[AI-assisted] fix(agents): resolve CLI runtime in preflight + memory-flush gates by adele-with-a-b · Pull Request #86224 · openclaw/openclaw

adele-with-a-b · 2026-05-24T23:15:27Z

Summary

runPreflightCompactionIfNeeded and runMemoryFlushIfNeeded in src/auto-reply/reply/agent-runner-memory.ts gate on isCliProvider(params.followupRun.run.provider, params.cfg). isCliProvider only returns true when the passed provider id is itself a key in agents.defaults.cliBackends. For an agent configured with a primary model on a canonical provider (e.g. anthropic/claude-opus-4-7) that resolves through a CLI runtime via models.providers.<provider>.agentRuntime.id (or per-agent / per-model agentRuntime policy), the gate evaluates isCli=false and preflight runs as if the agent were embedded — even though the conversation transcript and tokens live in the CLI runtime's own session, not in the OC transcript file.

This PR mirrors the resolution shape already used by the runtime dispatch path: resolve the CLI execution provider via resolveCliRuntimeExecutionProvider first, then ask isCliProvider about the resolved id. The two preflight/flush gates predate that resolve and were never updated.

Existing pattern

src/auto-reply/reply/agent-runner-execution.ts resolves the CLI execution provider before gating on isCliProvider:

const cliExecutionProvider =
  resolveCliRuntimeExecutionProvider({
    provider,
    cfg: runtimeConfig,
    agentId: params.followupRun.run.agentId,
    modelId: model,
  }) ?? provider;

if (isCliProvider(cliExecutionProvider, runtimeConfig)) {
  // ... CLI dispatch path
}

The fix in this PR uses the same shape at both agent-runner-memory.ts call sites.

Reproduction context

A long-lived agent routed through a CLI runtime via agentRuntime.id accumulates context inside the CLI's own session. The model reports a large promptTokens to OC, which OC stores as entry.totalTokens. Once that crosses the preflight threshold (contextWindow - reserveTokensFloor - softThresholdTokens), preflight fires and tries to compact the OC-side transcript file. For agents whose real conversation lives outside the OC transcript file (the CLI runtime's own session), OC's transcript file only contains delivery-mirror records, and the compaction safeguard's containsRealConversationMessages filter strips those, returning compacted:false reason:"no real conversation messages". Preflight then throws Preflight compaction required but failed: no real conversation messages, surfacing as a generic dispatch error.

The runtime dispatch path doesn't hit this because it had already been migrated to resolveCliRuntimeExecutionProvider. The preflight + flush gates had not.

Verification

pnpm tsgo:core and pnpm tsgo:core:test green.
pnpm exec vitest run src/auto-reply/reply/agent-runner-memory.test.ts: 26 passed (existing 24 + 2 new). Both new tests fail without the source fix and pass with it.
pnpm exec oxfmt --check clean on both touched files.

The two new tests configure agents.defaults.cliBackends["claude-cli"] plus models.providers.anthropic.agentRuntime.id = "claude-cli" and assert that a followupRun with provider="anthropic" short-circuits both runMemoryFlushIfNeeded (returns the entry without invoking runEmbeddedPiAgent) and runPreflightCompactionIfNeeded (returns the entry without invoking compactEmbeddedPiSession).

Real behavior proof

Behavior addressed: Telegram dispatch on a long-lived claude-cli-routed Anthropic agent surfaced "Something went wrong while processing your request." The underlying error in the gateway log was Preflight compaction required but failed: no real conversation messages. Root cause: runPreflightCompactionIfNeeded evaluated isCliProvider("anthropic", cfg) (false because cliBackends keys on claude-cli, not anthropic), so preflight ran on agents whose real conversation lives in claude-cli's session JSONL — not in OC's transcript file (which holds only delivery-mirror records for that surface). The compactor's containsRealConversationMessages filter then strips delivery-mirror records and returns compacted:false, so preflight throws.

Real environment tested: A live OpenClaw stack on macOS 14 with three claude-cli-routed Anthropic agents wired to Telegram. Anthropic configured as primary; routes through auth.profiles["anthropic:claude-cli"] so dispatch resolves to the claude-cli runtime. Six historical failures observed today across one of the three agents over ~2h before manual recovery (delete the broken session-store binding + archive the all-delivery-mirror transcript file).

