perf(agents): add continuation-skip context injection by vincentkoc · Pull Request #61268 · openclaw/openclaw

vincentkoc · 2026-04-05T09:18:45Z

Summary

Problem: the embedded attempt path rebuilt workspace bootstrap context on every turn, even when a session had already completed a full assistant turn.
Why it matters: cold-cache, cache-invalidating, and non-Anthropic paths still pay the repeated bootstrap prompt cost, while heartbeat and post-compaction turns need different treatment.
What changed: add opt-in agents.defaults.contextInjection: "continuation-skip", driven by transcript-aware continuation detection that requires a completed assistant turn and forces reinjection after compaction.
What did NOT change (scope boundary): the default remains always, heartbeat runs still use their lightweight bootstrap path, and this PR does not attempt persisted bootstrap fingerprints or broader bootstrap splitting.

Change Type (select all)

Scope (select all touched areas)

Linked Issue/PR

Closes Performance: Workspace file injection wastes 93.5% of token budget #9157
Related perf(agents): add contextInjection option to skip workspace re-injection #55221
This PR fixes a bug or regression

Root Cause (if applicable)

Root cause: runEmbeddedAttempt() always resolved workspace bootstrap context before building the prompt, even when the session transcript already proved that bootstrap had been injected and processed on a prior assistant turn.
Missing detection / guardrail: there was no transcript-aware check for "safe continuation" vs. "first turn" vs. "post-compaction retry", so the only existing behavior was unconditional reinjection.
Contributing context (if known): older attempts drifted into weaker heuristics like "session file exists", which is wrong for pre-created, repaired, or just-compacted transcripts.

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/bootstrap-files.test.ts
- src/agents/pi-embedded-runner/run/attempt.spawn-workspace.context-injection.test.ts
- src/config/zod-schema.agent-defaults.test.ts
Scenario the test should lock in: only skip bootstrap reinjection when contextInjection is opt-in, the transcript already contains a completed assistant turn, the turn is not heartbeat, and no compaction entry trails the last assistant turn.
Why this is the smallest reliable guardrail: helper tests cover transcript edge cases directly, and the run-path seam test proves the real embedded attempt flow honors the helper outcome.
Existing test that already covers this (if any): None.
If no new test is added, why not: N/A.

User-visible / Behavior Changes

New opt-in config: agents.defaults.contextInjection: "continuation-skip".
When enabled, safe continuation turns skip workspace bootstrap reinjection.
Heartbeat runs still resolve lightweight bootstrap context.
Post-compaction retries still rebuild bootstrap context.
Default behavior is unchanged.

Diagram (if applicable)

Before:
[every embedded turn] -> [resolve bootstrap context] -> [build prompt]

After:
[embedded turn] -> [contextInjection mode]
  -> [safe continuation? yes] -> [skip bootstrap reinjection] -> [build prompt]
  -> [first turn / heartbeat / post-compaction] -> [resolve bootstrap context] -> [build prompt]

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: N/A

Repro + Verification

Environment

OS: macOS
Runtime/container: Node 22 / pnpm workspace
Model/provider: embedded runner test harness
Integration/channel (if any): N/A
Relevant config (redacted): agents.defaults.contextInjection: "continuation-skip"

Steps

Enable agents.defaults.contextInjection: "continuation-skip".
Run one embedded turn to establish a completed assistant response in the transcript.
Run a normal follow-up turn, then compare that with a heartbeat turn and a post-compaction retry.

Expected

Normal safe continuation turns skip workspace bootstrap reinjection.
Heartbeat turns still resolve lightweight bootstrap context.
Post-compaction retries re-inject bootstrap context.

Actual

Safe continuation turns now skip bootstrap reinjection only when the transcript proves that is safe.
Heartbeat and post-compaction turns still rebuild context as required.

Evidence

Attach at least one:

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

Human Verification (required)

What you personally verified (not just CI), and how:

Verified scenarios:
- pnpm exec oxlint on the touched source/test files passed.
- pnpm exec oxfmt --check on the touched files passed.
- pnpm test:serial src/agents/bootstrap-files.test.ts src/config/zod-schema.agent-defaults.test.ts src/agents/pi-embedded-runner/run/attempt.spawn-workspace.context-injection.test.ts passed.
- pnpm config:schema:check passed.
- pnpm config:docs:check passed.
Edge cases checked:
- missing or empty session transcript
- header-only / user-only transcript
- compaction after last assistant turn
- assistant turn after compaction
- heartbeat runs bypassing continuation skip
What you did not verify:
- pnpm build is still red on current main because of the existing unresolved import in src/agents/tool-display.ts for ../../apps/shared/OpenClawKit/Sources/OpenClawKit/Resources/tool-display.json; this PR does not touch that path.

