stay eligible. */
function isPaneAgentRunEligible(
run: PaneAgentRunKey | undefined,
currentRun: PaneAgentRunKey | undefined
): boolean {
return (
run === undefined ||
currentRun === undefined ||
run.authorityId !== currentRun.authorityId ||
run.incarnation === currentRun.incarnation
)
}
/** Shared evidence ranking primitive used by every pane-identity adapter. */
export function resolveCanonicalPaneAgentEvidence<A extends string = TuiAgent>(
input: PaneAgentIdentityInput<A>
): PaneAgentIdentity<A> {
const superseded: PaneAgentEvidenceSource[] = []
const floor = input.minimumSource ? sourceRankIndex(input.minimumSource) : Number.MAX_SAFE_INTEGER
const eligible = input.evidence.filter((item) => {
if (item.source === 'sibling' && input.allowSibling !== true) {
return false
}
if (sourceRankIndex(item.source) > floor) {
return false
}
if (isPaneAgentRunEligible(item.run, input.currentRun)) {
return true
}
superseded.push(item.source)
return false
})
for (const source of SOURCE_RANK) {
const matches = eligible.filter((item) => item.source === source)
if (matches.length === 0) {
continue
}
const agents = new Set(matches.map((item) => item.agent))
if (agents.size > 1) {
return { agent: null, source: null, ambiguousAt: source, supersededSources: superseded }
}
return { agent: matches[0].agent, source, supersededSources: superseded }
}
return { agent: null, source: null, supersededSources: superseded }
}
/** Freshness is judged on the authority's own clock: age at capture against its TTL. */
export function isForegroundProcessProofFresh(proof: ForegroundProcessProof): boolean {
return (
Number.isFinite(proof.capturedAgeMs) &&
Number.isFinite(proof.validForMs) &&
proof.capturedAgeMs >= 0 &&
proof.validForMs > 0 &&
proof.capturedAgeMs <= proof.validForMs
)
}
/** A proof only carries identity for the agent it names; a name mismatch is no proof at all. */
function processEvidenceFromProof(
input: CanonicalPaneAgentIdentityInput
): PaneAgentEvidence<TuiAgent> | null {
const proof = input.processProof
if (!proof || !isForegroundProcessProofFresh(proof)) {
return null
}
if (input.foregroundAgent && input.foregroundAgent !== proof.agent) {
return null
}
return { source: 'process', agent: proof.agent }
}
export function resolveCanonicalPaneAgentIdentity(
input: CanonicalPaneAgentIdentityInput
): CanonicalPaneAgentIdentity {
const processEvidence = processEvidenceFromProof(input)
// Coverage comes from authority-bearing sources that are still eligible for this run. A stale
// hook/launch row can remain in the input after a pane is replaced; it must not make a title-only
// answer look covered to a future action consumer.
const covered = Boolean(
(input.hookAgent && isPaneAgentRunEligible(input.hookRun, input.currentRun)) ||
(input.completedHookAgent &&
isPaneAgentRunEligible(input.completedHookRun, input.currentRun)) ||
processEvidence ||
(input.launchAgent && isPaneAgentRunEligible(input.launchRun, input.currentRun)) ||
(input.sleepingSessionAgent && isPaneAgentRunEligible(input.sleepingRun, input.currentRun))
)
// Keep stale evidence in the resolver so diagnostics still report which source was superseded,
// even when it no longer qualifies the pane as covered.
const hasAuthorityEvidence = Boolean(
input.hookAgent ||
input.completedHookAgent ||
processEvidence ||
input.launchAgent ||
input.sleepingSessionAgent
)
const titleEvidence = input.title ? collectAgentTitleEvidence(input.title) : null
const titleAgent = titleEvidence?.agent ?? null
if (!hasAuthorityEvidence) {
if (input.uncoveredFallback) {
const agent = input.uncoveredFallback.agent
// A legacy title parser may have picked the first token from an ambiguous or
// free-text-only title. Do not let that compatibility value bypass the canonical
// ambiguity fence when the caller marks it as title-only evidence.
