feat: add per-agent compaction overrides by kklouzal · Pull Request #88506 · openclaw/openclaw

kklouzal · 2026-05-31T03:05:04Z

Summary

Adds per-agent compaction and contextPruning overrides so each configured agent can resolve maintenance/runtime behavior from its active agent scope instead of always inheriting every value from agents.defaults.

This is for multi-agent installations where main, channel/proxy, local/self-hosted, and Codex/app-server agents need different compaction, pruning, timeout, thinking-level, and transcript-maintenance behavior.

Linked Issues And Related Work

Closes #52732

Related: #52751, #69988, #19501, #15826, #43834, #6571, #6572, #30452, #57901, #58077, #59863, #60512, #80784, #82008

Problem / Root Cause

Before this change, per-agent compaction settings were either rejected by schema or only partially honored. Several runtime paths still read agents.defaults.compaction / agents.defaults.contextPruning even when the active agents.list[] entry had its own settings.

That weakened these setups:

main/default agent compacts with one model/provider while another agent compacts with another
local/self-hosted agents use local provider aliases for compaction
Codex/app-server agents intentionally omit or ignore OpenClaw model overrides and keep native server-side compaction behavior
legacy/non-agent-shaped session keys still need an explicit non-main agentId preserved across queued/direct compaction and maintenance paths
queued/CLI/context-engine compaction should use compaction.thinkingLevel, not the active run's normal thinking level

What Changed

Adds strict schema/docs support for agents.list[].compaction and per-agent contextPruning.
Deep-merges per-agent compaction and contextPruning blocks over agents.defaults, so partial overrides inherit unspecified fields.
Resolves compaction model/provider/auth-profile/timeout/cache/pruning/reserve/thinking settings through the active session agent.
Threads sessionAgentId through queued/direct compaction, CLI compaction, context-engine maintenance, Codex app-server lifecycle, embedded cache-TTL finalization, memory flush, transcript rotation, and follow-up/notice paths.
Keeps explicit OpenClaw compaction.model behavior separate from Codex/app-server native runtime selection.
Adds regression coverage for legacy-session/non-main-agent scope in direct, queued, CLI, Codex lifecycle, cache-TTL, memory-flush, empty/partial merge, and compaction thinking-level propagation paths.

Reviewer Notes

The key invariant is:

A compaction run should resolve configuration from the same active agent scope as the session/run that triggered it, even when the session key is legacy-shaped and does not encode agent:<id>:.

That is why this PR explicitly passes agentId into compaction and maintenance params instead of relying only on parsing sessionKey.

Real behavior proof

Behavior or issue addressed: Refreshed live proof for the per-agent compaction override resolution work, covering model/provider, thinking level, timeout, reserve-token floor, small-context reserve-floor capping, clean config validation, focused compaction regression coverage, and actual model/gateway requests after the patch.
Current PR head: cfc4ff41b9fc7ee1709ca0429e2bb3f4c342868a (feat(compaction): add per-agent compaction overrides), a final single-commit squash on top of fork main at 26913e60a42e84dd256a682c5239be57f346e244.
Proof note: The detailed live proof below was collected before the final rebase/squash at f723ad186d7879d12c61562dd4c2dab3745b5ed3. The final squash preserves the intended branch content as one commit, but exact-head proof has not yet been rerun for cfc4ff41b9fc7ee1709ca0429e2bb3f4c342868a.
Real environment tested for proof refresh: Nano-01 / Orion proof host at /home/kklouzal/openclaw-pr-83637, detached checkout f723ad186d7879d12c61562dd4c2dab3745b5ed3, Node v24.15.0, pnpm 11.2.2, model github-copilot/gpt-5-mini, clean proof config at /home/kklouzal/openclaw-pr-83637/.artifacts/proof-refresh-f723ad18-20260527T050453Z/model-home/openclaw.json, redacted artifacts at /home/kklouzal/openclaw-pr-83637/.artifacts/proof-refresh-f723ad18-20260527T050453Z/redacted.
Exact steps or command run for that proof refresh: fetched the pushed fork branch, detached the proof checkout to f723ad186d7879d12c61562dd4c2dab3745b5ed3, ran pnpm install --frozen-lockfile, pnpm build:strict-smoke, focused compaction Vitest coverage, tsgo core and core-test typechecks, git diff --check, a source-import compaction config proof, clean openclaw config validate, openclaw models status --probe --json, openclaw infer model run --local --model github-copilot/gpt-5-mini, OPENCLAW_LIVE_MODELS=github-copilot/gpt-5-mini node scripts/test-live.mjs src/agents/models.profiles.live.test.ts, and OPENCLAW_LIVE_GATEWAY_MODELS=github-copilot/gpt-5-mini node scripts/test-live.mjs src/gateway/gateway-models.profiles.live.test.ts.
Evidence after fix: focused compaction suite passed 22 files / 779 tests; build:strict-smoke, tsgo core, tsgo core-test, and git diff --check passed; the compaction config proof returned status: pass; clean config validation passed; direct model output returned OPENCLAW_PR83637_F723AD18_OK; live direct model suite passed 17 tests; live gateway suite passed 68 tests.