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? (Yes)
Migration needed? (No)
If yes, exact upgrade steps: N/A

Risks and Mitigations

Risk: skipping bootstrap too early could drop required workspace instructions.
- Mitigation: continuation skip is opt-in, requires a completed assistant turn, and is disabled for heartbeat and post-compaction retries.
Risk: continuation detection adds transcript-read overhead.
- Mitigation: the scan is line-by-line JSONL, checks only minimal entry fields, and only runs when the opt-in mode is enabled.

Credits

Issue report and problem framing: @ivanenev in Performance: Workspace file injection wastes 93.5% of token budget #9157
Prior-art PR used as source material for the current-main re-roll: @cgdusek in perf(agents): add contextInjection option to skip workspace re-injection #55221
Fresh current-main re-roll, maintainer prep, and landing path: @vincentkoc
AI-assisted: yes

greptile-apps · 2026-04-05T09:35:54Z

Greptile Summary

This PR adds the opt-in agents.defaults.contextInjection: "continuation-skip" feature that skips workspace bootstrap reinjection on safe continuation turns — where the transcript already contains a completed assistant turn and the run is not a heartbeat or post-compaction retry. The implementation is well-scoped: the default remains "always", the new hasCompletedBootstrapTurn helper is unit-tested across all relevant edge cases (symlinks, malformed JSON, compaction ordering), and the embedded-runner seam test proves the flow is wired correctly.

Confidence Score: 5/5

Safe to merge; the only finding is a changelog section misplacement.

All findings are P2 (style/placement). The feature logic, tests, schema changes, and opt-in guard are all correct.

CHANGELOG.md — the new entry should be moved from ### Fixes to ### Changes.

Prompt To Fix All With AI

This is a comment left during a code review.
Path: CHANGELOG.md
Line: 81

Comment:
**Changelog entry placed in wrong section**

This line describes a new opt-in feature (`add opt-in agents.defaults.contextInjection: "continuation-skip"`), so it belongs in `### Changes`, not `### Fixes`. The `Fixes #9157` trailer is an issue reference, not a signal that the entry is a bug fix. Per the project's changelog guidelines, `### Fixes` is for bug fixes and `### Changes` is for new capabilities. This entry should be moved to the end of the `### Changes` block (around line 32–33).

How can I resolve this? If you propose a fix, please make it concise.

_{Reviews (1): Last reviewed commit: "Merge branch 'main' into issue-9157-cont..." | Re-trigger Greptile}

chatgpt-codex-connector

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 99e635d117

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

Open a pull request for review
Mark a draft as ready
Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

chatgpt-codex-connector

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 9d65c36128

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

Open a pull request for review
Mark a draft as ready
Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

chatgpt-codex-connector

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: d0f2f4391f

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

Open a pull request for review
Mark a draft as ready
Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

chatgpt-codex-connector

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: f1d4b1040e

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

Open a pull request for review
Mark a draft as ready
Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

chatgpt-codex-connector · 2026-04-06T04:31:29Z

+    const sessionLock = await acquireSessionWriteLock({
+      sessionFile: params.sessionFile,
+      maxHoldMs: resolveSessionLockMaxHoldFromTimeout({
+        timeoutMs: resolveRunTimeoutWithCompactionGraceMs({
+          runTimeoutMs: params.timeoutMs,


Guard early-acquired session lock with finally

This lock is now acquired before a long pre-session setup section, but it is only released in the later inner try/finally that starts at line 775. If anything throws between these points (for example resolveBootstrapContextForRun(...), tool/runtime setup, or prompt assembly), control never reaches that finally, so the session lock file stays held until the watchdog max-hold timeout and subsequent turns on the same session can fail with lock timeouts. Please wrap all post-acquire setup in a release-guaranteeing finally (or move lock acquisition back next to the guarded block).

Useful? React with 👍 / 👎.

ethangui · 2026-04-06T05:26:02Z

I’m worried the current continuation-skip proof is weaker than the name suggests.