const rejectTitleFallback =
input.uncoveredFallback.titleOnly === true &&
((titleEvidence?.reason === 'free-text-only' &&
(titleEvidence.freeTextNames?.length ?? 0) > 1) ||
titleEvidence?.reason === 'conflicting-anchored-names' ||
titleEvidence?.reason === 'conflicting-vendor-markers')
if (rejectTitleFallback) {
return {
agent: null,
source: null,
coverage: 'uncovered',
titleOnly: false,
...(titleEvidence?.reason === 'free-text-only' ? {} : { ambiguousAt: 'title' as const }),
supersededSources: []
}
}
const titleOnly =
input.uncoveredFallback.titleOnly ?? (agent !== null && agent === titleAgent)
return {
agent,
source: agent === null ? null : titleOnly ? 'title' : null,
coverage: 'uncovered',
titleOnly,
supersededSources: []
}
}
const siblingEvidence = [
...(input.siblingAgent ? [{ source: 'sibling' as const, agent: input.siblingAgent }] : []),
...(input.siblingAgents?.map((agent) => ({ source: 'sibling' as const, agent })) ?? []),
...(titleAgent ? [{ source: 'title' as const, agent: titleAgent }] : [])
]
const siblingResolved = resolveCanonicalPaneAgentEvidence<TuiAgent>({
evidence: siblingEvidence,
allowSibling: input.allowSibling,
minimumSource: input.minimumSource
})
return {
agent: siblingResolved.agent,
source: siblingResolved.source,
coverage: 'uncovered',
titleOnly: siblingResolved.source === 'title',
...(siblingResolved.ambiguousAt ? { ambiguousAt: siblingResolved.ambiguousAt } : {}),
supersededSources: siblingResolved.supersededSources
}
}
const resolved = resolveCanonicalPaneAgentEvidence<TuiAgent>({
evidence: [
...(input.hookAgent
? [
{
source: input.hookIsLive ? ('live-hook' as const) : ('completed-hook' as const),
agent: input.hookAgent,
...(input.hookRun ? { run: input.hookRun } : {})
}
]
: []),
...(input.completedHookAgent
? [
{
source: 'completed-hook' as const,
agent: input.completedHookAgent,
...(input.completedHookRun ? { run: input.completedHookRun } : {})
}
]
: []),
...(processEvidence ? [processEvidence] : []),
...(input.launchAgent
? [
{
source: 'launch' as const,
agent: input.launchAgent,
...(input.launchRun ? { run: input.launchRun } : {})
}
]
: []),
...(input.sleepingSessionAgent
? [
{
source: 'sleeping-session' as const,
agent: input.sleepingSessionAgent,
...(input.sleepingRun ? { run: input.sleepingRun } : {})
}
]
: []),
...(input.siblingAgent ? [{ source: 'sibling' as const, agent: input.siblingAgent }] : []),
...(input.siblingAgents?.map((agent) => ({ source: 'sibling' as const, agent })) ?? []),
...(titleAgent ? [{ source: 'title' as const, agent: titleAgent }] : [])
],
currentRun: input.currentRun,
minimumSource: input.minimumSource,
allowSibling: input.allowSibling
})
return {
agent: resolved.agent,
source: resolved.source,
coverage: covered ? 'covered' : 'uncovered',
titleOnly: resolved.source === 'title',
...(resolved.ambiguousAt ? { ambiguousAt: resolved.ambiguousAt } : {}),
supersededSources: resolved.supersededSources
}
}
/** Projects the host-local sidecar onto the optional wire shape. Returns undefined when there is
* nothing to publish — absence stays absence, and a bare `agentIdentity` with no sidecar is
* never treated as covered proof by any consumer. */
export function buildPaneAgentIdentityEvidenceWire(
identity: CanonicalPaneAgentIdentity,
run?: PaneAgentRunKey,
freshness?: { capturedAgeMs: number; validForMs: number }
): PaneAgentIdentityEvidenceWire | undefined {
if (identity.agent === null || identity.source === null) {
return undefined
}
return {
source: identity.source,
coverage: identity.coverage,
...(run ? { authorityId: run.authorityId, incarnation: run.incarnation } : {}),
...(freshness ? { freshness } : {}),
...(identity.coverage === 'uncovered' && identity.titleOnly
? { titleOnlyActionFallback: true as const }
: {})
}
}
@@ -18,7 +18,13 @@ const HELPERS = [
'resolveExplicitTerminalTitleAgentType',
'resolveCommittedTitleAgentType',
'resolvePaneAgentOwner',
'resolveCompatibleAgentTypeForOwner'
'resolveCompatibleAgentTypeForOwner',
'classifyTitleActivity',
'detectAgentStatusFromTitle',
'resolveAgentTypeFromTerminalTitle',
'resolvePaneAgentIdentity',
'resolveCanonicalPaneAgentIdentity',
'resolvePublishedPaneAgentIdentity'
] as const
const TEST_SUPPORT_PATHS = new Set([
@@ -242,6 +248,136 @@ const INVENTORY: readonly InventoryGroup[] = [
['src/renderer/src/components/terminal-pane/pty-connection/direct-ssh-retry-status.ts', 2],
['src/renderer/src/components/terminal-pane/pty-connection/title-spawn-bell.ts', 2]
]
},
{
helper: 'classifyTitleActivity',
classification: 'identity-consumer',
paths: [
['src/renderer/src/components/sidebar/smart-attention.ts', 3],
['src/renderer/src/components/sidebar/worktree-title-derived-agent-rows.ts', 2],
['src/renderer/src/components/status-bar/workspace-space-presentation.ts', 3],
['src/renderer/src/lib/active-agent-note-target.ts', 2],
['src/renderer/src/lib/worktree-status.ts', 3],
['src/renderer/src/store/slices/terminal-helpers.ts', 2]
]
},
{
helper: 'classifyTitleActivity',
classification: 'action-consumer',
paths: [
['src/renderer/src/components/terminal-pane/cache-timer-seeding.ts', 2],
['src/renderer/src/lib/agent-ready-wait.ts', 2],
['src/renderer/src/store/terminals/terminal-ephemeral-state.ts', 2]
]
},
{
helper: 'classifyTitleActivity',
classification: 'activity-only',
paths: [