Exact steps or command run after this patch:

Patched ~/.local/bin/openclaw-cli-runtime-preflight-gate-patch.py applied to the brew-installed OC dist (mirrors this PR's source change exactly — same anchored swap of isCliProvider(params.followupRun.run.provider, ...) for resolve-then-isCli).
Restarted the gateway: openclaw gateway restart --safe.
Enabled debug logging via OPENCLAW_LOG_LEVEL=debug so the preflightCompaction check / memoryFlush check verbose lines emit.
Sent one Telegram inbound to a claude-cli-routed agent on the same session-key shape that previously failed.
Sent one Telegram inbound to a different claude-cli-routed agent for cross-agent confirmation.
Inspected the gateway log for the verbose isCli= value at both gates.
Reverted OPENCLAW_LOG_LEVEL=debug and restarted the gateway.

Evidence after fix:

Pre-fix gateway log (six historical failures across ~2h, redacted):

[compaction-diag] end runId=<redacted> sessionKey=<redacted-agent-session-key> diagId=cmp-<redacted>
  trigger=budget provider=anthropic/claude-opus-4-7 attempt=1 maxAttempts=1
  outcome=failed reason=unknown detail=No_API_key_found_for_provider_anthropic_._Auth_store:_<path>
  durationMs=7540
[compaction] skipping — no real conversation messages (sessionKey=<redacted>)
message dispatch completed: ... outcome=error duration=17203ms
  error="Error: Preflight compaction required but failed: no real conversation messages"

Pre-fix verbose memoryFlush check (debug log, before patch was applied; raw provider="anthropic" misclassified):

memoryFlush check: sessionKey=<redacted> tokenCount=51762 contextWindow=1048576 threshold=1024576
  isHeartbeat=false isCli=false
  persistedFresh=true ...

Post-fix verbose memoryFlush check (same env, same agent kind, debug log after patch + this PR's source change):

memoryFlush check: sessionKey=<redacted> tokenCount=48024 contextWindow=1048576 threshold=1024576
  isHeartbeat=false isCli=true
  ...

Post-fix preflight short-circuits at the gate before its own verbose log emits — the early-return at the resolved-runtime-isCli check fires before the logVerbose line on the run-preflight branch.

Observed result after fix: All three claude-cli-routed agents now resolve isCli=true at both runPreflightCompactionIfNeeded and runMemoryFlushIfNeeded. The Telegram dispatch path no longer hits the OC-transcript compaction path, no Preflight compaction required but failed errors, no user-visible "Something went wrong" message. Six previously-failing inbounds no longer reproduce on the affected session-key shape (verified by sending three test messages across two agents post-patch — all returned the agent's reply normally).

What was not tested: The auth-profile-pinned-to-direct-API-while-auth-order-prefers-CLI configuration was added in the follow-up commit (per Codex review) and is covered by the new regression tests but was not exercised live — the live test stack uses the simpler auth.profiles["anthropic:claude-cli"] mapping rather than a direct-API pin override. No Crabbox / Testbox run; the live verification was on a personal stack with the three claude-cli-routed agents reproducing the exact original failure shape.

clawsweeper · 2026-05-24T23:18:14Z

Codex review: needs changes before merge. Reviewed June 24, 2026, 12:22 PM ET / 16:22 UTC.

Summary
The PR resolves effective CLI runtime routing before preflight-compaction and memory-flush gates, with regression tests for model policy, auth-profile, and persisted runtime override cases.

PR surface: Source +71, Tests +508. Total +579 across 2 files.

Reproducibility: yes. Current main's memory gate helper does not call the resolver that dispatch uses for configured model policy and auth-profile CLI routing, and the PR body provides redacted live logs for the same failure shape.

Review metrics: none identified.

Root-cause cluster
Relationship: canonical
Canonical: #86224
Summary: This PR remains the canonical open item for aligning auto-reply memory gates with dispatch-style CLI runtime resolution; nearby merged work only covers subsets or adjacent preflight/memory behavior.

Members:

same_root_cause: fix(auto-reply): per-model runtime policy in memory-flush gates #86361 - Closed narrower PR for per-model runtime policy in the same memory gates; the discussion explicitly deferred to this PR's resolver-based shape.
partial_overlap: fix(agents): preserve compatible CLI session runtime pins #92476 - Merged compatible CLI session runtime pin support and current helper shape, but it does not cover configured model-policy or auth-order CLI routing.
partial_overlap: fix(agents): match runtime policy entries when session provider is empty #85970 - Merged runtime-policy resolver fix that this PR depends on, but it fixes empty-provider policy matching rather than the memory gate classification itself.
partial_overlap: fix(reply): don't wedge preflight compaction on benign skip reasons #86663 - Closed workaround for benign preflight skip reasons in the same symptom family, but it does not align the memory gates with dispatch routing.
adjacent_distinct: fix(reply): flush memory for pending in-turn auto-compaction even when tokens drop below threshold #87088 - Open memory-flush PR for in-turn compaction catch-up; it touches nearby memory flush behavior but not CLI runtime classification.