Right now the skip decision seems to be:

contextInjection === "continuation-skip"
not a heartbeat run
transcript contains a prior openclaw:bootstrap-context:full marker
no later compaction entry

That proves a previous turn was executed with full bootstrap, but it does not prove the current prompt still semantically preserves the bootstrap content.

Concrete counterexample:

Turn 1 includes full bootstrap
User says “say hi in one word”
Assistant replies hi
Full-bootstrap marker is recorded
Turn 2 asks something that depends on bootstrap instructions
Bootstrap is now skipped

In that case, the surviving transcript may carry almost none of the bootstrap-derived constraints forward, even though the marker says a full-bootstrap turn happened historically.

So this looks more like a heuristic optimization than a safe continuation proof. That may still be a reasonable tradeoff, but I don’t think the current marker establishes semantic equivalence strongly enough to justify the “safe continuation” framing.

Possible directions:

narrow the claim/documentation to say this is heuristic and may trade some behavioral robustness for token savings
require stronger evidence than “one prior full-bootstrap turn completed”
persist a reduced/structured bootstrap state that is explicitly replayed on continuation turns instead of assuming the transcript implicitly carries it

@mbelinky

* refactor: dedupe discord native command auth * docs: add discord native command changelog note * fix(video): queue fal provider jobs * feat(agents): track video generation tasks * fix(discord): short-circuit bound thread self-loop drops * refactor: harden plugin metadata and bundled channel entry seams * test: fold xai extra params coverage into hot lane * fix: ignore unsupported image generation overrides * docs: document channel persisted auth metadata * test(live): prefer google models over big-pickle * Lobster: run workflows in process (openclaw#61523) * Lobster: run workflows in process * docs: note in-process lobster runtime * docs: add lobster changelog attribution * Lobster: add managed TaskFlow mode (openclaw#61555) * test: split inline provider model coverage * docs: update Lobster in-process mode and REM preview tooling * test: speed up nodes camera coverage * fix: defer plugin sync after git switch * test: optimize macos release-to-dev smoke lane * fix(openai): avoid em dashes in gpt-5 overlay (openclaw#61560) * feat(agents): detach video generation completion * feat(video): add runway provider * docs(video): document runway support * fix: clarify dirty dev update error * fix: ignore unsupported video generation overrides * refactor: add metadata-first channel configured-state probes * fix(video): guard active async generation tasks * docs(providers): surface new video provider pages * feat(qa): add live suite runner and harness * feat(qa): improve qa lab debugger ui * fix: restore pnpm check type safety * test: trim slow agent web and lifecycle coverage * fix: restore green checks * fix(qa): stop embedded control ui reload loop * test: reset guest git root before dev update * test: speed up openai tool id preservation replay coverage * fix: restore qa lab config typing * matrix: align bundled channel metadata * docs: note Matrix persisted auth detection * docs: add changelog note for qa lab config fix * refactor(video): share async task status helpers * memory-core: checkpoint mode-first dreaming refactor * Dreaming: simplify sweep flow and add diary surface * docs: rewrite video generation docs for readability * docs(faq): add gpt-5.4 fast mode entry * feat(memory): add Bedrock embedding provider for memory search (openclaw#61547) * feat(memory): add Bedrock embedding provider for memory search Add Amazon Bedrock as a native embedding provider for memory search. Supports Titan Embed Text v1/v2 and Cohere Embed models via AWS SDK. - New embeddings-bedrock.ts: BedrockRuntimeClient + InvokeModel - Auth via AWS default credential chain (same as Bedrock inference) - Auto-selected in 'auto' mode when AWS credentials are detected - Titan V2: configurable dimensions (256/512/1024), normalization - Cohere: native batch support with search_query/search_document types - 16 new tests covering all model types, auth detection, edge cases Closes openclaw#26289 * fix(memory): harden bedrock embedding selection --------- Co-authored-by: Vincent Koc <vincentkoc@ieee.org> * docs(openai): clarify gpt-5.4 fast mode * test: speed up models config env provider coverage * test: speed up sanitize session history policy smoke * build: refresh lockfile for control ui deps * refactor: narrow bundled channel entry surfaces * test: speed up sanitize session history coverage * fix: skip old-process config writes after git switch * fix(update): bootstrap pnpm for dev preflight * fix(memory-qmd): restore qmd compatibility defaults * test: speed up image tool auth-heavy coverage * test: seed channel setup contract registry in helper tests * Dreaming: update multiphase stats and UI polish * test: add irc runtime api smoke coverage * feat(bedrock-mantle): add IAM credential auth via @aws/bedrock-token-… (openclaw#61563) * feat(bedrock-mantle): add IAM credential auth via @aws/bedrock-token-generator Mantle previously required a manually-created API key (AWS_BEARER_TOKEN_BEDROCK). This adds