openclaw - ✅(Solved) Fix feat: include usage/cost metadata in agent.wait terminal response [2 pull requests, 2 comments, 2 participants]

openclaw2026-03-18 05:20:02

ON THIS PAGE

Recommended Tools

×6

Utilities matched from this issue’s tags and category — try them while you read without losing context.

GitHub issue graph ai analysis

Paste a GitHub issue URL. We fetch that issue, discover linked issues from bodies/comments/timeline, collect linked pull requests, and produce a structured English report.

The report is written in English Markdown for sharing and archival.

GitHub issue URL

Helpful · Quick feedback

GitHub stats

openclaw/openclaw#49494•Fetched 2026-04-08 00:54:44

View on GitHub

Comments

Participants

Timeline

Reactions

Author

shivasymbl

Participants

Ryce

shivasymbl

Timeline (top)

cross-referenced ×3commented ×2referenced ×1

The agent.wait WebSocket response currently returns only { status, startedAt, endedAt, error }. External orchestrators like Paperclip need per-run token usage and cost data to enforce budgets and track spend — but this data never reaches the wire despite being computed internally.

Error Message

// agent.ts L790-796 respond(true, { runId, status: snapshot.status, startedAt: snapshot.startedAt, endedAt: snapshot.endedAt, error: snapshot.error, });

Root Cause

Paperclip's OpenClaw Gateway adapter (@paperclipai/adapter-openclaw-gateway) already has parseUsage() ready to consume usage, costUsd, provider, and model from response metadata — it just gets undefined today.
Paperclip PR #884 fixes the meta extraction priority chain on the consuming side.
Paperclip PR #949 (merged) added full billing ledger support including biller vs provider distinction, budget enforcement, and cost UI — all waiting for upstream cost data.
Without this, OpenClaw agents in Paperclip show $0 spend and budget enforcement never triggers.

Fix Action

Fix / Workaround

Paperclip PR #884 — consuming side meta extraction fix
Paperclip PR #949 — billing ledger and cost control plane
Paperclip PR #386 — heartbeat cost recording fix (merged in v0.3.1)

PR fix notes

PR #884: feat: openclaw-gateway adapter — include contextUsagePercent in run result metadata (ALE-1129)

Repository: paperclipai/paperclip
Author: alexkerber
State: closed | merged: False
Link: https://github.com/paperclipai/paperclip/pull/884

Description (problem / solution / changelog)

Summary

Resolves ALE-1129.

The OpenClaw gateway cloud adapter now correctly surfaces context window usage (percentage, tokens used, and context window size) in the run result's usage field.

Changes

`packages/adapters/openclaw-gateway/src/server/execute.ts`

1. Extend parseUsage() with context window fields

Extracts contextUsagePercent, contextTokensUsed, contextTokensMax from the raw usage record
Derives contextUsagePercent from token counts when not explicitly provided: Math.round(contextTokensUsed / contextTokensMax * 100)
These are forwarded into AdapterExecutionResult.usage which Paperclip stores and surfaces in the live-agents dashboard

2. Fix meta lookup to also cover agent.wait response Previously, result.meta.agentMeta was only read from acceptedPayload (the initial agent response). For most runs, OpenClaw sends the final result metadata inside the agent.wait terminal snapshot. The code now checks both:

acceptedPayload.result.meta  → acceptedPayload.meta
  → latestResultPayload.result.meta  → latestResultPayload.meta

3. Use lastCallUsage.input as contextTokensUsed OpenClaw's agentMeta.usage.input is the accumulated sum of input tokens across all tool-use loops in a run. This overstates the actual context window consumption. agentMeta.lastCallUsage.input reflects only the last API call — the true context snapshot at end of run. When lastCallUsage.input is available and no explicit contextTokensUsed is provided, it is used as the preferred source.

4. Merge all context fields into rawUsage Explicit contextUsagePercent / contextTokensUsed / contextTokensMax at both agentMeta and meta levels are merged into the rawUsage object so they override the accumulated token fallbacks.

`server/src/tests/openclaw-gateway-adapter.test.ts`

Added a new test that:

Creates a mock gateway returning agentMeta.lastCallUsage (with input: 35000) alongside agentMeta.usage.input (accumulated: 50000) in the agent.wait response
Asserts result.usage.contextTokensUsed === 35000 (lastCallUsage wins)
Asserts result.usage.inputTokens === 50000 (accumulated usage unchanged)

Test results

Test Files  34 passed (34)
     Tests  144 passed (144)

Changed files

packages/adapters/openclaw-gateway/src/server/execute.ts (modified, +66/-3)
server/src/__tests__/openclaw-gateway-adapter.test.ts (modified, +101/-0)

PR #949: feat(costs): add billing, quota, and budget control plane

Repository: paperclipai/paperclip
Author: cryppadotta
State: closed | merged: True
Link: https://github.com/paperclipai/paperclip/pull/949