Proposal only: this assessment does not dispatch repair, suppress jobs, mutate sibling items, close, or merge anything.

Merge readiness
Overall: 🦐 gold shrimp
Proof: 🦞 diamond lobster
Patch quality: 🦐 gold shrimp
Result: needs maintainer review before merge.

Overall follows the weaker of proof and patch quality, so missing proof can cap an otherwise strong patch.

Rank-up moves:

[P2] Rebase onto current main and repair followupUsesCliRuntime instead of the old two call sites.
Rerun the focused memory-gate, dispatch, follow-up runner, and tsgo validation after the rebase.

Risk before merge

[P1] The PR head is conflicting and edits the older duplicated call-site shape; a manual conflict resolution that misses current main's followupUsesCliRuntime helper could leave configured model-policy and auth-order CLI routing broken in the memory gates.
[P1] These gates decide whether embedded compaction or memory flush mutates OpenClaw transcript/session state for a turn, so resolver divergence can still surface as auth-provider failures or session-state drift for CLI-owned conversations.

Maintainer options:

Repair the shared helper before merge (recommended)
Rebase onto current main and update followupUsesCliRuntime to use the same provider-compatible override plus auth-profile-aware CLI runtime resolution that dispatch uses.
Pause for owner boundary review
If maintainers want a different canonical helper boundary, pause the PR because the current conflicting head cannot be merged safely as-is.

Copy recommended automerge instruction

@clawsweeper automerge

Special instructions:
Rebase onto current main and replace followupUsesCliRuntime with dispatch-equivalent effective CLI runtime resolution, preserving compatible and stale session override behavior, model-policy routing, auth-profile pinning, auth.order CLI routing, and pi/openclaw cases with focused tests.

Next step before merge

[P2] A narrow automated repair can rebase the PR and move the resolver parity into the current shared helper with focused routing tests.

Security
Cleared: The diff changes TypeScript runtime classification logic and tests only, with no dependency, workflow, credential, package, download, or supply-chain surface added.

Review findings

[P2] Move resolver parity into the shared memory gate helper — src/auto-reply/reply/agent-runner-memory.ts:731-745

Review details

Best possible solution:

Rebase the PR and implement dispatch-equivalent CLI runtime classification once in current main's followupUsesCliRuntime helper, preserving provider-compatible override, model-policy, auth-profile, and pi/openclaw regression coverage.

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

Yes. Current main's memory gate helper does not call the resolver that dispatch uses for configured model policy and auth-profile CLI routing, and the PR body provides redacted live logs for the same failure shape.

Is this the best way to solve the issue?

No for the current branch as-is. The resolver-based direction is the right fix, but it needs to move into current main's shared followupUsesCliRuntime helper after rebasing.

Full review comments:

[P2] Move resolver parity into the shared memory gate helper — src/auto-reply/reply/agent-runner-memory.ts:731-745
Current main routes both preflight and flush through followupUsesCliRuntime, but this branch still adds resolver blocks at the old two call sites and GitHub reports the head as conflicting. Rebase and update the helper instead, or the conflict resolution can leave configured model-policy/auth-order CLI routing unfixed in the current memory gates.
Confidence: 0.87

Overall correctness: patch is incorrect
Overall confidence: 0.86

AGENTS.md: found and applied where relevant.

Codex review notes: model internal, reasoning high; reviewed against 3217165be770.