Redacted command summary and compaction proof

{
 "pr": 83637,
 "head": "f723ad186d7879d12c61562dd4c2dab3745b5ed3",
 "host": "Nano-01",
 "artifact": "/home/kklouzal/openclaw-pr-83637/.artifacts/proof-refresh-f723ad18-20260527T050453Z",
 "commands": {
 "01-install": "0",
 "02-build-strict-smoke": "0",
 "03-config-validate": "1",
 "04-focused-compaction-tests": "0",
 "05-tsgo-core": "0",
 "06-tsgo-core-test": "0",
 "07-diff-check": "0",
 "08-compaction-config-proof": "0",
 "09-clean-config-validate": "0",
 "10-models-status": "0",
 "11-model-run": "0",
 "12-live-models": "0",
 "13-live-gateway-models": "0"
 },
 "markers": {
 "model": true,
 "liveModels": true,
 "liveGateway": true,
 "compactionConfig": true
 }
}

{
 "status": "pass",
 "head": "f723ad186d7879d12c61562dd4c2dab3745b5ed3",
 "default": {
 "timeoutMs": 90000,
 "think": "minimal",
 "target": {
 "provider": "github-copilot",
 "model": "gpt-5-mini"
 },
 "floor": 20000,
 "settings": {
 "applied": {
 "didOverride": true,
 "compaction": {
 "reserveTokens": 20000,
 "keepRecentTokens": 1024
 }
 },
 "reserve": 20000,
 "keep": 1024
 }
 },
 "flat": {
 "timeoutMs": 61000,
 "think": "off",
 "target": {
 "provider": "openai",
 "model": "gpt-5.4-mini"
 },
 "floor": 0,
 "settings": {
 "applied": {
 "didOverride": true,
 "compaction": {
 "reserveTokens": 2048,
 "keepRecentTokens": 512
 }
 },
 "reserve": 2048,
 "keep": 512
 }
 },
 "nested": {
 "timeoutMs": 122000,
 "think": "high",
 "target": {
 "provider": "github-copilot",
 "model": "gpt-5.4-mini",
 "authProfileId": "copilot-auth"
 },
 "floor": 24000,
 "settings": {
 "applied": {
 "didOverride": true,
 "compaction": {
 "reserveTokens": 24000,
 "keepRecentTokens": 2048
 }
 },
 "reserve": 24000,
 "keep": 2048
 }
 },
 "smallWindow": {
 "applied": {
 "didOverride": true,
 "compaction": {
 "reserveTokens": 8384,
 "keepRecentTokens": 512
 }
 },
 "reserve": 8384,
 "keep": 512
 },
 "runtime": {
 "provider": "openai",
 "model": "gpt-5.4-mini",
 "thinkLevel": "off"
 }
}

Observed result after fix: per-agent compaction settings resolved independently: defaults produced timeoutMs=90000, think=minimal, and github-copilot/gpt-5-mini; flat-agent overrode to timeoutMs=61000, think=off, provider/model openai/gpt-5.4-mini, and dropped the mismatched auth profile; nested-agent overrode to timeoutMs=122000, think=high, github-copilot/gpt-5.4-mini, and retained the matching auth profile; small-context capping reduced reserve tokens to 8384 for a 16K context budget. Real direct and gateway model requests completed successfully after the patch.
What was not tested: this run did not restart the production Dev-01 gateway or mutate the live Discord proof agents. The default host config on Nano-01 still contains newer slot shapes unrelated to feat: add per-agent compaction overrides (AI-assisted) #83637 and failed validation there; the clean proof config passed and was used for model/gateway evidence. Codex app-server native compaction still owns server-side compaction model behavior, so OpenClaw-selected model override usage is proven through source/runtime resolution and focused tests rather than by claiming Codex server-side model selection.