Description (problem / solution / changelog)

Summary

add billing, finance, quota, and budget control plane work
add agent, project, and company budget enforcement plus budget approvals
add /costs UI for providers, billers, finance, quotas, and budgets

Verification

pnpm install
pnpm -r typecheck
pnpm test:run
pnpm build
browser smoke test against http://127.0.0.1:3104/PAP/dashboard and /PAP/costs

Changed files

cli/src/__tests__/company-delete.test.ts (modified, +2/-0)
doc/plans/2026-03-14-billing-ledger-and-reporting.md (added, +468/-0)
doc/plans/2026-03-14-budget-policies-and-enforcement.md (added, +611/-0)
packages/adapter-utils/src/billing.test.ts (added, +28/-0)
packages/adapter-utils/src/billing.ts (added, +20/-0)
packages/adapter-utils/src/index.ts (modified, +3/-0)
packages/adapter-utils/src/types.ts (modified, +47/-1)
packages/adapters/claude-local/package.json (modified, +3/-1)
packages/adapters/claude-local/src/cli/quota-probe.ts (added, +124/-0)
packages/adapters/claude-local/src/server/execute.ts (modified, +7/-1)
packages/adapters/claude-local/src/server/index.ts (modified, +12/-0)
packages/adapters/claude-local/src/server/quota.ts (added, +531/-0)
packages/adapters/codex-local/package.json (modified, +2/-1)
packages/adapters/codex-local/src/cli/quota-probe.ts (added, +112/-0)
packages/adapters/codex-local/src/server/execute.ts (modified, +15/-3)
packages/adapters/codex-local/src/server/index.ts (modified, +11/-0)
packages/adapters/codex-local/src/server/quota.ts (added, +556/-0)
packages/adapters/cursor-local/src/server/execute.ts (modified, +20/-3)
packages/adapters/gemini-local/src/server/execute.ts (modified, +8/-2)
packages/adapters/opencode-local/src/server/execute.ts (modified, +6/-1)
packages/adapters/pi-local/src/server/execute.ts (modified, +6/-1)
packages/db/package.json (modified, +1/-0)
packages/db/src/migration-runtime.ts (modified, +56/-6)
packages/db/src/migration-status.ts (modified, +17/-1)
packages/db/src/migrations/0031_zippy_magma.sql (added, +51/-0)
packages/db/src/migrations/0032_pretty_doctor_octopus.sql (added, +102/-0)
packages/db/src/migrations/0033_shiny_black_tarantula.sql (added, +2/-0)
packages/db/src/migrations/0034_fat_dormammu.sql (added, +2/-0)
packages/db/src/migrations/meta/0031_snapshot.json (added, +7242/-0)
packages/db/src/migrations/meta/0032_snapshot.json (added, +7733/-0)
packages/db/src/migrations/meta/0033_snapshot.json (added, +9038/-0)
packages/db/src/migrations/meta/0034_snapshot.json (added, +9039/-0)
packages/db/src/migrations/meta/_journal.json (modified, +28/-0)
packages/db/src/schema/agents.ts (modified, +2/-0)
packages/db/src/schema/budget_incidents.ts (added, +42/-0)
packages/db/src/schema/budget_policies.ts (added, +43/-0)
packages/db/src/schema/companies.ts (modified, +2/-0)
packages/db/src/schema/cost_events.ts (modified, +19/-0)
packages/db/src/schema/finance_events.ts (added, +67/-0)
packages/db/src/schema/index.ts (modified, +3/-0)
packages/db/src/schema/projects.ts (modified, +2/-0)
packages/shared/src/constants.ts (modified, +71/-1)
packages/shared/src/index.ts (modified, +45/-0)
packages/shared/src/types/agent.ts (modified, +3/-0)
packages/shared/src/types/budget.ts (added, +99/-0)
packages/shared/src/types/company.ts (modified, +3/-1)
packages/shared/src/types/cost.ts (modified, +77/-0)
packages/shared/src/types/dashboard.ts (modified, +6/-0)
packages/shared/src/types/finance.ts (added, +60/-0)
packages/shared/src/types/index.ts (modified, +11/-1)
packages/shared/src/types/project.ts (modified, +3/-1)
packages/shared/src/types/quota.ts (added, +26/-0)
packages/shared/src/validators/budget.ts (added, +37/-0)
packages/shared/src/validators/cost.ts (modified, +9/-1)
packages/shared/src/validators/finance.ts (added, +34/-0)
packages/shared/src/validators/index.ts (modified, +12/-0)
scripts/dev-runner.mjs (modified, +48/-10)
server/package.json (modified, +1/-1)
server/src/__tests__/budgets-service.test.ts (added, +311/-0)
server/src/__tests__/companies-route-path-guard.test.ts (modified, +3/-0)
server/src/__tests__/costs-service.test.ts (added, +226/-0)
server/src/__tests__/monthly-spend-service.test.ts (added, +90/-0)
server/src/__tests__/quota-windows-service.test.ts (added, +56/-0)
server/src/__tests__/quota-windows.test.ts (added, +812/-0)
server/src/adapters/registry.ts (modified, +4/-0)
server/src/index.ts (modified, +7/-10)
server/src/routes/agents.ts (modified, +15/-0)
server/src/routes/companies.ts (modified, +20/-1)
server/src/routes/costs.ts (modified, +207/-5)
server/src/routes/issues.ts (modified, +13/-0)
server/src/services/agents.ts (modified, +67/-7)
server/src/services/approvals.ts (modified, +16/-0)
server/src/services/budgets.ts (added, +958/-0)
server/src/services/companies.ts (modified, +67/-10)
server/src/services/costs.ts (modified, +221/-61)
server/src/services/dashboard.ts (modified, +9/-0)
server/src/services/finance.ts (added, +134/-0)
server/src/services/heartbeat.ts (modified, +342/-84)
server/src/services/index.ts (modified, +2/-0)
server/src/services/quota-windows.ts (added, +64/-0)
ui/src/App.tsx (modified, +1/-0)
ui/src/api/budgets.ts (added, +20/-0)
ui/src/api/costs.ts (modified, +41/-9)
ui/src/components/AccountingModelCard.tsx (added, +69/-0)
ui/src/components/ApprovalCard.tsx (modified, +4/-1)
ui/src/components/ApprovalPayload.tsx (modified, +25/-1)
ui/src/components/BillerSpendCard.tsx (added, +145/-0)
ui/src/components/BudgetIncidentCard.tsx (added, +100/-0)
ui/src/components/BudgetPolicyCard.tsx (added, +219/-0)
ui/src/components/BudgetSidebarMarker.tsx (added, +13/-0)
ui/src/components/ClaudeSubscriptionPanel.tsx (added, +140/-0)
ui/src/components/CodexSubscriptionPanel.tsx (added, +157/-0)
ui/src/components/FinanceBillerCard.tsx (added, +44/-0)
ui/src/components/FinanceKindCard.tsx (added, +43/-0)
ui/src/components/FinanceTimelineCard.tsx (added, +71/-0)
ui/src/components/ProviderQuotaCard.tsx (added, +416/-0)
ui/src/components/QuotaBar.tsx (added, +65/-0)
ui/src/components/SidebarAgents.tsx (modified, +16/-8)
ui/src/components/SidebarProjects.tsx (modified, +2/-0)
ui/src/context/LiveUpdatesProvider.tsx (modified, +4/-0)

