Best possible solution:

Land this or an equivalent narrow persisted leaf-removal fix after redacted real behavior proof is added and focused embedded-runner validation stays green.

Do we have a high-confidence way to reproduce the issue?

Yes, source-reproducible: a file-backed SessionManager with a trailing user leaf reaches the orphan-repair block, and current main only calls branch() or resetLeaf() before rebuilding active messages. I did not run a live OpenClaw session in this read-only review.

Is this the best way to solve the issue?

Yes for the code direction, but not yet for merge readiness. Rewriting the file-backed SessionManager entries while retaining the branch/reset fallback is the narrowest maintainable fix I saw, but the external PR still needs real behavior proof.

What I checked:

• current_main_in_memory_only: Current main repairs a trailing user leaf by calling sessionManager.branch(...) or sessionManager.resetLeaf() and rebuilding active in-memory messages; this block does not remove the persisted file entry or call a rewrite hook. (src/agents/pi-embedded-runner/run/attempt.ts:3197, 333f65fc8a12)
• helper_contract_requests_removal: mergeOrphanedTrailingUserPrompt returns removeLeaf: true for empty, already-queued, and merged orphaned user content, so the runtime path intentionally removes the active leaf after carrying or dropping the prompt text. (src/agents/pi-embedded-runner/run/attempt.prompt-helpers.ts:437, 333f65fc8a12)
• current_tests_cover_merge_not_persistence: Current tests cover single-leaf merge and removeLeaf behavior, but current main has no removeSessionManagerLeafEntry helper or persisted JSONL rewrite coverage for this orphan-user repair path. (src/agents/pi-embedded-runner/run/attempt.test.ts:689, 333f65fc8a12)
• dependency_contract_checked: OpenClaw pins @earendil-works/pi-coding-agent to 0.74.0; the upstream v0.74.0 SessionManager keeps private fileEntries, byId, and leafId, _rewriteFile() rewrites fileEntries, and branch()/resetLeaf() only move leafId. (package.json:1759, 333f65fc8a12)
• pr_implements_narrow_rewrite_path: The PR adds removeSessionManagerLeafEntry, which removes the leaf entry from fileEntries, deletes it from byId, moves leafId to the parent, calls _rewriteFile(), and falls back to branch/reset when rewrite internals are unavailable. (src/agents/pi-embedded-runner/run/attempt.ts:563, e42bceefd658)
• real_behavior_proof_missing: The PR body lists lint, a focused Vitest command, and tsgo:test:src; the author follow-up also states the current blocker is the real behavior proof gate, not the fix direction. (2b4518a0d0c2)

Likely related people:

• vincentkoc: Commit fcae3bf9433b98b22c1d995644e9e33f0621d8f8 added the active-turn queued user prompt preservation helper and tests in the same embedded-runner path this PR extends. (role: introduced current orphan-merge behavior; confidence: high; commits: fcae3bf9433b; files: src/agents/pi-embedded-runner/run/attempt.ts, src/agents/pi-embedded-runner/run/attempt.prompt-helpers.ts, src/agents/pi-embedded-runner/run/attempt.test.ts)
• steipete: Recent GitHub path history and local blame show Peter Steinberger repeatedly touching attempt.ts and adjacent runtime-state/transcript areas, including current-line snapshot ownership in this checkout. (role: recent embedded-runner and transcript-state contributor; confidence: medium; commits: e79e5dbbdf5a, c6ddb1afb7bc, 694ca50e9775; files: src/agents/pi-embedded-runner/run/attempt.ts, src/agents/pi-embedded-runner/run/attempt.prompt-helpers.ts, src/agents/pi-embedded-runner/transcript-file-state.ts)
• jalehman: Recent session repair/replay work touches the same embedded-runner and transcript consistency area, so this is a useful secondary routing candidate for review coordination. (role: adjacent session repair contributor; confidence: medium; commits: 4bfd7416f0f9, 75e7fc97f804; files: src/agents/pi-embedded-runner/run/attempt.ts, src/agents/session-file-repair.ts)

Remaining risk / open question:

• The contributor has not supplied after-fix evidence from a real OpenClaw setup, which is required before merge for an external bug-fix PR.
• The implementation intentionally relies on private SessionManager internals from the pinned Pi dependency; that matches existing local repair patterns but remains coupled to upstream internals.
• The broader stale-reply transcript cluster has adjacent open work, so this PR should stay scoped to persisted orphaned-user leaf removal rather than claiming to solve every stale-context symptom.
• No live OpenClaw run or test command was executed in this read-only review; the reproduction assessment is source-backed plus issue discussion evidence.

Codex review notes: model gpt-5.5, reasoning high; reviewed against 333f65fc8a12.