['src/renderer/src/store/slices/workspace-cleanup-local-evidence.ts', 3],
['src/renderer/src/store/terminals/terminal-tab-presentation.ts', 4]
]
},
{
helper: 'classifyTitleActivity',
classification: 'evidence-producer',
paths: [
['src/renderer/src/lib/agent-send-title-status.ts', 2],
['src/renderer/src/lib/agent-status-terminal-title.ts', 2]
]
},
{
helper: 'classifyTitleActivity',
classification: 'parser-implementation',
paths: [
['src/renderer/src/lib/agent-status.ts', 5],
'src/renderer/src/lib/pane-agent-evidence.ts'
]
},
{
helper: 'detectAgentStatusFromTitle',
classification: 'evidence-producer',
paths: [
['src/main/runtime/orca-runtime-apply-tracked-pty-title.ts', 2],
['src/main/runtime/orca-runtime-get-pty-record-for-pane-key.ts', 2],
['src/main/runtime/orca-runtime-get-unpersisted-tracked-title-for-pty.ts', 2],
['src/main/runtime/orca-runtime-maybe-hydrate-headless-from-renderer.ts', 2],
['src/main/runtime/orca-runtime-record-agent-prompt-lifecycle-state.ts', 2],
['src/main/runtime/runtime-terminal-agent-status-query.ts', 3],
['src/main/runtime/runtime-worktree-status-projection.ts', 4],
['src/main/runtime/terminal-wait-detection.ts', 2],
['src/renderer/src/components/terminal-pane/agent-completion-title-observer.ts', 2],
['src/renderer/src/components/terminal-pane/pty-connection/shell-command-inference.ts', 4],
['src/renderer/src/components/terminal-pane/pty-output-title-observer.ts', 2],
['src/shared/terminal-output-side-effects.ts', 3]
]
},
{
helper: 'detectAgentStatusFromTitle',
classification: 'action-consumer',
paths: [
['src/renderer/src/components/terminal-pane/pty-connection/agent-task-complete-notify.ts', 2],
[
'src/renderer/src/components/terminal-pane/pty-connection/command-inferred-pane-agent.ts',
3
],
['src/renderer/src/components/terminal-pane/pty-connection/interrupt-input-intent.ts', 3]
]
},
{
helper: 'detectAgentStatusFromTitle',
classification: 'parser-implementation',
paths: [
['src/renderer/src/components/terminal-pane/title-agent-identity.ts', 2],
'src/renderer/src/lib/agent-status.ts',
['src/renderer/src/lib/pane-agent-evidence.ts', 3],
['src/shared/agent-decorative-title-signature.ts', 2],
'src/shared/agent-detection.ts',
['src/shared/agent-title-owner.ts', 2],
['src/shared/agent-title-status.ts', 6]
]
},
{
helper: 'resolveAgentTypeFromTerminalTitle',
classification: 'identity-consumer',
paths: [
['src/renderer/src/components/sidebar/worktree-agent-row-type.ts', 2],
'src/renderer/src/components/sidebar/worktree-title-derived-agent-rows.ts',
['src/renderer/src/lib/worktree-status.ts', 2]
]
},
{
helper: 'resolvePaneAgentIdentity',
classification: 'parser-implementation',
paths: ['src/shared/pane-agent-identity-resolver.ts']
},
{
helper: 'resolvePaneAgentIdentity',
classification: 'identity-consumer',
paths: [['src/shared/published-pane-agent-identity.ts', 2]]
},
{
helper: 'resolveCanonicalPaneAgentIdentity',
classification: 'parser-implementation',
paths: ['src/shared/pane-agent-identity-adapter.ts']
},
{
helper: 'resolveCanonicalPaneAgentIdentity',
classification: 'identity-consumer',
paths: [['src/shared/agent-status-identity.ts', 2]]
},
{
helper: 'resolveCanonicalPaneAgentIdentity',
classification: 'identity-consumer',
paths: [['src/shared/terminal-title-agent-type.ts', 2]]
},
{
helper: 'resolvePublishedPaneAgentIdentity',
classification: 'parser-implementation',
paths: [
'src/shared/published-pane-agent-identity.ts',
['src/main/runtime/orca-runtime-write-orchestration-pointer-pty.ts', 2]
]
}
]
@@ -1,4 +1,5 @@
import { describe, expect, it } from 'vitest'
import { PANE_AGENT_SOURCE_RANK } from './pane-agent-identity-adapter'
import {
PANE_AGENT_EVIDENCE_SOURCES,
type PaneAgentEvidence,
@@ -11,6 +12,14 @@ const resolve = (evidence: PaneAgentEvidence[], extra = {}) =>
const H = 'authority-a'
describe('resolvePaneAgentIdentity', () => {
it('keeps every evidence source ranked exactly once', () => {
expect(PANE_AGENT_SOURCE_RANK).toBe(PANE_AGENT_EVIDENCE_SOURCES)
expect(new Set(PANE_AGENT_SOURCE_RANK).size).toBe(PANE_AGENT_SOURCE_RANK.length)
for (const source of PANE_AGENT_EVIDENCE_SOURCES) {
expect(PANE_AGENT_SOURCE_RANK.indexOf(source)).toBeGreaterThanOrEqual(0)
}
})
describe('a display title is the last thing consulted', () => {
it.each(PANE_AGENT_EVIDENCE_SOURCES.filter((s) => s !== 'title' && s !== 'sibling'))(
'lets %s outrank a conflicting title',
@@ -153,6 +162,17 @@ describe('resolvePaneAgentIdentity', () => {
})
})
it('fails loudly when an evidence source is missing from the rank', () => {
expect(() =>
resolve([
{
source: 'future-source' as PaneAgentEvidence['source'],
agent: 'codex'
}
])
).toThrow('Unknown pane-agent evidence source')
})
describe('input order does not decide the answer', () => {
it('resolves the same regardless of how evidence is listed', () => {
const evidence: PaneAgentEvidence[] = [
+6 -69
View File
@@ -1,5 +1,10 @@
import { resolveCanonicalPaneAgentEvidence } from './pane-agent-identity-adapter'
import type { PaneAgentEvidenceSource } from './pane-agent-evidence-sources'
import type { TuiAgent } from './tui-agent'
export { PANE_AGENT_EVIDENCE_SOURCES } from './pane-agent-evidence-sources'
export type { PaneAgentEvidenceSource } from './pane-agent-evidence-sources'
/**
* One place that answers "which agent is in this pane".