Validation

Acceptance commands requested by ClawSweeper:

node scripts/test-projects.mjs src/agents/agent-scope.test.ts src/agents/pi-settings.test.ts src/agents/pi-embedded-runner/compaction-runtime-context.test.ts src/agents/command/cli-compaction.test.ts src/auto-reply/reply/agent-runner-memory.test.ts

Result:

[test] starting test/vitest/vitest.auto-reply.config.ts
Test Files 1 passed (1)
Tests 25 passed (25)

[test] starting test/vitest/vitest.agents.config.ts
Test Files 4 passed (4)
Tests 115 passed (115)

[test] passed 2 Vitest shards in 19.18s

Schema baseline:

node --import tsx scripts/generate-base-config-schema.ts --check
[base-config-schema] ok

Config docs baseline:

node --import tsx scripts/generate-config-doc-baseline.ts --check
OK docs/.generated/config-baseline.sha256

Note: this docs-baseline check initially caught stale committed hash drift. Regenerated docs/.generated/config-baseline.sha256, committed it as 04170e48eb01fc29f7f3acb1a830c5742b41c3ea, and reran the check successfully.

Additional gate:

git diff --check
passed

Final Merge/Polish Validation

After conflict cleanup, final branch polish, and latest main merge repair, the old branch line was pushed at f239d594d22b56837bb371770898c30759e81515. That line was later rebased and squashed into the current formal PR head cfc4ff41b9fc7ee1709ca0429e2bb3f4c342868a.

Final local gates on the prior pushed line:

node scripts/test-projects.mjs src/agents/skills.test.ts src/commands/channel-setup/plugin-install.test.ts src/plugins/plugin-registry.test.ts test/scripts/check-extension-package-tsc-boundary.test.ts test/scripts/docker-all-scheduler.test.ts
Test Files 5 passed; Tests 101 passed

node --max-old-space-size=8192 --import tsx scripts/generate-plugin-sdk-api-baseline.ts --check
OK docs/.generated/plugin-sdk-api-baseline.sha256

PATH="/tmp/openclaw-corepack-bin:" pnpm prompt:snapshots:check
Prompt snapshots are current (7 files)

PATH="/tmp/openclaw-corepack-bin:" OPENCLAW_ADDITIONAL_BOUNDARY_SHARD=1/4 OPENCLAW_ADDITIONAL_BOUNDARY_CONCURRENCY=4 node scripts/run-additional-boundary-checks.mjs
6/6 shard checks passed

git diff --check
passed

Current Branch State

Base branch: main
Current PR head: cfc4ff41b9fc7ee1709ca0429e2bb3f4c342868a
Branch shape: one squashed commit over fork main at 26913e60a42e84dd256a682c5239be57f346e244.
Local final-shape checks before PR creation:
- git diff --stat origin/main...fork/feat/agent-compaction-overrides: 82 files changed, 3613 insertions, 398 deletions.
- git rev-list --left-right --count origin/main...fork/feat/agent-compaction-overrides: 1 1 because the branch is based on the pinned green fork-main commit one commit behind upstream.
- Branch verified as one squashed commit over fork/main.
- Anchored conflict-marker scan clean.
- git diff --check origin/main...fork/feat/agent-compaction-overrides: passed.

clawsweeper · 2026-05-31T03:06:54Z

Codex review: needs real behavior proof before merge. Reviewed June 29, 2026, 5:41 PM ET / 21:41 UTC.

Summary
The PR adds per-agent compaction/contextPruning config resolution across agent scope, compaction runtime, Codex app-server lifecycle, CLI, memory, context-engine paths, docs, generated baselines, and regression tests.

Reproducibility: yes. for the PR blockers: source inspection of the current PR head shows the missing contextPruning export/schema support, defaults-only cache-TTL read, and duplicate agentId key. I did not run tests because this review is read-only.

Review metrics: 1 noteworthy metric.

Per-agent config surface: 1 added, 1 claimed but incomplete. The diff adds agents.list[].compaction schema/help coverage while the promised agents.list[].contextPruning surface is still missing, so maintainers need to review the public config contract before merge.

Stored data model
Persistent data-model change detected: persistent cache schema: src/agents/embedded-agent-runner/run/attempt.thread-helpers.ts, persistent cache schema: src/config/zod-schema.agent-defaults.ts, persistent cache schema: src/config/zod-schema.agent-runtime.ts, serialized state: extensions/codex/src/app-server/run-attempt.test.ts, serialized state: extensions/codex/src/app-server/session-binding.ts, serialized state: src/agents/command/cli-compaction.test.ts, and 10 more. Confirm migration or upgrade compatibility proof before merge.