Code Example

// agent-wait-dedupe.ts L3-7
export type AgentWaitTerminalSnapshot = {
  status: "ok" | "error" | "timeout";
  startedAt?: number;
  endedAt?: number;
  error?: string;
};

---

// agent.ts L790-796
respond(true, {
  runId,
  status: snapshot.status,
  startedAt: snapshot.startedAt,
  endedAt: snapshot.endedAt,
  error: snapshot.error,
});

---

export type AgentWaitTerminalSnapshot = {
  status: "ok" | "error" | "timeout";
  startedAt?: number;
  endedAt?: number;
  error?: string;
  meta?: {
    usage?: {
      input: number;
      output: number;
      cacheRead?: number;
      cacheWrite?: number;
    };
    lastCallUsage?: {
      input: number;
      output: number;
    };
    costUsd?: number;
    provider?: string;
    model?: string;
  };
};

RAW_BUFFERClick to expand / collapse

Summary

The Gap

OpenClaw already tracks this internally:

normalizeUsage() in src/agents/usage.ts handles every provider format
estimateUsageCost() in src/utils/usage-format.ts computes per-model costs
loadSessionCostSummary() in src/infra/session-cost-usage.ts aggregates per-session costs from JSONL transcripts
Gateway methods usage.cost and sessions.usage expose this data to OpenClaw's own UI

But agent.wait doesn't include any of it:

// agent-wait-dedupe.ts L3-7
export type AgentWaitTerminalSnapshot = {
  status: "ok" | "error" | "timeout";
  startedAt?: number;
  endedAt?: number;
  error?: string;
};

// agent.ts L790-796
respond(true, {
  runId,
  status: snapshot.status,
  startedAt: snapshot.startedAt,
  endedAt: snapshot.endedAt,
  error: snapshot.error,
});

Proposed Change

Extend AgentWaitTerminalSnapshot to include an optional meta field:

export type AgentWaitTerminalSnapshot = {
  status: "ok" | "error" | "timeout";
  startedAt?: number;
  endedAt?: number;
  error?: string;
  meta?: {
    usage?: {
      input: number;
      output: number;
      cacheRead?: number;
      cacheWrite?: number;
    };
    lastCallUsage?: {
      input: number;
      output: number;
    };
    costUsd?: number;
    provider?: string;
    model?: string;
  };
};

At the end of an agent run, accumulate session usage from the JSONL transcript (the infrastructure for this already exists in session-cost-usage.ts) and attach it to the terminal snapshot before sending the agent.wait response.