automatic bearer token generation from IAM credentials using the official @aws/bedrock-token-generator package. Auth priority: 1. Explicit AWS_BEARER_TOKEN_BEDROCK env var (manual API key from Console) 2. IAM credentials via getTokenProvider() → Bearer token (instance roles, SSO profiles, access keys, EKS IRSA, ECS task roles) Token is cached in memory (1hr TTL, generated with 2hr validity) and in process.env.AWS_BEARER_TOKEN_BEDROCK for downstream sync reads. Falls back gracefully when package is not installed or credentials are unavailable — Mantle provider simply not registered. Closes openclaw#45152 * fix(bedrock-mantle): harden IAM auth --------- Co-authored-by: Vincent Koc <vincentkoc@ieee.org> * refactor(update): extract package manager bootstrap logic * feat: add comfy workflow media support * fix: stabilize line and feishu ci shards * feat: add music generation tooling * chore: remove stray finder metadata * docs: document music generation async flow * fix(memory-qmd): streamline compatibility coverage * test: speed up dispatch-from-config thread fallback coverage * docs: improve music generation docs * docs: reorder changelog highlights * fix: skip stale post-switch update follow-ups * test: harden macos release-to-dev smoke verification * fix: route comfy music through shared tool * refactor: remove comfy music tool shim * Gateway: bound websocket shutdown close (openclaw#61565) Merged via squash. Prepared head SHA: 9040dd5 Co-authored-by: mbelinky <132747814+mbelinky@users.noreply.github.com> Reviewed-by: @mbelinky * Docs: clarify Matrix quiet push rules * memory: chunk daily dreaming ingestion (openclaw#61583) Merged via squash. Prepared head SHA: 88816a0 Co-authored-by: mbelinky <17249097+mbelinky@users.noreply.github.com> Co-authored-by: mbelinky <132747814+mbelinky@users.noreply.github.com> Reviewed-by: @mbelinky * fix: stop old cli after package-to-git switch * fix(gateway): accept music generation internal events * docs: update unreleased provider notes * fix(agents): keep large read tool results visible * feat: add vydra media provider * fix(agents): ignore unsupported music generation hints * fix(agents): preserve latest read output during compaction * docs: update changelog for read visibility fixes * test: fix current-main prep blockers (openclaw#61582) Merged via squash. Prepared head SHA: 49f7b12 Reviewed-by: @mbelinky * test: use explicit node entrypoint in macos update smoke * fix: exit after package-to-git handoff * fix: prune staged feishu sdk types from npm pack * fix(qa): harden new scenario suite * fix(agents): prefer overflow compaction for fresh reads * perf(auto-reply): lazy-load TTS helpers on demand * test(plugin-sdk): tighten ACP command dispatch guards * docs(web): clarify control ui language picker * test(auto-reply): split ACP and reply-dispatch regressions * memory: trim generic daily chunk headings (openclaw#61597) * memory: trim generic daily chunk headings * docs: tag dreaming heading cleanup changelog * docs: attribute dreaming heading cleanup changelog * fix(cli): narrow post-update root * fix(ui): localize control ui strings * Lobster: harden embedded runtime integration (openclaw#61566) Merged via squash. Prepared head SHA: a6f4830 Co-authored-by: mbelinky <132747814+mbelinky@users.noreply.github.com> Reviewed-by: @mbelinky * fix(matrix): reuse raw default account key during onboarding promotion * fix: unblock comfy live plugin loading * fix(agents): extend subagent announce timeout * fix(agents): carry async media wake attachments structurally * fix(tasks): hide internal completion wake rows * test(auto-reply): isolate reply abort dispatch seams * test: fix reply dispatch mock contract * fix(ui): localize more control ui strings * fix: deliver async media generation results directly * perf(test): trim send-policy and abort hot paths * perf(agents): isolate subagent announce origin helper * fix(discord): raise default media cap * Matrix: recover from pinned dispatcher runtime failures (openclaw#61595) Merged via squash. Prepared head SHA: f9a2d9b Co-authored-by: gumadeiras <5599352+gumadeiras@users.noreply.github.com> Co-authored-by: gumadeiras <5599352+gumadeiras@users.noreply.github.com> Reviewed-by: @gumadeiras * fix: harden async media completion delivery * fix: gate async media direct delivery behind config * docs: add changelog note for async media delivery flag * perf(test): trim announce and sessions tool imports * fix: resolve global bundled plugin facade fallback (openclaw#61297) (thanks @openperf) * fix(gateway): resolve globally-installed bundled plugins in facade-runtime * fix: resolve global bundled plugin facade fallback (openclaw#61297) (thanks @openperf) --------- Co-authored-by: Ayaan Zaidi <hi@obviy.us> * chore: prepare 2026.4.6-beta.1 release * style: trim facade fallback comment noise * test: stabilize browser and provider ci shards * fix: restore latest-main ci gates * (chore): delete dream-diary-preview file * perf(test): trim runReplyAgent misc mock imports * fix(ci): harden control ui locale refresh rebases * Matrix: clear undici test override after transport test * chore(ui): refresh zh-CN control ui locale * chore(ui): refresh pt-BR control ui locale * chore(ui): refresh zh-TW control ui locale * chore(ui): refresh de control ui locale * fix: support corepack cmd shim