*
@@ -23,27 +28,6 @@ import type { TuiAgent } from './tui-agent'
* launch, a recognized command at a shell prompt, a host-confirmed foreground change, a new
* provider session. Never by a title changing, and never by transport loss.
*/
export const PANE_AGENT_EVIDENCE_SOURCES = [
/** A live provider hook for a turn in progress. The agent is running and said so. */
'live-hook',
/** The pane's foreground process, as read on the execution host. */
'process',
/** Orca launched, resumed, or accepted a command for this agent. A fact Orca owns. */
'launch',
/** A provider hook from a turn that finished. Still authoritative about identity. */
'completed-hook',
/** A sleeping session record restored for this pane. */
'sleeping-session',
/** Another pane in the same tab. Tab-level surfaces only; never pane-scoped routing. */
'sibling',
/** Parsed from the terminal title. A decoration channel; anyone can type an agent's name. */
'title'
] as const
export type PaneAgentEvidenceSource = (typeof PANE_AGENT_EVIDENCE_SOURCES)[number]
/** Authority order, strongest first. Position here is the ONLY place precedence is expressed. */
const SOURCE_RANK: readonly PaneAgentEvidenceSource[] = PANE_AGENT_EVIDENCE_SOURCES
/**
* Which agent run a piece of evidence belongs to.
*
@@ -112,52 +96,5 @@ export type PaneAgentIdentity<A extends string = TuiAgent> = {
export function resolvePaneAgentIdentity<A extends string = TuiAgent>(
input: PaneAgentIdentityInput<A>
): PaneAgentIdentity<A> {
const superseded: PaneAgentEvidenceSource[] = []
const floor = input.minimumSource
? SOURCE_RANK.indexOf(input.minimumSource)
: Number.MAX_SAFE_INTEGER
const eligible = input.evidence.filter((item) => {
if (item.source === 'sibling' && input.allowSibling !== true) {
return false
}
// Why the floor: an action consumer must not be able to act on a title, at any rank. Dropping
// the evidence entirely rather than ranking it lower makes misuse impossible rather than
// unlikely — a caller cannot accidentally consult it by reordering.
if (SOURCE_RANK.indexOf(item.source) > floor) {
return false
}
if (input.currentRun === undefined || item.run === undefined) {
// Why eligible: absence means "this peer does not publish run keys", not "this is stale".
// Treating unknown as superseded would blank every row from an older host.
return true
}
if (item.run.authorityId !== input.currentRun.authorityId) {
// Why eligible and NOT superseded: runs from different authorities are incomparable, not
// older. A restarted main counts from its own floor, so `incarnation` alone would falsely
// equate unrelated runs. Incomparable evidence is treated as unknown, like an absent key.
return true
}
if (item.run.incarnation === input.currentRun.incarnation) {
return true
}
superseded.push(item.source)
return false
})
for (const source of SOURCE_RANK) {
const matches = eligible.filter((item) => item.source === source)
if (matches.length === 0) {
continue
}
const agents = new Set(matches.map((item) => item.agent))
if (agents.size > 1) {
// Why null and not the first: two observations of the same class naming different agents is
// a genuine conflict, and picking one would make the answer depend on array order — the very
// property this resolver exists to remove. Fall through to nothing rather than guess.
return { agent: null, source: null, ambiguousAt: source, supersededSources: superseded }
}
return { agent: matches[0].agent, source, supersededSources: superseded }
}
return { agent: null, source: null, supersededSources: superseded }
return resolveCanonicalPaneAgentEvidence(input)
}
@@ -0,0 +1,284 @@
import { readFileSync } from 'node:fs'
import { join } from 'node:path'
import { describe, expect, it } from 'vitest'
import { glob } from 'tinyglobby'
import { isTestFile, stripComments } from './source-scan/source-tree-scan'
/**
* Surface half of the identity inventory ratchet: every consumer decision point from the closed
* 65-row inventory (rows 3265 — the direct title/native-chat selectors, tab projections, mobile
* sync graph, lifecycle selectors, status/OSC ingress, worktree status, attention, and
* title-reset paths) is pinned to a marker symbol in its file. The helper-name census
* (`pane-agent-identity-inventory.test.ts`) is necessary but not sufficient — these files reach
* identity through direct reads a name census cannot see. Moving or renaming a marker means the
* inventory row must be re-classified, deliberately, before review.