Why This Matters

Paperclip's OpenClaw Gateway adapter (@paperclipai/adapter-openclaw-gateway) already has parseUsage() ready to consume usage, costUsd, provider, and model from response metadata — it just gets undefined today.
Paperclip PR #884 fixes the meta extraction priority chain on the consuming side.
Paperclip PR #949 (merged) added full billing ledger support including biller vs provider distinction, budget enforcement, and cost UI — all waiting for upstream cost data.
Without this, OpenClaw agents in Paperclip show $0 spend and budget enforcement never triggers.

Scope

~30-40 lines across:

agent-wait-dedupe.ts — type extension
Agent run lifecycle — accumulate usage from session transcript at run completion
agent.ts — include meta in the agent.wait response

Backward compatible — meta is optional, existing consumers ignore unknown fields.

Paperclip PR #884 — consuming side meta extraction fix
Paperclip PR #949 — billing ledger and cost control plane
Paperclip PR #386 — heartbeat cost recording fix (merged in v0.3.1)

extent analysis

Fix Plan

To fix the issue, we need to extend the AgentWaitTerminalSnapshot type and accumulate session usage from the JSONL transcript at the end of an agent run. Here are the concrete steps:

Extend the AgentWaitTerminalSnapshot type in agent-wait-dedupe.ts to include an optional meta field:

export type AgentWaitTerminalSnapshot = {
  status: "ok" | "error" | "timeout";
  startedAt?: number;
  endedAt?: number;
  error?: string;
  meta?: {
    usage?: {
      input: number;
      output: number;
      cacheRead?: number;
      cacheWrite?: number;
    };
    lastCallUsage?: {
      input: number;
      output: number;
    };
    costUsd?: number;
    provider?: string;
    model?: string;
  };
};

Accumulate session usage from the JSONL transcript at the end of an agent run using the existing infrastructure in session-cost-usage.ts. For example:

const sessionUsage = await loadSessionCostSummary(runId);
const meta = {
  usage: sessionUsage.usage,
  costUsd: sessionUsage.costUsd,
  provider: sessionUsage.provider,
  model: sessionUsage.model,
};

Include the meta field in the agent.wait response in agent.ts:

respond(true, {
  runId,
  status: snapshot.status,
  startedAt: snapshot.startedAt,
  endedAt: snapshot.endedAt,
  error: snapshot.error,
  meta,
});

Verification

To verify that the fix worked, you can check the agent.wait response for the presence of the meta field and its contents. You can also test the consuming side, such as the Paperclip adapter, to ensure that it can correctly extract and use the usage and cost data from the response metadata.

Extra Tips

Make sure to handle cases where the meta field is not present or is empty, to ensure backward compatibility with existing consumers.
Consider adding logging or monitoring to track the usage and cost data being sent in the agent.wait response, to help with debugging and troubleshooting.

Vote matrix · Quick signals

Works

Did the solution work? Tap to confirm.

Easy Fix

Was it a quick fix?

Time Saver

Did it save you time?

Blocking

Was it severely blocking?

Common Issue

Are others likely hitting this too?

Flaky / Intermittent

Is it intermittent?

Verified / Reproducible

Can you reproduce it reliably?

#api #ssr #installation #tensor shape #autograd error #cache error #pipeline error #runtime error #dependency conflict #environment setup

Still need to ship something?

×6

Another batch ranked right after the header list — different links, same matching logic.

openclaw - ✅(Solved) Fix feat: include usage/cost metadata in agent.wait terminal response [2 pull requests, 2 comments, 2 participants]

Recommended Tools

GitHub issue graph ai analysis

Error Message

Root Cause

Fix Action

Fix / Workaround

PR fix notes

PR #884: feat: openclaw-gateway adapter — include contextUsagePercent in run result metadata (ALE-1129)

Description (problem / solution / changelog)

Summary

Changes

packages/adapters/openclaw-gateway/src/server/execute.ts

server/src/__tests__/openclaw-gateway-adapter.test.ts

Test results

Changed files

PR #949: feat(costs): add billing, quota, and budget control plane

Description (problem / solution / changelog)

Summary

Verification

Changed files

Code Example

Summary

The Gap

Proposed Change

Why This Matters

Scope

Related

extent analysis

Fix Plan

Verification

Extra Tips

Still need to ship something?

RELATED_DISCOVERY

TRENDING

`packages/adapters/openclaw-gateway/src/server/execute.ts`

`server/src/tests/openclaw-gateway-adapter.test.ts`