ewilliams28/openclaw
1
0
Fork 0
mirror of https://github.com/openclaw/openclaw.git synced 2026-08-12 21:53:00 -06:00
Code Issues Packages Projects Releases Wiki Activity

feat(agents): report per-run stats (code-mode engagement, round trips, cost) in agent JSON envelopes (#114688)

Browse Source 
* feat(agents): add per-run stats to embedded agent run meta

Adds codeModeEngaged, assistantTurns, bridgeCalls, and costUsd to
EmbeddedAgentRunMeta.agentMeta and mirrors them on the agent exec --json
envelope. Code-mode engagement is stamped from the tool-surface truth,
round trips accumulate across attempts beside usage, bridge counts come
from the run's tool-search catalog counters, and cost reuses the shared
model pricing helpers (cache tiers included, omitted without cost data).

* fix(agents): accumulate bridge call counts across run attempts

Attempt cleanup clears the per-attempt tool-search catalog, so retries and
fallbacks discarded earlier bridge counts. Fold each attempt's bridgeCalls
into the run accumulator beside assistantTurns and stamp the cumulative
totals into agentMeta, matching the documented per-run contract.
This commit is contained in:
Peter Steinberger
2026-07-27 23:48:01 -04:00
committed by GitHub
parent e1ced6de50
commit cf4cb0ac85
19 changed files with 374 additions and 3 deletions
Download Patch File Download Diff File Expand all files Collapse all files
docs/tools/code-mode.md
+18
View File
@@ -971,6 +971,24 @@ breakdown (`openclaw`/`mcp`/`client` counts), cumulative search/describe/call
counts for the run's catalog, and the model-visible tool names (`exec`,
`wait`, and retained direct-only tools).
The run metadata (`meta.agentMeta` in `openclaw agent --json`, mirrored on the
`agent exec --json` envelope) adds per-run stats:
- `codeModeEngaged`: `true` only when code mode actually owned the model tool
surface. This is the reliable engagement signal — do not infer engagement
from config or tool names: the shell tool is also named `exec`, the
`"auto"` tier engages per model capability, and a model routed through a
native harness surface (for example OpenAI-family models on their harness)
reports `codeModeEngaged: false` even with `tools.codeMode.enabled=true`,
making the silent no-op observable.
- `assistantTurns`: completed assistant/provider round trips across the run.
- `bridgeCalls`: the run's cumulative inner bridge counts
(`{ search, describe, call }`). These calls never reach the provider;
provider-visible outer tool calls remain in `meta.toolSummary.calls`.
- `costUsd`: estimated USD cost from the run's accumulated usage and the
model's cost config (cache read/write tiers included); omitted when the
model has no cost data.
Telemetry must not include secrets, raw environment values, or unredacted
tool inputs beyond existing OpenClaw trajectory policy.