*/
type SurfaceRow = {
/** Row number in the closed consumer inventory. */
row: number
path: string
marker: string
}
const SURFACE_ROWS: readonly SurfaceRow[] = [
{
row: 32,
path: 'src/renderer/src/components/terminal-pane/native-chat-leaf-title-agent.ts',
marker: 'resolveNativeChatLeafTitleAgent'
},
{
row: 32,
path: 'src/renderer/src/components/terminal-pane/use-terminal-pane-chat-state.ts',
marker: 'resolveNativeChatLeafTitleAgent'
},
{
row: 33,
path: 'src/renderer/src/components/terminal-pane/pty-connection/pane-agent-identity.ts',
marker: 'installPaneAgentIdentity'
},
{ row: 34, path: 'src/main/runtime/orchestration/groups.ts', marker: 'terminalIsAgent' },
{
row: 35,
path: 'src/renderer/src/lib/active-agent-note-target.ts',
marker: 'getActiveTerminalNoteTarget'
},
{
row: 36,
path: 'src/renderer/src/components/terminal-pane/terminal-agent-paste-bracketing.ts',
marker: 'resolveProtectedMultilinePasteOptionsForPane'
},
{
row: 37,
path: 'src/renderer/src/components/terminal-pane/command-code-output-ownership.ts',
marker: 'canCommandCodeOutputOwnPane'
},
{ row: 38, path: 'src/renderer/src/lib/agent-ready-wait.ts', marker: 'waitForAgentReady' },
{
row: 39,
path: 'src/renderer/src/lib/agent-paste-draft.ts',
marker: 'getSettingsForAgentTabRuntimeOwner'
},
{
row: 40,
path: 'src/renderer/src/lib/agent-followup-delivery.ts',
marker: 'sendFollowupPromptWhenAgentReady'
},
{
row: 41,
path: 'src/renderer/src/lib/codex-session-restart.ts',
marker: 'markLiveCodexSessionsForRestart'
},
{
row: 41,
path: 'src/renderer/src/lib/codex-pane-restart-eligibility.ts',
marker: 'isCodexForegroundProcess'
},
{
row: 42,
path: 'src/renderer/src/components/native-chat/native-chat-availability.ts',
marker: 'canToggleNativeChat'
},
{
row: 43,
path: 'src/renderer/src/components/native-chat/native-chat-pane-resolution.ts',
marker: 'resolveNativeChatSession'
},
{
row: 44,
path: 'src/renderer/src/components/terminal-pane/terminal-agent-session-continuation.ts',
marker: 'canContinueAgentSessionInNewSession'
},
{
row: 45,
path: 'src/renderer/src/components/terminal-pane/terminal-agent-session-fork.ts',
marker: 'prepareAgentSessionForkFromPane'
},
{
row: 46,
path: 'src/renderer/src/components/terminal-pane/agent-interrupt-inference.ts',
marker: 'isPlainEscapeKeyEvent'
},
{
row: 46,
path: 'src/renderer/src/components/terminal-pane/agent-question-answered-inference.ts',
marker: 'inferQuestionAnsweredFromCurrentStatus'
},
{
row: 47,
path: 'src/renderer/src/components/terminal-pane/terminal-keyboard-protocol-pane-agent.ts',
marker: 'resolvePaneKeyboardProtocolAgent'
},
{
row: 47,
path: 'src/renderer/src/components/terminal-pane/terminal-pane-manager-options.ts',
marker: 'resolvePaneKeyboardProtocolAgent'
},
{
row: 48,
path: 'src/renderer/src/components/tab-bar/tab-agent-types-by-tab-id.ts',
marker: 'selectTabAgentTypesByTabId'
},
{
row: 49,
path: 'src/renderer/src/components/terminal-pane/terminal-tab-agent-type-index.ts',
marker: 'createTerminalTabAgentTypeSelector'
},
{
row: 50,
path: 'src/renderer/src/lib/tab-agent-status-index.ts',
marker: 'selectLiveTabAgentPanes'
},
{
row: 51,
path: 'src/renderer/src/components/tab-bar/terminal-tab-activity-status.ts',
marker: 'resolveTerminalTabActivityStatus'
},
{
row: 52,
path: 'src/renderer/src/lib/workspace-tab-agent-metadata.ts',
marker: 'maxAgentActivityAt'
},
{
row: 52,
path: 'src/renderer/src/lib/workspace-tab-palette-entry-builder.ts',
marker: 'buildSearchableWorkspaceTabEntries'
},
{
row: 53,
path: 'src/renderer/src/lib/running-agent-targets.ts',
marker: 'deriveRunningAgentSendTargets'
},
{
row: 54,
path: 'src/renderer/src/runtime/sync-runtime-graph.ts',
marker: 'buildMobileSessionTabSnapshots'
},
{
row: 55,
path: 'src/renderer/src/lib/agent-hibernation-pane-eligibility.ts',
marker: 'toRuntimePtyId'
},
{
row: 56,
path: 'src/renderer/src/lib/resume-sleeping-agent-session.ts',
marker: 'resumeSleepingAgentSessionsForWorktree'
},
{
row: 57,
path: 'src/renderer/src/lib/automation-session-reuse.ts',
marker: 'findReusableAutomationSession'
},
{
row: 58,
path: 'src/renderer/src/components/terminal-pane/pty-connection/cold-restore-resume-startup.ts',
marker: 'bindBuildColdRestoreAgentResumeStartup'
},
{
row: 59,
path: 'src/main/agent-hooks/server/server-authority-evidence.ts',
marker: 'recordCurrentAuthorityObservation'
},
{
row: 59,
path: 'src/main/runtime/orca-runtime-write-orchestration-pointer-pty.ts',