Label changes

Label justifications:

P1: The bug can break live agent replies by running embedded maintenance against sessions whose actual runtime and auth live in a CLI backend.
merge-risk: 🚨 auth-provider: The diff changes how auth-profile and provider routing are classified before choosing embedded maintenance versus CLI-owned execution.
merge-risk: 🚨 session-state: The diff controls whether preflight compaction and memory flush mutate OpenClaw transcript/session state for CLI-owned conversations.
rating: 🦐 gold shrimp: Overall readiness is 🦐 gold shrimp; proof is 🦞 diamond lobster and patch quality is 🦐 gold shrimp.
status: ⏳ waiting on author: ClawSweeper has contributor-facing work open and is waiting for author action. Sufficient (logs): The PR body includes redacted live gateway logs showing the pre-fix failure, post-fix isCli=true classification, and successful replies on a real claude-cli-routed setup.
proof: sufficient: Contributor real behavior proof is sufficient. The PR body includes redacted live gateway logs showing the pre-fix failure, post-fix isCli=true classification, and successful replies on a real claude-cli-routed setup.

Evidence reviewed

PR surface:

Source +71, Tests +508. Total +579 across 2 files.

View PR surface stats

Area	Files	Added	Removed	Net
Source	1	74	3	+71
Tests	1	508	0	+508
Docs	0	0	0	0
Config	0	0	0	0
Generated	0	0	0	0
Other	0	0	0	0
Total	2	582	3	+579

Acceptance criteria:

[P1] node scripts/run-vitest.mjs src/auto-reply/reply/agent-runner-memory.test.ts.
[P1] node scripts/run-vitest.mjs src/auto-reply/reply/agent-runner-execution.test.ts src/auto-reply/reply/followup-runner.test.ts.
[P1] pnpm tsgo:core.
[P1] pnpm tsgo:core:test.
[P1] git diff --check.

What I checked:

Root policy read: Root AGENTS.md was read completely; its ClawSweeper PR review policy applies because this change touches provider routing, auth, session state, and auto-reply memory gates. (AGENTS.md:1, 3217165be770)
Scoped policy read: The relevant scoped agents guide was read; it mainly affects test/import-performance expectations for agent-runtime adjacent code. (src/agents/AGENTS.md:1, 3217165be770)
Current-main memory gate gap: Current main funnels both preflight and flush through followupUsesCliRuntime, but that helper only checks direct CLI providers and compatible persisted runtime pins; it does not use the dispatch resolver for configured model policy or auth-profile/order CLI routing. (src/auto-reply/reply/agent-runner-memory.ts:263, 3217165be770)
Dispatch contract: The dispatch path resolves a provider-compatible session override, then falls through to resolveCliRuntimeExecutionProvider with the selected auth profile before checking isCliProvider. (src/auto-reply/reply/agent-runner-execution.ts:2208, 3217165be770)
Follow-up runner sibling contract: The follow-up runner uses the same dispatch-style chain of session runtime override, auth-profile-aware CLI runtime resolution, and isCliProvider classification. (src/auto-reply/reply/followup-runner.ts:904, 3217165be770)
Resolver covers missing cases: resolveCliRuntimeExecutionProvider handles configured runtime policy and auth-profile/order derived CLI runtime resolution, which is the missing branch in current-main followupUsesCliRuntime. (src/agents/model-runtime-aliases.ts:232, 3217165be770)

Likely related people:

vincentkoc: Current-main blame points to this author for followupUsesCliRuntime and dispatch resolver code, and merged PR history shows a follow-up memory-path commit for compatible CLI session pins. (role: recent area contributor; confidence: high; commits: 303b2f794f6c, e9720c27fa69; files: src/auto-reply/reply/agent-runner-memory.ts, src/auto-reply/reply/agent-runner-execution.ts)
yu-xin-c: Authored the merged provider-compatible CLI session runtime pin work that changed the dispatch and memory-path invariants this PR must preserve. (role: adjacent feature contributor; confidence: high; commits: 1917cc4b8591, 9974641d1ee1; files: src/auto-reply/reply/agent-runner-execution.ts, src/auto-reply/reply/followup-runner.ts, src/auto-reply/reply/agent-runner-memory.ts)
potterdigital: Authored the merged runtime-policy resolver fix and the closed narrower PR that reported the same memory-gate CLI runtime classification gap. (role: adjacent runtime-policy contributor; confidence: medium; commits: efd88dc00d8d; files: src/agents/model-runtime-policy.ts, src/agents/model-runtime-aliases.ts, src/auto-reply/reply/agent-runner-memory.ts)

What the crustacean ranks mean

🦀 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.

How this review workflow works

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.

clawsweeper · 2026-05-24T23:26:45Z

ClawSweeper PR egg

✨ Hatched: 🌱 uncommon Mossy Proofling

Hatch command

Comment @clawsweeper hatch when this PR is hatchable.

