Best possible solution:

Keep the PR open under the tracker, rebase it onto the current stack, fix the storage-safety blockers, and add redacted real SQLite run-target proof before any merge decision.

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

Yes. Source inspection of PR head shows the explicit compaction file target being dropped when a session key exists, SQLite reads using a schema-creating opener, and session cleanup deleting rows before an enforced export receipt.

Is this the best way to solve the issue?

No. The SQLite adapter seam is the intended owner boundary, but the current draft is not the best mergeable shape until it preserves current target behavior, adds no-create read paths, guards transcript deletion, rebases, and provides real proof.

Full review comments:

• [P1] Preserve explicit compact session files — src/context-engine/delegate.ts:67
delegateCompactionToRuntime now omits params.sessionFile whenever any session key is present. A ContextEngine caller that supplies sessionKey plus an explicit sessionFile but no fully resolved runtime target will be resolved through the configured/default store instead of compacting the caller's transcript file.
  Confidence: 0.9
• [P1] Use no-create opens for SQLite read helpers — src/config/sessions/session-accessor.sqlite.ts:104
  The SQLite read helpers call the normal agent database opener, which creates directories/files, applies schema DDL, caches the handle, and registers the DB. Doctor/startup preview reads need a no-create/read-only path that returns empty results when the database is absent.
  Confidence: 0.9
• [P1] Require export receipts before transcript deletes — src/config/sessions/session-accessor.sqlite.ts:751
  Deleting a sessions row cascades transcript_events, but lifecycle cleanup only counts transcript state as archived and does not require a durable export receipt first. Once SQLite transcripts are live, this can silently drop conversation history.
  Confidence: 0.88

Overall correctness: patch is incorrect
Overall confidence: 0.88

AGENTS.md: found and applied where relevant.

Codex review notes: model internal, reasoning high; reviewed against 3ab8d6aa609a.

Label justifications:

• P2: This is normal-priority storage migration work with significant future blast radius but no immediate production storage flip in the draft PR.
• merge-risk: 🚨 compatibility: The PR changes embedded/CLI/compaction runtime session parameter contracts and SQLite open/schema behavior that can affect existing callers and upgrades.
• merge-risk: 🚨 session-state: The diff rewires session identity, compaction target resolution, and transcript/session persistence paths.
• merge-risk: 🚨 security-boundary: SQLite transcript rows can contain sensitive conversation history, and lifecycle deletion/export ordering affects the privacy and retention boundary.
• rating: 🧂 unranked krab: Overall readiness is 🧂 unranked krab; proof is 🧂 unranked krab and patch quality is 🧂 unranked krab.
• status: 📣 needs proof: The PR needs real behavior proof before ClawSweeper can clear the contributor ask. Needs real behavior proof before merge: The PR body lists focused tests and autoreview only; it still needs redacted terminal, log, live-output, recording, or linked-artifact proof from a real SQLite run-target path before merge. After adding proof, update the PR body; ClawSweeper should re-review automatically. If it does not, the PR author or someone with repository write access can comment @clawsweeper re-review.

PR surface:

Source +2956, Tests +2164, Generated +220. Total +5340 across 44 files.

Security concerns:

• [medium] Sensitive transcript retention guard missing — src/config/sessions/session-accessor.sqlite.ts:751
  The SQLite cleanup path can delete session rows under cascading transcript foreign keys before a durable transcript export receipt is enforced, risking loss of sensitive conversation history without the required retention artifact.
  Confidence: 0.86

What I checked:

