PR #62536: fix(logging): route subagent announce output through subsystem logger

• Repository: openclaw/openclaw
• Author: irresi
• State: open | merged: False
• Link: https://github.com/openclaw/openclaw/pull/62536

Description (problem / solution / changelog)

Summary

• Problem: Gateway file logs record full subagent announce/completion assistant response bodies as info lines, polluting operational logs with chat content.
• Why it matters: Logs become noisy and hard to debug; large multi-line summaries bloat file size; user-facing text mixed into system logs is misleading and a potential privacy concern.
• What changed: Announce delivery output now routes through createSubsystemLogger("agents/announce") — metadata at info, full body at debug. Announce-related error/warning logging in subagent-announce.ts and subagent-announce-delivery.ts also moved from defaultRuntime to the subsystem logger.
• What did NOT change (scope boundary): Normal (non-announce) agent delivery logging, channel delivery, session transcript writes, or any LLM behavior.

Change Type (select all)

• Bug fix
• Feature
• Refactor required for the fix
• Docs
• Security hardening
• Chore/infra

Scope (select all touched areas)

• Gateway / orchestration
• Skills / tool execution
• Auth / tokens
• Memory / storage
• Integrations
• API / contracts
• UI / DX
• CI/CD / infra

Linked Issue/PR

• Closes #62500
• Related #
• This PR fixes a bug or regression

Root Cause (if applicable)

• Root cause: deliverAgentCommandResult() in delivery.ts calls runtime.log(output) for announce run payloads. In the gateway process, runtime.log() → console.log() → enableConsoleCapture() intercepts all console.log calls and forwards them to the tslog root logger at info level via getLogger().info(formatted), which writes to the rolling date file.
• Missing detection / guardrail: No distinction between announce-internal output and normal agent output in the delivery logging path. Announce runs use the same logPayload() / logNestedOutput() path as user-facing agent runs.
• Contributing context (if known): The announce delivery files (subagent-announce.ts, subagent-announce-delivery.ts) had no subsystem logger at all — errors went through defaultRuntime.error() which also hits the console capture path.

Regression Test Plan (if applicable)

• Coverage level that should have caught this:
  • Unit test
  • Seam / integration test
  • End-to-end test
  • Existing coverage already sufficient
• Target test or file: src/agents/command/delivery.test.ts
• Scenario the test should lock in: When inputProvenance.sourceTool === "subagent_announce", runtime.log() must NOT receive the full payload body; instead announceLog.info() receives metadata and announceLog.debug() receives the full body.
• Why this is the smallest reliable guardrail: Tests the exact decision point in logPayload() without requiring a live gateway or real subagent completion.
• Existing test that already covers this (if any): None — this gap is why the log pollution went undetected.
• If no new test is added, why not: A new test is added.

User-visible / Behavior Changes

• openclaw logs output no longer includes full subagent completion prose at default log level.
• Announce metadata (session, runId, payload size) still appears at info.
• Full announce body is available at debug level for troubleshooting.

Diagram (if applicable)

Before:
[subagent completes] → deliverAgentCommandResult → runtime.log(fullBody)
  → console.log(fullBody) → enableConsoleCapture → getLogger().info(fullBody)
  → openclaw-YYYY-MM-DD.log (info, no subsystem, full chat text)

After:
[subagent completes] → deliverAgentCommandResult → announceLog.info(metadata)
  → logToFile(fileLogger, "info", metadata)    ← clean operational log
  → announceLog.debug(fullBody)
  → logToFile(fileLogger, "debug", fullBody)   ← only at debug level

Security Impact (required)

• New permissions/capabilities? No
• Secrets/tokens handling changed? No
• New/changed network calls? No
• Command/tool execution surface changed? No
• Data access scope changed? No

Repro + Verification

Environment

• OS: macOS
• Runtime/container: Node 22+ / pnpm workspace
• Model/provider: Any
• Integration/channel (if any): Any channel with subagent spawning

Steps

1. Start a gateway with logging.level: debug or default info
2. Spawn a subagent that produces a multi-line completion summary
3. Let the subagent complete and announce back to the parent session
4. Inspect gateway logs via openclaw logs --plain --local-time

Expected

• Gateway file logs show only announce metadata (session, runId, chars) at info level
• Full completion body appears only at debug level

Actual

• Before this fix: full completion body appears at info level with no subsystem tag
• After this fix: metadata-only at info, full body at debug

Evidence

• Failing test/log before + passing after
• Trace/log snippets
• Screenshot/recording
• Perf numbers (if relevant)

Human Verification (required)

• Verified scenarios:
  • New unit test confirms announce payloads route through subsystem logger instead of runtime.log
  • All 14 existing announce-related tests pass (delivery + announce + announce-delivery)
  • pnpm check lint/format clean
• Edge cases checked:
  • Non-announce agent runs still use the original runtime.log() path (unchanged)
  • opts.json mode still uses runtime.log(JSON.stringify(...)) (unchanged)
  • Nested non-announce runs still use logNestedOutput() (unchanged)
• What you did not verify:
  • Full end-to-end gateway run with a real subagent completion (verified via code path analysis and unit test)

Review Conversations

• I replied to or resolved every bot review conversation I addressed in this PR.
• I left unresolved only the conversations that still need reviewer or maintainer judgment.

Compatibility / Migration

• Backward compatible? Yes
• Config/env changes? No
• Migration needed? No

Risks and Mitigations

• Risk: Announce errors previously written via defaultRuntime.error() now go through subsystem logger, which uses rawConsole instead of patched console.
  • Mitigation: Both paths write to stderr and file log at error level. The only difference is the subsystem logger adds a [agents/announce] tag, which is an improvement for filtering.

AI-assisted: Built with Claude. All changes reviewed and understood. 14/14 related tests passing.

Changed files

• src/agents/command/delivery.test.ts (modified, +88/-0)
• src/agents/command/delivery.ts (modified, +16/-0)
• src/agents/subagent-announce-delivery.ts (modified, +5/-3)
• src/agents/subagent-announce.ts (modified, +19/-4)