Hatchability rules:

Merged PRs are hatchable.
Open PRs are hatchable when they are status: 👀 ready for maintainer look, status: 🚀 automerge armed, or labeled clawsweeper:automerge.
Closed unmerged PRs are hatchable only when one of those hatchable labels is still present in the durable record.

Rarity: 🌱 uncommon.
Trait: collects tiny proofs.
Image traits: location diff observatory; accessory shell-shaped keyboard; palette charcoal, cyan, and signal green; mood celebratory; pose nestled inside a glowing shell; shell soft velvet shell; lighting subtle sparkle highlights; background little resolved-comment flags.
Share on X: post this hatch
Copy: My PR egg hatched a 🌱 uncommon Mossy Proofling in ClawSweeper.

What is this egg doing here?

Eggs appear after the PR passes real-behavior proof. It is here for vibes, not verdicts: it does not change labels, ratings, merge decisions, or automation.
The shell reacts to review momentum: open follow-up work warms it up, re-review makes it wobble, and a clean final review lets it hatch.
Hatchability usually comes from sufficient real-behavior proof, no blocking P0/P1/P2 findings, no security attention needed, and clean correctness. A merged PR is already final, so merge makes the egg hatchable independently.
The hatch is seeded from this repository and PR number, so the same PR keeps the same creature; the reviewed head SHA can only change safe visual details.
Rarity is just collectible sparkle: 🥚 common, 🌱 uncommon, 💎 rare, ✨ glimmer, and 🌈 legendary.

Mirrors the dispatch path's call to resolveCliRuntimeExecutionProvider, which uses the run's selected auth profile (via resolveRunAuthProfile) to handle the case where auth.order prefers a CLI profile but the session is pinned to a direct-API auth profile. Without authProfileId, the memory gates would resolve "claude-cli" while the dispatch path keeps the direct embedded route — the gates then incorrectly skip preflight compaction / memory flush even when dispatch runs embedded. Adds two regression tests configuring auth.order with a CLI profile first while the run pins authProfileId to the direct-API profile, asserting both gates proceed. Addresses Codex review on PR openclaw#86224.

potterdigital · 2026-05-25T07:25:56Z

Drive-by note from a parallel investigation — I had #86361 open targeting the per-model runtime path before this PR appeared; closed it because your resolver-based shape is the right one.

Production validation: I've been running an equivalent resolveCliRuntimeExecutionProvider-based gate as a bundle patch on a 24/7 production gateway (claude-cli backend, Claude subscription OAuth) since 2026-05-24. Composition differs from yours — mine is isCliProvider(p) || resolveCliRuntimeExecutionProvider(...) !== void 0, yours is isCliProvider(resolveCliRuntimeExecutionProvider(...) ?? p) — but they behave the same at the gate boundary. No recurrence of the 1.3M-context 404 cascade the original bug was tracking.

On the persisted-override P2: My prod failure path was the per-model agentRuntime.id route, so I never exercised the session-persisted agentRuntimeOverride branch clawsweeper flagged. Can't add evidence on that branch from my config — flagging in case it's useful that my prod validation doesn't cover it.

Adjacent merge context: #85970 (merged yesterday) patched modelEntryMatchKind in model-runtime-policy.ts for empty-provider session entries. That policy table is what resolveCliRuntimeExecutionProvider consults via resolveConfiguredRuntime — without #85970, the resolver call here could silently miss per-model agentRuntime.id entries on empty-provider sessions. Different files (src/agents/ vs src/auto-reply/), so no merge conflict, but the two fixes ride together.

@clawsweeper

adele-with-a-b · 2026-05-25T15:27:26Z

Thanks for the production validation — 24h on a real gateway with the equivalent gate logic is exactly the kind of signal that's hard to get from tests alone.

Good call flagging the agentRuntimeOverride session-persisted branch. My config doesn't exercise that path either (I hit the bug via auth.profiles mapping, not per-model runtime overrides). I'll add a unit test covering the persisted-override case before requesting final review so clawsweeper's P2 has coverage regardless of prod configs.

Re #85970 adjacency: agreed, these ride together. The empty-provider session entries that #85970 fixes would have caused resolveConfiguredRuntime to return undefined on lookup, which means resolveCliRuntimeExecutionProvider falls through to the ?? p passthrough — effectively reverting to the broken isCliProvider(configured_provider) behavior for those sessions. Good that it merged first. I'll add a note in the PR description calling out the dependency.