Root-cause cluster
Relationship: fixed_by_candidate
Canonical: #52732
Summary: This PR is the active open candidate for the canonical per-agent compaction/contextPruning issue, with partial overlap to broader compaction model-routing work.

Members:

canonical: Per-agent compaction and contextPruning overrides in agents.list #52732 - The live closing issue reference points to this issue and its title/body match the requested per-agent compaction/contextPruning feature.
superseded: feat: add per-agent compaction and contextPruning config overrides #52751 - This earlier same-feature PR is closed unmerged and was followed by broader implementation attempts.
superseded: feat: add per-agent compaction overrides (AI-assisted) #83637 - This earlier broad implementation PR is closed unmerged and the current PR describes itself as the squashed follow-up line.
partial_overlap: Safeguard compaction ignores compaction.model config — uses session model instead #57901 - That issue focuses on safeguard compaction model routing, which this PR touches, but it is narrower than the per-agent config feature.
partial_overlap: [Feature]: Compaction: configurable model override + flush-then-reset mode #30452 - That issue covers broader compaction model and memory-flush controls beyond this PR's per-agent override focus.

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

Merge readiness
Overall: 🧂 unranked krab
Proof: 🦪 silver shellfish
Patch quality: 🧂 unranked krab
Result: blocked until stronger real behavior proof is added.

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

Rank-up moves:

[P1] Fix the contextPruning resolver/schema/cache-TTL wiring and duplicate agentId key.
Refresh the branch until GitHub can evaluate the merge result cleanly.
[P1] Add redacted exact-head real behavior proof after the repair.

Proof guidance:

[P1] Needs stronger real behavior proof before merge: The PR body includes terminal/live proof from an earlier commit and explicitly says exact-head proof was not rerun for cfc4ff4; a redacted exact-head terminal log, live output, recording, or linked artifact is still needed 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.

Risk before merge

[P1] The PR expands strict user-facing agents.list[] config/default resolution, so schema, docs, generated baselines, fresh-install behavior, and upgrade behavior need maintainer review together before merge.
[P1] Provider/model/auth-profile routing changes span default compaction, safeguard compaction, Codex app-server, CLI compaction, and memory flush, so a bad scope decision can use the wrong model or credentials.
[P1] Session-agent identity is threaded through compaction, maintenance, transcript, memory, and context-pruning paths, so stale or missing agent IDs can mis-associate session side effects.
[P1] The branch is draft and currently conflicting, and the PR body only has real behavior proof from an earlier pre-squash commit.

Maintainer options:

Repair blockers and rerun exact-head proof (recommended)
Fix the missing contextPruning schema/resolver/cache-TTL wiring and duplicate key, then add redacted terminal, log, live output, or linked artifact proof for the repaired current head.
Approve the config contract explicitly
Maintainers can accept the new per-agent config surface only after reviewing strict validation, generated baselines, docs, and upgrade behavior as one public contract.
Pause or replace the draft
If the broad conflicting draft remains hard to repair, sponsor a narrower replacement that preserves the useful implementation and closes this branch later.

Next step before merge

[P1] Human review is needed because the external draft is conflicting, lacks exact-head proof, and needs maintainer approval of a compatibility-sensitive config contract after code blockers are fixed.

Security
Cleared: No concrete security or supply-chain regression was found; the blockers are correctness, compatibility, auth-provider routing, session scope, and proof freshness.

Review findings

[P1] Export and use the scoped context-pruning resolver — src/agents/embedded-agent-runner/extensions.ts:14
[P1] Accept per-agent contextPruning in strict schema — src/config/zod-schema.agent-runtime.ts:1141
[P2] Use the active agent when appending cache-TTL markers — src/agents/embedded-agent-runner/run/attempt.thread-helpers.ts:53-55

Review details

Best possible solution:

Land a refreshed implementation that exports shared per-agent compaction/contextPruning resolvers, updates strict schema/docs/baselines, preserves existing defaults, and proves fresh-install plus upgrade behavior at the exact head.

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

Yes for the PR blockers: source inspection of the current PR head shows the missing contextPruning export/schema support, defaults-only cache-TTL read, and duplicate agentId key. I did not run tests because this review is read-only.

Is this the best way to solve the issue?