marker: 'resolvePaneAgentIdentityField'
},
{
row: 59,
path: 'src/renderer/src/hooks/ipc-events/agent-status-event-applicator.ts',
marker: 'createAgentStatusEventApplicator'
},
{
row: 59,
path: 'src/renderer/src/store/slices/agent-status-authority-actions.ts',
marker: 'transferAgentPaneAuthority'
},
{
row: 59,
path: 'src/renderer/src/store/slices/pane-foreground-agent.ts',
marker: 'createPaneForegroundAgentSlice'
},
{
row: 59,
path: 'src/renderer/src/hooks/ipc-events/agent-status-routing.ts',
marker: 'isAgentStatusForRecentlyClosedTab'
},
{
row: 60,
path: 'src/renderer/src/components/terminal-pane/pty-connection/title-spawn-bell.ts',
marker: 'installTitleSpawnBell'
},
{ row: 61, path: 'src/renderer/src/lib/worktree-status.ts', marker: 'getWorktreeStatus' },
{
row: 62,
path: 'src/main/runtime/runtime-worktree-status-projection.ts',
marker: 'getLeafWorktreeStatus'
},
{
row: 63,
path: 'src/renderer/src/components/sidebar/smart-attention.ts',
marker: 'buildAttentionByWorktree'
},
{
row: 64,
path: 'src/renderer/src/components/status-bar/workspace-space-presentation.ts',
marker: 'countWorkspaceSpaceActiveAgents'
},
{ row: 65, path: 'src/renderer/src/store/slices/terminal-helpers.ts', marker: 'getResetTitle' },
{
row: 6,
path: 'src/renderer/src/runtime/web-session-tabs-sync.ts',
marker: 'applyWebSessionTabs'
}
]
describe('pane agent identity surface inventory (rows 6, 3265)', () => {
it('every pinned surface still carries its marker symbol', () => {
for (const row of SURFACE_ROWS) {
const source = stripComments(readFileSync(join(process.cwd(), row.path), 'utf8'))
expect({ row: row.row, path: row.path, hasMarker: source.includes(row.marker) }).toEqual({
row: row.row,
path: row.path,
hasMarker: true
})
}
})
})
/**
* Identity-observation rebind audit. Advancing a pane incarnation without a positive replacement
* proof is how a legitimate reclaim and a stale-hook bug get conflated (see
* `PaneReplacementProof` in pane-agent-identity-adapter.ts). Every existing sequencer `rebind`
* call is pinned here by file and count: today they are the retired-pane `restart` disposition
* (three ingress paths) and the renderer pane-key transfer. Adding a rebind call, or changing
* these, requires updating this audit — and per the migration plan, a `replacementProof`.
*/
const IDENTITY_SEQUENCER_REBIND_RE = /\b(?:observations|rendererAgentStatusObservations)\.rebind\(/g
const EXPECTED_REBIND_SITES: readonly (readonly [path: string, occurrences: number])[] = [
['src/main/agent-hooks/server/server-ingest-normalization.ts', 1],
['src/main/agent-hooks/server/server-ingest-remote.ts', 1],
['src/main/agent-hooks/server/server-lifecycle.ts', 1],
['src/renderer/src/store/slices/agent-status-authority-actions.ts', 1]
]
describe('identity observation rebind audit', () => {
it('pins every identity-sequencer rebind call site by file and count', async () => {
const files = await glob(['src/**/*.{ts,tsx}', 'mobile/src/**/*.{ts,tsx}'], {
ignore: ['**/*.test.*', '**/*.spec.*']
})
const actual: [string, number][] = []
for (const path of files.sort()) {
if (isTestFile(path)) {
continue
}
const source = stripComments(readFileSync(join(process.cwd(), path), 'utf8'))
const occurrences = source.match(IDENTITY_SEQUENCER_REBIND_RE)?.length ?? 0
if (occurrences > 0) {
actual.push([path, occurrences])
}
}
expect(actual).toEqual(EXPECTED_REBIND_SITES.map((site) => [...site]))
}, 30_000)
})
@@ -0,0 +1,156 @@
import { existsSync, readdirSync, readFileSync } from 'node:fs'
import { homedir } from 'node:os'
import { join } from 'node:path'
import { createHash, randomBytes } from 'node:crypto'
import { describe, expect, it } from 'vitest'
import { collectAgentTitleEvidence } from './agent-title-evidence'
import { resolveCanonicalPaneAgentIdentity } from './pane-agent-identity-adapter'
import type { TuiAgent } from './tui-agent'
/**
* Title regression gates for the identity-ladder migration.
*
* Two layers: a controlled fixture table that always runs (CI-safe), and a local characterization
* gate over the machine's real recorded corpus. The corpus gate is not a CI prerequisite tied to
* one developer's home — when no history exists it reports `corpus unavailable — skipped`
* explicitly, never a silently green zero-title run. Raw titles never reach logs or failure
* output; changed titles are reported as salted hashes plus old/new agent summaries only.