Mirrors the dispatch path's call to resolveCliRuntimeExecutionProvider, which uses the run's selected auth profile (via resolveRunAuthProfile) to handle the case where auth.order prefers a CLI profile but the session is pinned to a direct-API auth profile. Without authProfileId, the memory gates would resolve "claude-cli" while the dispatch path keeps the direct embedded route — the gates then incorrectly skip preflight compaction / memory flush even when dispatch runs embedded. Adds two regression tests configuring auth.order with a CLI profile first while the run pins authProfileId to the direct-API profile, asserting both gates proceed. Addresses Codex review on PR openclaw#86224.

Mirrors the three-tier precedence from runReplyAgent's dispatch path: 1. agentRuntimeOverride === "pi" → use canonical provider (forces embedded) 2. agentRuntimeOverride is a CLI alias for provider → use the override 3. Fall through to resolveCliRuntimeExecutionProvider (the auth-profile-aware resolver already wired into both gates) Without (1) and (2), sessions with persisted agentRuntimeOverride: "claude-cli" were being classified as non-CLI by the memory gates while dispatch routed them through CLI — the same embedded transcript compaction failure mode that this PR's earlier commit fixed for auth-profile mappings. Adds regression tests for the CLI-override case at both gates. Addresses ClawSweeper P2 review comment on PR openclaw#86224.

adele-with-a-b · 2026-05-25T15:52:45Z

@clawsweeper re-review

Pushed 26befa57 addressing the P2 — both memory gates now apply the same three-tier sessionRuntimeOverride precedence as the dispatch path:

agentRuntimeOverride === "pi" → use canonical provider (forces embedded)
agentRuntimeOverride is a CLI alias for the provider → use the override directly
Fall through to resolveCliRuntimeExecutionProvider (the auth-profile-aware resolver from the previous commit)

Plus 3 regression tests covering CLI-override-skips-flush, CLI-override-skips-preflight, and pi-override-runs-flush-despite-CLI-auth-order. pnpm tsgo:core + tsgo:core:test + oxfmt --check all clean; vitest run agent-runner-memory.test.ts is 32 passing.

Also rebased onto current main to clear the unrelated check-dependencies failure on the prior head (extensions/acpx + canvas + diffs unused-files were a transient main-side issue, gone post-rebase).

clawsweeper · 2026-05-25T15:52:47Z

🦞🧹
ClawSweeper re-review requested.

I asked ClawSweeper to review this item again.
Action: item re-review queued (workflow sweep.yml, event repository_dispatch).
Result: the existing ClawSweeper review comment will be edited in place when the review finishes.

Re-review progress:

State: Complete
Detail: The targeted re-review finished, the durable review comment was updated, and the synced verdict was routed.
Run: https://github.com/openclaw/clawsweeper/actions/runs/26408932489
Updated: 2026-05-25T16:03:04.149Z

runPreflightCompactionIfNeeded and runMemoryFlushIfNeeded both check isCliProvider on the canonical provider id stored on the followup run. isCliProvider only reports true when that id is itself a key in agents.defaults.cliBackends, so an agent configured with a primary model on a canonical provider (e.g. "anthropic") that resolves through a CLI runtime via models.providers.<provider>.agentRuntime.id evaluates isCli=false. Preflight then runs as if the agent were embedded. The runtime dispatch path in runReplyAgent already resolves the runtime provider via resolveCliRuntimeExecutionProvider before checking isCliProvider (see src/auto-reply/reply/agent-runner-execution.ts around the isCliProvider gate). The preflight + memory-flush gates predate that resolve and were never updated. Mirror the established shape at both call sites so a CLI-routed agent skips the embedded-only preflight transcript compaction. Adds two regression tests asserting that a config with agents.defaults.cliBackends["claude-cli"] plus models.providers.anthropic.agentRuntime.id = "claude-cli" causes both runMemoryFlushIfNeeded and runPreflightCompactionIfNeeded to short-circuit when the followupRun carries provider="anthropic".

Mirrors the dispatch path's call to resolveCliRuntimeExecutionProvider, which uses the run's selected auth profile (via resolveRunAuthProfile) to handle the case where auth.order prefers a CLI profile but the session is pinned to a direct-API auth profile. Without authProfileId, the memory gates would resolve "claude-cli" while the dispatch path keeps the direct embedded route — the gates then incorrectly skip preflight compaction / memory flush even when dispatch runs embedded. Adds two regression tests configuring auth.order with a CLI profile first while the run pins authProfileId to the direct-API profile, asserting both gates proceed. Addresses Codex review on PR openclaw#86224.