No. The direction is plausible, but this is not the best merge state until the schema/resolver blockers, duplicate key, branch conflict, exact-head proof, and maintainer config-contract review are addressed.

Full review comments:

[P1] Export and use the scoped context-pruning resolver — src/agents/embedded-agent-runner/extensions.ts:14
resolveAgentContextPruningConfig is imported here, but the PR head does not export that helper from agent-scope-config.ts, so this branch will not typecheck. The same path still reads defaults-only pruning settings, so the resolver needs to exist and be used with the active agent id.
Confidence: 0.96
[P1] Accept per-agent contextPruning in strict schema — src/config/zod-schema.agent-runtime.ts:1141
The PR promises agents.list[].contextPruning, but AgentEntrySchema only adds compaction and still omits contextPruning. A user config with the requested pruning override would still be rejected as an unrecognized per-agent key.
Confidence: 0.95
[P2] Use the active agent when appending cache-TTL markers — src/agents/embedded-agent-runner/run/attempt.thread-helpers.ts:53-55
The caller now passes agentId, but shouldAppendAttemptCacheTtl ignores it and still reads agents.defaults.contextPruning. That means per-agent cache-TTL overrides are not honored for the marker that drives the pruning lifecycle.
Confidence: 0.9
[P1] Remove the duplicate agentId property — src/agents/embedded-agent-runner/run.ts:1992
This object literal sets agentId twice in the same runPostCompactionSideEffects call. Typecheck will reject the duplicate key, so the branch cannot be merged until one entry is removed.
Confidence: 0.93

Overall correctness: patch is incorrect
Overall confidence: 0.94

AGENTS.md: found and applied where relevant.

Codex review notes: model internal, reasoning high; reviewed against 15de9d881a84.

Label changes

Label justifications:

P2: This is a normal-priority feature/config repair for multi-agent compaction behavior with meaningful but bounded blast radius.
merge-risk: 🚨 compatibility: The PR changes strict user-facing config/default resolution and generated config baselines, which can affect validation and upgrades.
merge-risk: 🚨 auth-provider: The diff changes compaction provider, model, and auth-profile selection across multiple runtime paths.
merge-risk: 🚨 session-state: The diff threads session-agent identity through compaction, transcript, memory, and context-engine side effects.
rating: 🧂 unranked krab: Overall readiness is 🧂 unranked krab; proof is 🦪 silver shellfish and patch quality is 🧂 unranked krab.
status: 📣 needs proof: The PR needs real behavior proof before ClawSweeper can clear the contributor ask. Needs stronger real behavior proof before merge: The PR body includes terminal/live proof from an earlier commit and explicitly says exact-head proof was not rerun for cfc4ff4; a redacted exact-head terminal log, live output, recording, or linked artifact is still needed 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.

Evidence reviewed

What I checked:

Repository policy read: Root AGENTS.md plus scoped guidance for extensions, agents, embedded-runner, plugins, plugin SDK, docs, and tests were read; the config/provider/session-state compatibility rules directly apply to this review. (AGENTS.md:31, 15de9d881a84)
Live PR state: GitHub reports this PR is open, draft, conflicting, head cfc4ff4, and closing Per-agent compaction and contextPruning overrides in agents.list #52732. (cfc4ff41b9fc)
Current main does not implement the central request: Current main's strict AgentEntrySchema lists accepted agents.list[] keys without compaction or contextPruning, so current main still rejects the requested per-agent config shape. (src/config/zod-schema.agent-runtime.ts:1033, 15de9d881a84)
PR head imports a missing resolver: The PR imports resolveAgentContextPruningConfig, but the PR-head agent-scope config exports resolveAgentCompactionConfig and not the context-pruning counterpart. (src/agents/embedded-agent-runner/extensions.ts:14, cfc4ff41b9fc)
PR head exports only compaction scoped resolver: The PR-head agent-scope helper exports resolveAgentCompactionConfig but has no resolveAgentContextPruningConfig export for the imported symbol or runtime path. (src/agents/agent-scope-config.ts:201, cfc4ff41b9fc)
PR head omits per-agent contextPruning schema: The PR defines AgentContextPruningSchema and adds compaction to AgentEntrySchema, but agents.list[] still does not accept contextPruning. (src/config/zod-schema.agent-runtime.ts:1141, cfc4ff41b9fc)

