Summary

• Problem: long-lived main sessions could crash before provider dispatch when compaction token estimation hit malformed replay history and estimateTokens() read missing .length fields.
• Why it matters: once a session contained one malformed history block, every later prompt attempt could fail in pre-prompt compaction, making the session effectively unrecoverable.
• What changed: added a guarded estimateMessageTokens() path in src/agents/compaction.ts and switched preemptive compaction plus embedded compaction metrics/sanity checks to reuse it.
• What did NOT change (scope boundary): this PR does not redesign replay-history normalization or patch @mariozechner/pi-coding-agent; it only hardens OpenClaw’s local compaction estimation path.

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 Main session prompt crash: Cannot read properties of undefined (reading 'length') in compaction token estimation #63612

• This likely also addresses the malformed-history / reading 'length' Telegram manifestation discussed in [Bug]: Telegram direct lane repeatedly throws 'Cannot read properties of undefined (reading "length")' on 2026.4.9 #64053, and may partially reduce the Telegram crash facet mentioned in [Bug]: 2026.4.9 multi-channel degradation on Linux: Discord direct-session overflow + slash lag + Telegram 'reading length' lane crashes #64034, but it does not address the broader Discord overflow/lag symptoms tracked there.

• This PR fixes a bug or regression

Root Cause (if applicable)

• Root cause: compaction-side token estimation assumed replayed message blocks always had fully normalized shapes, but malformed assistant/toolResult blocks could still reach estimation and trigger unchecked .length reads.
• Missing detection / guardrail: OpenClaw had replay sanitization and some downstream try/catch sites, but no shared safe estimator for all compaction-related estimateTokens() call sites.
• Contributing context (if known): long-lived main sessions exercise pre-prompt compaction on every turn, so one malformed history block could repeatedly fail the recovery path itself.

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/compaction.test.ts, src/agents/pi-embedded-runner/run/preemptive-compaction.test.ts
• Scenario the test should lock in: malformed assistant/toolResult history blocks do not throw during token estimation or pre-prompt compaction checks.
• Why this is the smallest reliable guardrail: the crash happens in pure estimation logic before provider dispatch, so unit coverage at the estimation and precheck seam is enough to lock in the failure mode.
• Existing test that already covers this (if any): none
• If no new test is added, why not: N/A

User-visible / Behavior Changes

Long-lived sessions with malformed replay history now fail soft in compaction token estimation instead of crashing before reply generation.

Diagram (if applicable)

Before:
[new user turn] -> [pre-prompt compaction estimation] -> [throws on malformed block] -> [session cannot reply]

After:
[new user turn] -> [guarded token estimation] -> [invalid block counted as 0/safe fallback] -> [reply flow continues]

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)
• If any Yes, explain risk + mitigation:

Repro + Verification

Environment

• OS: macOS
• Runtime/container: local Node 22+/pnpm workspace
• Model/provider: N/A for repro; crash occurs before provider dispatch
• Integration/channel (if any): embedded main session
• Relevant config (redacted): default compaction path with long-lived session history

Steps

1. Build or replay a session history containing malformed assistant/toolResult blocks.
2. Trigger a new turn that runs pre-prompt compaction estimation.
3. Observe the behavior before and after the patch.

Expected

• Token estimation tolerates malformed blocks and the session continues.

Actual

• Before this fix, estimation could throw Cannot read properties of undefined (reading 'length') and abort the reply before provider dispatch.

Evidence

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

Human Verification (required)

• Verified scenarios: ran pnpm test src/agents/compaction.test.ts, pnpm test src/agents/pi-embedded-runner/run/preemptive-compaction.test.ts, and pnpm check.
• Edge cases checked: malformed assistant content entries, missing assistant content arrays, malformed toolResult content during pre-prompt estimation.
• What you did not verify: no live reproduction against a real damaged long-lived session transcript.

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)
• If yes, exact upgrade steps:

Risks and Mitigations

• Risk: local fallback token estimation may slightly differ from upstream estimateTokens() for malformed messages.
  • Mitigation: fallback is only used on malformed inputs where the previous behavior was to throw; valid messages still use upstream estimation first.