Mirrors the three-tier precedence from runReplyAgent's dispatch path: 1. agentRuntimeOverride === "pi" → use canonical provider (forces embedded) 2. agentRuntimeOverride is a CLI alias for provider → use the override 3. Fall through to resolveCliRuntimeExecutionProvider (the auth-profile-aware resolver already wired into both gates) Without (1) and (2), sessions with persisted agentRuntimeOverride: "claude-cli" were being classified as non-CLI by the memory gates while dispatch routed them through CLI — the same embedded transcript compaction failure mode that this PR's earlier commit fixed for auth-profile mappings. Adds regression tests for the CLI-override case at both gates. Addresses ClawSweeper P2 review comment on PR openclaw#86224.

… in memory gates

…lution The preflight-compaction and memory-flush gates short-circuited on agentRuntimeOverride === "pi" by forcing canonical/embedded routing, but runReplyAgent's dispatch path (agent-runner-execution.ts:2056, followup-runner.ts:726) has no such "pi" branch: it resolves embedded-vs-CLI purely from a provider-compatible persisted override plus the auth-profile-aware resolver. For a session pinned to "pi" on a CLI-routed config, the gate ran embedded maintenance while dispatch ran CLI -- out of sync. Option A: make the gate match dispatch. Remove the gate's "pi" short-circuit so it falls through to the same config-driven resolution dispatch uses. Now the gate skips embedded maintenance exactly when dispatch routes to CLI, and runs it when dispatch runs embedded -- in both directions. Flip the test that codified the divergence (pi override + CLI-preferring auth.order now SKIPS, matching dispatch) and add the inverse regression (pi override + no CLI route still RUNS embedded), proving gate and dispatch agree both ways.

adele-with-a-b · 2026-06-08T16:05:14Z

Rebased onto current upstream/main (c0f7292) and addressed the P1 gate↔dispatch divergence.

Root cause of the P1: the preflight/memory-flush gates imported resolveSessionRuntimeOverrideForProvider from session-runtime-compat.ts (which collapses "pi"→"openclaw" and returns "openclaw"), while dispatch imports a same-named but different function from agent-runner-execution.ts (which returns undefined for "pi"). The gate then papered over this with a === "pi" short-circuit forcing canonical/embedded, which dispatch has no equivalent for — so a "pi"-pinned session on a CLI-routed config ran embedded maintenance in the gate but went CLI in dispatch.

Fix (maintainer option 1 — align the gate with dispatch): removed the gate's "pi" short-circuit at both sites so the gate falls through to the same config-driven resolveCliRuntimeExecutionProvider(...) resolution dispatch uses. The gate now skips embedded maintenance exactly when dispatch routes to CLI, and runs it when dispatch runs embedded — verified in both directions.

Tests: flipped the test that codified the divergence (pi + CLI-preferring auth.order now SKIPS, matching dispatch) and added the inverse regression (pi + no CLI route still RUNS embedded). All green on the rebased tree: agent-runner-memory.test.ts 50/50, agent-runner-execution.test.ts + followup-runner.test.ts 254/254, tsgo:core + tsgo:core:test clean.

@clawsweeper re-review

clawsweeper · 2026-06-08T16:05:17Z

🦞🧹
ClawSweeper re-review requested.

I asked ClawSweeper to review this item again.
Action: item re-review queued (workflow sweep.yml, event repository_dispatch).
Result: the existing ClawSweeper review comment will be edited in place when the review finishes.

Re-review progress:

State: Complete
Detail: The targeted re-review finished, the durable review comment was updated, and the synced verdict was routed.
Run: https://github.com/openclaw/clawsweeper/actions/runs/27150606016
Updated: 2026-06-08T16:14:23.699Z

openclaw-barnacle · 2026-07-12T04:28:56Z

This pull request has been automatically marked as stale due to inactivity.
Please add updates or it will be closed.

openclaw-barnacle · 2026-07-19T04:32:40Z

Closing due to inactivity.
If you believe this PR should be revived, post in #clawtributors on Discord to talk to a maintainer.
That channel is the escape hatch for high-quality PRs that get auto-closed.