Likely related people:

steipete: Recent history shows major agent runtime internalization, normalization/config refactors, compaction planning work, and review involvement on compaction model alias handling across the same resolver/runtime boundary. (role: recent area contributor; confidence: high; commits: bb46b79d3c14, 00d8d7ead059, 6443d067645f; files: src/agents/agent-scope-config.ts, src/config/zod-schema.agent-runtime.ts, src/agents/embedded-agent-runner/extensions.ts)
vincentkoc: Recent commits touch agent-scope helper isolation, config schema cleanup, and compaction reserve/helper cleanup adjacent to this per-agent config resolver work. (role: recent adjacent contributor; confidence: high; commits: a372e4a15240, 42bcb3ecb0fe, 9e3db6bedd28; files: src/agents/agent-scope-config.ts, src/config/zod-schema.agent-runtime.ts, src/agents/embedded-agent-runner/extensions.ts)
Val Alexander: Added per-agent bootstrap profile overrides through the same schema/resolver/docs pattern this PR extends for compaction and context pruning. (role: adjacent feature contributor; confidence: medium; commits: 930852af2959; files: src/agents/agent-scope-config.ts, src/config/zod-schema.agent-runtime.ts, docs/gateway/config-agents.md)
pick-cat: Authored the merged compaction model alias-resolution change in the shared compaction target resolver that this PR must preserve while adding per-agent scope. (role: adjacent compaction model contributor; confidence: medium; commits: b677ea6726d7; files: src/agents/embedded-agent-runner/compaction-runtime-context.ts, src/agents/embedded-agent-runner/compaction-runtime-context.test.ts)
jalehman: Earlier compaction provider and notification work in the related compaction surface was reviewed or co-authored through this owner boundary. (role: adjacent reviewer and merger; confidence: medium; commits: 12331f04631f, f48d040bf523; files: docs/concepts/compaction.md, src/plugins/compaction-provider.ts, src/agents/embedded-agent-runner/compaction-runtime-context.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.

Review history (1 earlier review cycle)

reviewed 2026-06-21T08:32:55.369Z sha cfc4ff4 :: needs real behavior proof before merge. :: [P1] Export and use the scoped context-pruning resolver | [P1] Accept per-agent contextPruning in strict schema | [P2] Use the active agent when appending cache-TTL markers | [P1] Remove the duplicate agentId property

openclaw-barnacle · 2026-07-03T04:37:07Z

This assigned pull request has been automatically marked as stale after being open for 27 days.
Please add updates or it will be closed.

openclaw-barnacle · 2026-07-13T04:37:11Z

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.

feat(compaction): add per-agent compaction overrides

cfc4ff4

kklouzal mentioned this pull request May 31, 2026

feat: add per-agent compaction overrides (AI-assisted) #83637

Closed

kklouzal marked this pull request as draft June 4, 2026 04:49

ferminquant mentioned this pull request Jun 15, 2026

feat(compaction): add compaction.fallbacks: string[] for ordered model fallback chain #93125

Closed

1 task

clawsweeper Bot mentioned this pull request Jun 17, 2026

Safeguard compaction ignores compaction.model config — uses session model instead #57901

Open

clawsweeper Bot mentioned this pull request Jun 30, 2026

feat(config): add compaction.thinkingLevel to override thinking level during compaction #98074

Merged

jalehman self-assigned this Jul 2, 2026

openclaw-barnacle Bot added the stale Marked as stale due to inactivity label Jul 3, 2026

openclaw-barnacle Bot closed this Jul 13, 2026

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

Uh oh!

feat: add per-agent compaction overrides#88506

feat: add per-agent compaction overrides#88506
kklouzal wants to merge 1 commit into
openclaw:mainfrom
kklouzal:feat/agent-compaction-overrides

kklouzal commented May 31, 2026

Uh oh!

clawsweeper Bot commented May 31, 2026 •

edited

Loading

Uh oh!

openclaw-barnacle Bot commented Jul 3, 2026

Uh oh!

openclaw-barnacle Bot commented Jul 13, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

Uh oh!

Uh oh!

Conversation

kklouzal commented May 31, 2026

Summary

Linked Issues And Related Work

Problem / Root Cause

What Changed

Reviewer Notes

Real behavior proof

Validation

Final Merge/Polish Validation

Current Branch State

Uh oh!

clawsweeper Bot commented May 31, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

openclaw-barnacle Bot commented Jul 3, 2026

Uh oh!

openclaw-barnacle Bot commented Jul 13, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

clawsweeper Bot commented May 31, 2026 •

edited

Loading