*/
const RECORDED_HISTORY_DIR = 'terminal-history'
const QUARANTINE_DIR = '.recovery-quarantine'
function orcaAppSupportCandidates(): string[] {
if (process.platform === 'darwin') {
return [join(homedir(), 'Library', 'Application Support', 'Orca')]
}
if (process.platform === 'win32') {
return [join(process.env.APPDATA ?? join(homedir(), 'AppData', 'Roaming'), 'Orca')]
}
return [
join(process.env.XDG_CONFIG_HOME ?? join(homedir(), '.config'), 'Orca'),
join(process.env.XDG_DATA_HOME ?? join(homedir(), '.local', 'share'), 'Orca')
]
}
/** Deliberately shallow: `terminal-history/<session>/checkpoint.json` only, with the hidden
* quarantine subtree excluded BY NAME so a future recursive rewrite cannot silently turn
* quarantined recovery data into product regressions. */
function loadRecordedTitleCorpus(): { checkpointCount: number; titles: string[] } | null {
const root = orcaAppSupportCandidates()
.map((candidate) => join(candidate, RECORDED_HISTORY_DIR))
.find((candidate) => existsSync(candidate))
if (!root) {
return null
}
let checkpointCount = 0
const titles = new Set<string>()
for (const entry of readdirSync(root, { withFileTypes: true })) {
if (!entry.isDirectory() || entry.name === QUARANTINE_DIR || entry.name.startsWith('.')) {
continue
}
const checkpointPath = join(root, entry.name, 'checkpoint.json')
if (!existsSync(checkpointPath)) {
continue
}
const parsed: unknown = JSON.parse(readFileSync(checkpointPath, 'utf8'))
checkpointCount += 1
const lastTitle = (parsed as { lastTitle?: unknown }).lastTitle
if (typeof lastTitle === 'string' && lastTitle.length > 0) {
titles.add(lastTitle)
}
}
return { checkpointCount, titles: [...titles] }
}
/** What the canonical adapter answers when a title is all a pane has (the uncovered lane). */
function canonicalTitleOnlyAgent(title: string): TuiAgent | null {
return resolveCanonicalPaneAgentIdentity({ title }).agent
}
describe('controlled title fixtures (always run)', () => {
const FIXTURES: readonly { name: string; title: string; expected: TuiAgent | null }[] = [
{
name: 'mandatory adversarial owner suffix beats the agent names in task text',
title: 'STA-4011 Linux Antigravity Commit Messages - grok',
expected: 'grok'
},
{
name: 'task text mentioning other agents is not identity',
title: 'Compare Antigravity with Gemini 3.7 Flash',
expected: null
},
{
name: 'owner suffix still answers over mentioned agents',
title: 'Compare Antigravity with Gemini 3.7 Flash… - grok',
expected: 'grok'
},
{ name: 'Claude status sigil is a vendor marker', title: '✳', expected: 'claude' },
{ name: 'Claude management screen is not identity', title: 'claude agents', expected: null },
{ name: 'a shell title names no agent', title: 'zsh', expected: null },
{ name: 'a default worktree-ish title names no agent', title: 'my-claude-fix', expected: null },
{
name: 'conflicting vendor markers resolve to nothing',
title: '✳ | ✦ two sigils',
expected: null
},
{
name: 'conflicting anchored names resolve to nothing',
title: 'OC | something… - grok',
expected: null
},
{ name: 'a bare Pi title anchors as Pi', title: 'pi', expected: 'pi' },
{ name: 'an OMP status title anchors as OMP', title: 'omp ready', expected: 'omp' },
{
// Wrapper-frame π/OMP separators are handled by the synthetic-title path, not this
// evidence parser; pinned so a parser change here is a deliberate decision.
name: 'a π wrapper frame is declined by the evidence parser',
title: 'π : ready',
expected: null
}
]
for (const fixture of FIXTURES) {
it(fixture.name, () => {
expect(collectAgentTitleEvidence(fixture.title).agent).toBe(fixture.expected)
// The adapter's title-only lane must give the very same answer — phase 1 changes no
// parser semantics, only provenance.
expect(canonicalTitleOnlyAgent(fixture.title)).toBe(fixture.expected)
})
}
})
describe('recorded title corpus characterization (local gate)', () => {
it('the canonical title-only lane matches the shipped parser on every recorded title', (ctx) => {
const corpus = loadRecordedTitleCorpus()
if (corpus === null) {
console.info('corpus unavailable — skipped (no recorded terminal history on this machine)')
ctx.skip()
return
}
// A machine WITH history must never pass on an empty read — that would be a silently green
// zero-title run, not a characterization.
expect(corpus.checkpointCount).toBeGreaterThan(0)
expect(corpus.titles.length).toBeGreaterThan(0)
console.info(
`corpus: ${corpus.checkpointCount} checkpoints, ${corpus.titles.length} distinct titles`
)
const salt = randomBytes(16).toString('hex')
const changed: { titleHash: string; oldAgent: string | null; newAgent: string | null }[] = []
for (const title of corpus.titles) {
const oldAgent = collectAgentTitleEvidence(title).agent
const newAgent = canonicalTitleOnlyAgent(title)
if (oldAgent !== newAgent) {
changed.push({
titleHash: createHash('sha256').update(`${salt}:${title}`).digest('hex').slice(0, 16),
oldAgent,
newAgent
})
}
}
// Report hashes and agent summaries only; a reviewer who needs the raw value inspects the
// protected corpus on the machine that owns it.