on windows * test: add windows dev-update smoke lanes * chore(ui): refresh es control ui locale * chore(ui): refresh ja-JP control ui locale * chore(ui): refresh ko control ui locale * chore(ui): refresh fr control ui locale * test: capture windows npm debug tails in smoke logs * chore(ui): refresh tr control ui locale * chore(ui): refresh uk control ui locale * chore(ui): refresh id control ui locale * chore(ui): refresh pl control ui locale * fix: restore plugin boundary and ui locale ci gates * fix(ci): stabilize control ui locale checks * chore: release 2026.4.5 * perf(test): split subagent command coverage * fix(ci): patch main regression surfaces * fix: install bun in npm release preflight * test: fix subagent command result assertions * perf(test): split allowlist and models command coverage * fix(openai): allow qa image generation mock routing * feat(qa): execute ten new repo-backed scenarios * fix(matrix): harden startup auth bootstrap (openclaw#61383) Merged via squash. Prepared head SHA: d8011a9 Co-authored-by: gumadeiras <5599352+gumadeiras@users.noreply.github.com> Co-authored-by: gumadeiras <5599352+gumadeiras@users.noreply.github.com> Reviewed-by: @gumadeiras * Docs: clarify Matrix autoJoin invite scope * fix(discord): narrow binding runtime imports * fix: stabilize contract loader seams * test: tighten allowlist fixture typing * fix(qa): support image understanding inputs * feat(qa): add attachment understanding scenario * docs(matrix): clarify historyLimit default * feat(memory-wiki): restore llm wiki stack * chore: update appcast for 2026.4.5 * perf(test): split reply command coverage * perf(reply): lazy load compact runtime * refactor(reply): extract subagent text helper * style(reply): normalize subagent import order * fix: restore protocol and extension ci * chore: bump version to 2026.4.6 * fix(config): normalize channel streaming config shape (openclaw#61381) * feat(config): add canonical streaming config helpers * refactor(runtime): prefer canonical streaming accessors * feat(config): normalize preview channel streaming shape * test(config): lock streaming normalization followups * fix(config): polish streaming migration edges * chore(config): refresh streaming baseline hash * docs(memory): add promote-explain and rem-harness CLI reference * build: refresh pnpm lockfile * fix: stop emitting post-background exec updates (openclaw#61627) (thanks @openperf) * fix(exec ): stop emitting tool updates after session is backgrounded When an exec session is backgrounded (background: true), the owning agent run resolves its tool-call promise and may finish. The stdout handler's emitUpdate() closure, however, kept invoking opts.onUpdate(), delivering tool_execution_update events to a listener whose active run had already ended. This surfaced as an unhandled rejection and crashed the gateway process. Guard emitUpdate() with a session.backgrounded || session.exited check so that post-background output is still captured via appendOutput() but no longer forwarded to the (now-stale) agent-loop callback. Fixes openclaw#61592 * style: trim exec backgrounding comments * fix: stop emitting post-background exec updates (openclaw#61627) (thanks @openperf) * fix: place exec changelog entry at end of fixes (openclaw#61627) (thanks @openperf) --------- Co-authored-by: Ayaan Zaidi <hi@obviy.us> * test(memory-core): align dreaming expectations * test(memory-wiki): share plugin test helpers * test(memory-core): share workspace test helper * test(memory-core): reuse narrative workspace helper * test(plugin-sdk): share temp dir test helper * test(plugin-sdk): reuse temp dir helpers in facade tests * test(memory-core): reuse workspace helper in dreaming tests * perf(agents): add continuation-skip context injection (openclaw#61268) * test(agents): cover continuation bootstrap reuse * perf(agents): add continuation-skip context injection * docs(changelog): note context injection reuse * perf(agents): bound continuation bootstrap scan * fix(agents): require full bootstrap proof for continuation skip * fix(agents): decide continuation skip under lock * fix(commands): re-export subagent chat message type * fix(agents): clean continuation rebase leftovers * test(memory-core): reuse workspace helper in temp dir tests * docs: add contextInjection config key to reference * test(scripts): share temp dir helpers * test(scripts): reuse temp dir helpers in runtime tests * test(scripts): reuse temp dir helpers in repo fixtures * test(scripts): add async temp dir helper * fix: restore main ci type checks * test(root): reuse temp repo helper in clawhub release tests * test(root): clean up pre-commit temp repos * fix(matrix): pass deviceId through health probe to prevent storage-meta overwrite (openclaw#61317) (openclaw#61581) Merged via squash. Prepared head SHA: b0495dc Co-authored-by: MoerAI <26067127+MoerAI@users.noreply.github.com> Co-authored-by: gumadeiras <5599352+gumadeiras@users.noreply.github.com> Reviewed-by: @gumadeiras * test(root): share temp dir helper across root tests * test(root): reuse temp dir helper in scoped vitest config * test(root): reuse temp dir helper in launcher e2e * test(tooling): reuse temp dir helpers in script tests * test(unit): reuse temp dir helper in install-sh version tests * test: reset telegram dispatch mocks between cases * test(plugins): reuse tracked temp helpers in runtime staging tests * test(plugins): reuse tracked temp helpers in path resolution tests * test(plugins): reuse tracked temp helpers in fixture tests * test(plugins): share async temp helpers in marketplace tests * test(plugins): reuse tracked temp helpers in loader fixture tests * test(plugins): share suite temp root helper in install path tests * test(plugins): reuse suite temp root helper in install fixture tests * test(plugins): reuse tracked temp helpers in package contract tests * test(plugins): reuse suite temp helper in bundle contract test * test(infra): reuse shared temp dir helpers in small file tests * test(infra): reuse temp dir helper in utility file tests * test(infra): reuse temp dir helper in run-node tests * test(infra): reuse temp dir helpers in install source tests * test(infra): reuse temp dir helper in install path safety tests * perf(test): split reply command coverage * perf(test): trim subagent command imports * test: remove legacy commands monolith * test(infra): reuse temp dir helper in node path tests * test(infra): reuse temp dir helper in state and watch tests * test(infra): reuse temp dir helper in sentinel and provider tests * test(infra): share temp dir cleanup in git metadata tests * test(infra): share tracked temp dirs in apns tests * test(infra): reuse temp dir helper in fs safety tests * test(infra): reuse temp dir helper in update status tests * test(infra): share sync temp dir helper in approval tests * test(infra): share suite temp root tracker in infra tests * test(infra): reuse suite temp root tracker in update tests * test(infra): reuse suite temp root tracker in provider auth tests * test(infra): reuse suite temp root tracker in install tests * test(infra): reuse suite temp root tracker in startup checks * test(infra): reuse temp dir helper in global update tests * test(infra): reuse temp dir helper in clawhub tests * test(core): reuse shared temp dir helpers in utils tests * test(infra): reuse temp dir helper in node pairing tests * test(infra): reuse suite temp root tracker in device pairing tests * test(core): reuse shared temp dir helper in logger tests * test(e2e): reuse suite temp root tracker in docker setup tests * test(infra): reuse suite temp root tracker in session cost tests * test(config): reuse temp dir helper in config surface tests * test(config): reuse temp dir helper in disk budget tests * test(config): share session test fixture helper * test(config): reuse suite temp root tracker in session key normalization tests * test(config): reuse suite temp root tracker in store pruning integration tests * test(config): reuse shared temp dir helpers in sessions tests * test(config): reuse shared temp dir helper in store read tests * perf(test): split subagent command coverage * perf(test): trim secrets runtime coverage * perf(test): split extra params resolver coverage * fix(anthropic): restore OAuth guard in service-tier stream wrappers (openclaw#60356) Merged via squash. Prepared head SHA: 7d58bef Co-authored-by: openperf <80630709+openperf@users.noreply.github.com> Co-authored-by: jalehman <550978+jalehman@users.noreply.github.com> Reviewed-by: @jalehman * perf(test): split extra params wrapper coverage * perf(secrets): trim runtime import walls * perf(test): split security audit coverage * refactor: dedupe plugin and outbound helpers * refactor: share gateway auth and approval helpers * refactor: share command config resolution * refactor: consolidate status reporting helpers * fix: resolve upstream sync conflicts (branding, firecrawl, lockfile) Resolve 7 merge conflicts from sync/upstream-2026-04-06 (v2026.4.5): - pnpm-lock.yaml: keep our platform-channel + upstream's qa-channel/qa-lab - app-render.ts: add upstream session-key imports, deduplicate agentLogoUrl - control-ui-bootstrap.ts: keep our branding (resolveUiBrand, title, agentId) - control-ui-bootstrap.test.ts: keep our test expectations + upstream null checks - schema.base.generated.ts: keep our Firecrawl + profile config entries - schema.labels.ts: keep our Firecrawl + profile labels Includes CVE-2026-33579 fix (callerScopes in /pair approve handler). Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> --------- Co-authored-by: Peter Steinberger <steipete@gmail.com> Co-authored-by: Gustavo Madeira Santana <gumadeiras@gmail.com> Co-authored-by: Mariano <132747814+mbelinky@users.noreply.github.com> Co-authored-by: Vincent Koc <vincentkoc@ieee.org> Co-authored-by: Vignesh Natarajan <vignesh.natarajan92@gmail.com> Co-authored-by: wirjo <daniel@wirjo.com> Co-authored-by: Mariano <mbelinky@gmail.com> Co-authored-by: mbelinky <17249097+mbelinky@users.noreply.github.com> Co-authored-by: gumadeiras <5599352+gumadeiras@users.noreply.github.com> Co-authored-by: Chunyue Wang <80630709+openperf@users.noreply.github.com> Co-authored-by: Ayaan Zaidi <hi@obviy.us> Co-authored-by: Vignesh <mailvgnsh@gmail.com> Co-authored-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com> Co-authored-by: ToToKr <friendnt@g.skku.edu> Co-authored-by: MoerAI <26067127+MoerAI@users.noreply.github.com> Co-authored-by: jalehman <550978+jalehman@users.noreply.github.com> Co-authored-by: github-actions[bot] <github-actions[bot]@users.noreply.github.com> Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>