openclaw-barnacle Bot added size: S triage: needs-real-behavior-proof Candidate: external PR needs after-fix proof from a real setup. labels May 24, 2026

openclaw-barnacle Bot added size: M proof: supplied External PR includes structured after-fix real behavior proof. and removed size: S triage: needs-real-behavior-proof Candidate: external PR needs after-fix proof from a real setup. labels May 25, 2026

This was referenced May 25, 2026

fix Codex preflight compaction auth routing #86292

Closed

fix(auto-reply): per-model runtime policy in memory-flush gates #86361

Closed

adele-with-a-b force-pushed the fix/agent-runner-memory-resolve-cli-runtime-before-gate branch from 62b2fb1 to 2d4f556 Compare May 25, 2026 15:41

openclaw-barnacle Bot removed the proof: sufficient ClawSweeper judged the real behavior proof convincing. label May 25, 2026

clawsweeper Bot added the proof: sufficient ClawSweeper judged the real behavior proof convincing. label May 25, 2026

adele-with-a-b added 5 commits June 8, 2026 10:55

fix(agents): honor agentRuntimeOverride only when provider-compatible…

8d1e2c3

… in memory gates

adele-with-a-b force-pushed the fix/agent-runner-memory-resolve-cli-runtime-before-gate branch from 872f6d0 to c0f7292 Compare June 8, 2026 16:04

openclaw-barnacle Bot removed the proof: sufficient ClawSweeper judged the real behavior proof convincing. label Jun 8, 2026

clawsweeper Bot added the proof: sufficient ClawSweeper judged the real behavior proof convincing. label Jun 8, 2026

clawsweeper Bot mentioned this pull request Jun 26, 2026

fix(cli): enable pre-compaction memory flush for CLI backends and hide fake context bar (#96858) #96918

Closed

clawsweeper Bot mentioned this pull request Jul 5, 2026

CLI-backed helper paths still bypass CLI dispatch on latest main #57326

Open

openclaw-barnacle Bot added the stale Marked as stale due to inactivity label Jul 12, 2026

openclaw-barnacle Bot closed this Jul 19, 2026

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

Uh oh!

[AI-assisted] fix(agents): resolve CLI runtime in preflight + memory-flush gates#86224

[AI-assisted] fix(agents): resolve CLI runtime in preflight + memory-flush gates#86224
adele-with-a-b wants to merge 5 commits into
openclaw:mainfrom
adele-with-a-b:fix/agent-runner-memory-resolve-cli-runtime-before-gate

adele-with-a-b commented May 24, 2026 •

edited

Loading

Uh oh!

clawsweeper Bot commented May 24, 2026 •

edited

Loading

Uh oh!

clawsweeper Bot commented May 24, 2026 •

edited

Loading

Uh oh!

potterdigital commented May 25, 2026 •

edited

Loading

Uh oh!

adele-with-a-b commented May 25, 2026 •

edited

Loading

Uh oh!

adele-with-a-b commented May 25, 2026

Uh oh!

clawsweeper Bot commented May 25, 2026 •

edited

Loading

Uh oh!

adele-with-a-b commented Jun 8, 2026

Uh oh!

clawsweeper Bot commented Jun 8, 2026 •

edited

Loading

Uh oh!

openclaw-barnacle Bot commented Jul 12, 2026

Uh oh!

openclaw-barnacle Bot commented Jul 19, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

Uh oh!

Uh oh!

Conversation

adele-with-a-b commented May 24, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Summary

Existing pattern

Reproduction context

Verification

Real behavior proof

Uh oh!

clawsweeper Bot commented May 24, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

clawsweeper Bot commented May 24, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Hatch command

Uh oh!

potterdigital commented May 25, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

adele-with-a-b commented May 25, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

adele-with-a-b commented May 25, 2026

Uh oh!

clawsweeper Bot commented May 25, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

adele-with-a-b commented Jun 8, 2026

Uh oh!

clawsweeper Bot commented Jun 8, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

openclaw-barnacle Bot commented Jul 12, 2026

Uh oh!

openclaw-barnacle Bot commented Jul 19, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

adele-with-a-b commented May 24, 2026 •

edited

Loading

clawsweeper Bot commented May 24, 2026 •

edited

Loading

clawsweeper Bot commented May 24, 2026 •

edited

Loading

potterdigital commented May 25, 2026 •

edited

Loading

adele-with-a-b commented May 25, 2026 •

edited

Loading

clawsweeper Bot commented May 25, 2026 •

edited

Loading

clawsweeper Bot commented Jun 8, 2026 •

edited

Loading