• Repository policy applied: Root and scoped AGENTS.md files were read from disk; storage/session compatibility, persistent data migration, plugin-SDK boundary, and proof rules apply to this PR. (AGENTS.md:30, 3ab8d6aa609a)
• Live PR state: GitHub reports this PR open, draft, conflicting/dirty, head 1c59b42, with maintainer, proof-needed, and storage/session/security merge-risk labels. (1c59b42c0362)
• Current-main seam state: Current main has the file-backed run session target seam, but not the SQLite runtime target branch added by this PR. (src/agents/run-session-target.ts:20, 3ab8d6aa609a)
• Explicit compaction target regression: PR head drops runtimeContext.sessionFile and only forwards params.sessionFile when no session key exists, changing explicit compaction-file targeting. (src/context-engine/delegate.ts:67, 1c59b42c0362)
• Current-main compaction comparison: Current main unconditionally forwards params.sessionFile into compactEmbeddedAgentSessionDirect, so the PR changes existing compaction target behavior. (src/context-engine/delegate.ts:58, 3ab8d6aa609a)
• SQLite reads use creating opener: The PR's SQLite read helpers call openOpenClawAgentDatabase, whose opener creates directories, opens the SQLite file, ensures schema, caches/registers the DB, and applies permissions. (src/config/sessions/session-accessor.sqlite.ts:104, 1c59b42c0362)

Likely related people:

• jalehman: Authored the canonical tracker, this PR head, the merged file-backed embedded run target seam, the merged session accessor seam, and current-main adjacent session metadata accessor work. (role: stack coordinator and recent area contributor; confidence: high; commits: 1c59b42c0362, 0dfa22c6e0fc, ef47dd610c87; files: src/agents/run-session-target.ts, src/config/sessions/session-accessor.ts, src/context-engine/delegate.ts)
• Shakker: Current-main blame for the central run-session-target and delegate files points at a broad recent commit that carried this area forward after the stack moved. (role: recent current-main committer; confidence: medium; commits: 8ecbf83c678c; files: src/agents/run-session-target.ts, src/context-engine/delegate.ts)
• Vincent Koc: Committed the revert that deferred the earlier SQLite session metadata migration, which the tracker cites as lessons for this staged reland. (role: adjacent storage migration contributor; confidence: medium; commits: 7f1d82ab2518; files: src/state/openclaw-agent-db.ts, src/config/sessions/store.ts)
• Peter Steinberger: History for the context-engine delegate includes earlier runtime import and compaction-boundary maintenance that is adjacent to this PR's delegate changes. (role: adjacent context-engine contributor; confidence: low; commits: 87fa88ac3d35; files: src/context-engine/delegate.ts)

• 🦀 challenger crab: rare, exceptional readiness with strong proof, clean implementation, and convincing validation.
• 🦞 diamond lobster: very strong readiness with only minor maintainer review expected.
• 🐚 platinum hermit: good normal PR, likely mergeable with ordinary maintainer review.
• 🦐 gold shrimp: useful signal, but proof or patch confidence is still limited.
• 🦪 silver shellfish: thin signal; proof, validation, or implementation needs work.
• 🧂 unranked krab: not merge-ready because proof is missing/unusable or there are serious correctness or safety concerns.
• 🌊 off-meta tidepool: rating does not apply to this item.

Shiny media proof means a screenshot, video, or linked artifact directly shows the changed behavior. Runtime, network, CSP, and security claims still need visible diagnostics.

• ClawSweeper keeps one durable marker-backed review comment per issue or PR.
• Re-runs edit this comment so the latest verdict, findings, and automation markers stay together instead of adding duplicate bot comments.
• A fresh review can be triggered by eligible @clawsweeper re-review comments, exact-item GitHub events, scheduled/background review runs, or manual workflow dispatch.
• PR/issue authors and users with repository write access can comment @clawsweeper re-review or @clawsweeper re-run on an open PR or issue to request a fresh review only.
• Maintainers can also comment @clawsweeper review to request a fresh review only.
• Fresh-review commands do not start repair, autofix, rebase, CI repair, or automerge.
• Maintainer-only repair and merge flows require explicit commands such as @clawsweeper autofix, @clawsweeper automerge, @clawsweeper fix ci, or @clawsweeper address review.
• Maintainers can comment @clawsweeper explain to ask for more context, or @clawsweeper stop to stop active automation.