expect(changed).toEqual([])
}, 60_000)
})
+36
View File
@@ -37,6 +37,42 @@ describe('resolvePaneAgentOwner', () => {
).toBe('omp')
})
it('preserves the pre-tranche precedence for every conflicting owner tier', () => {
expect(
resolvePaneAgentOwnerRecord({
launchAgent: 'claude',
hookAgent: 'codex',
siblingHookAgent: 'gemini',
completedHookAgent: 'pi',
sleepingSessionAgent: 'omp'
})
).toEqual({ agent: 'claude', ownerIsLaunch: true })
expect(
resolvePaneAgentOwnerRecord({
hookAgent: 'claude',
siblingHookAgent: 'codex',
completedHookAgent: 'gemini',
siblingCompletedHookAgent: 'pi',
sleepingSessionAgent: 'omp'
})
).toEqual({ agent: 'claude', ownerIsLaunch: false })
expect(
resolvePaneAgentOwnerRecord({
siblingHookAgent: 'codex',
completedHookAgent: 'claude',
siblingCompletedHookAgent: 'gemini',
sleepingSessionAgent: 'omp'
})
).toEqual({ agent: 'codex', ownerIsLaunch: false })
expect(
resolvePaneAgentOwnerRecord({
completedHookAgent: 'claude',
siblingCompletedHookAgent: 'codex',
sleepingSessionAgent: 'gemini'
})
).toEqual({ agent: 'claude', ownerIsLaunch: false })
})
it('returns null when no owner evidence exists', () => {
expect(resolvePaneAgentOwner({})).toBeNull()
expect(resolvePaneAgentOwner({ launchAgent: null, hookAgent: undefined })).toBeNull()
+4 -3
View File
@@ -48,9 +48,10 @@ const PANE_OWNER_RANK: readonly {
]
/**
* The single authoritative resolver for "which agent owns this pane", shared by
* the tab-icon resolver, the terminal-pane display/renderer owner, and the
* mirrored-tab title owner so they cannot drift apart.
* Compatibility owner lookup shared by the existing consumer surfaces.
*
* Tranche 0 intentionally preserves this pre-migration precedence byte-for-byte; switching
* these consumers to canonical evidence belongs to tranche 1.
*
* Why this precedence: launch intent is the authoritative bootstrap before any
* process signal exists, so it leads. Once launch metadata is gone — a mirrored
+13 -10
View File
@@ -10,6 +10,7 @@ import {
getPiCompatibleSyntheticAgentLabel,
isLegacyPiCompatibleTitle
} from './pi-compatible-synthetic-title'
import { resolveCanonicalPaneAgentIdentity } from './pane-agent-identity-adapter'
import { memoizeTitleClassification } from './terminal-title-classification-memo'
import type { TuiAgent } from './tui-agent'
@@ -217,10 +218,10 @@ function computeAgentLabel(title: string): string | null {
return null
}
// Maps getAgentLabel()'s product labels to TuiAgent ids — the fallback for
// agents whose foreground PROCESS name isn't self-identifying (Claude Code runs
// as `node`, but its "✳ Claude Code" title resolves here). Agents whose process
// name already matches (codex, etc.) never reach this path.
/** Pure in `title` — memoized so repeated selector reads skip the regex ladder. */
export const getAgentLabel: (title: string) => string | null =
memoizeTitleClassification(computeAgentLabel)
const TITLE_LABEL_TO_AGENT: Partial<Record<string, TuiAgent>> = {
'Claude Code': 'claude',
OpenClaude: 'openclaude',
@@ -240,10 +241,6 @@ const TITLE_LABEL_TO_AGENT: Partial<Record<string, TuiAgent>> = {
OMP: 'omp'
}
/** Pure in `title` — memoized so repeated selector reads skip the regex ladder. */
export const getAgentLabel: (title: string) => string | null =
memoizeTitleClassification(computeAgentLabel)
function hasGenericClaudeStatusPrefix(title: string): boolean {
return (
containsAgentSpinnerGlyph(title) ||
@@ -266,7 +263,13 @@ function isGenericClaudeStatusClaim(title: string, titleAgent: TuiAgent | null):
export function resolveTerminalTitleAgentType(title: string): TuiAgent | null {
const label = getAgentLabel(title)
return label ? (TITLE_LABEL_TO_AGENT[label] ?? null) : null
const parsed = label ? (TITLE_LABEL_TO_AGENT[label] ?? null) : null
return resolveCanonicalPaneAgentIdentity({
title,
// Preserve this public title-parser adapter's historical answer; pane identity
// consumers pass raw titles to the canonical resolver and enforce its fence.
uncoveredFallback: { agent: parsed, titleOnly: false }
}).agent
}
/**
@@ -283,6 +286,6 @@ function computeExplicitTerminalTitleAgentType(title: string): TuiAgent | null {
return titleAgent
}
/** Pure in `title` — memoized so repeated selector reads skip the regex ladder. */
/** Pure in `title` — memoized so repeated selector reads skip the canonical/title parse. */
export const resolveExplicitTerminalTitleAgentType: (title: string) => TuiAgent | null =
memoizeTitleClassification(computeExplicitTerminalTitleAgentType)