Workflow Auto.md after Compaction

[lavish steppe]

Hey, I found the following in my OpenClaw installation (v2026.2.26) at dist/pi-embedded-CQnl8oWA.js:
js// src/auto-reply/reply/post-compaction-audit.ts
const DEFAULT_REQUIRED_READS = ["WORKFLOW_AUTO.md", /memory/\d{4}-\d{2}-\d{2}.md/];
After every compaction, my agent gets a message starting with ⚠️ Post-Compaction Audit: The following required startup files were not read... referencing WORKFLOW_AUTO.md. Is this a legitimate built-in OpenClaw feature? I previously reported it as a prompt injection and was told it's not from OpenClaw — but the source code suggests otherwise. Can someone clarify?

[nova crown]

To help others find answers, you can mark your question as solved via Right click solution message -> Apps -> ✅ Mark Solution

[limpid turtle]

Yes — that warning is coming from OpenClaw itself, not a prompt injection.

In v2026.2.26 there’s a built-in “post-compaction read audit” in the auto-reply pipeline:

• The default required reads are hard-coded as:

• WORKFLOW_AUTO.md

• memory/YYYY-MM-DD.md (regex for daily memory files)
  Source: src/auto-reply/reply/post-compaction-audit.ts lines 4–8

• After an auto-compaction happens, the reply runner sets a “pending audit” flag, then on the next turn it:

1. reads the last ~100 lines of the session JSONL
2. extracts any read tool calls + their path/file_path
3. checks whether those required files were read
4. if not, it enqueues the exact system event you’re seeing
   Source: src/auto-reply/reply/agent-runner.ts lines 724–741

So if you’re seeing “⚠️ Post-Compaction Audit: The following required startup files were not read…”, it specifically means: after the compaction, the agent did not issue read tool calls for those paths (in the recent session log).