@gumadeiras

* feat(memory-wiki): add import gateway methods * feat(memory-wiki): add shared memory search bridge * feat(memory-wiki): add prompt supplement integration * feat(memory-wiki): surface imported source provenance * feat(memory-wiki): lint imported provenance gaps * feat(memory-wiki): allow per-call search corpus overrides * feat(memory-core): bridge wiki corpus into memory tools * feat(memory-wiki): compile related backlinks blocks * docs(memory-wiki): prefer shared corpus recall guidance * docs(memory-wiki): document shared recall and backlinks * feat(memory-wiki): generate dashboard report pages * test: isolate exec approval suite from bundled plugins * fix(sandbox): harden EXDEV rename fallback * Gateway: keep outbound session metadata in owner store * revert(memory-wiki): back out llm wiki stack * fix: align models status provider auth reporting * fix(ci): narrow control ui locale refresh push runs * style: format remaining local edits * fix: prevent duplicate block reply delivery for text_end channels (openclaw#61530) * fix(gateway): bound silent local pairing scopes * fix: resolve repo check drift * fix: clean rebase leftovers * test: isolate agent runtime seams * docs: refine unreleased changelog * feat(video): add xai and alibaba providers * Revert "fix(gateway): bound silent local pairing scopes" This reverts commit 7f1b159. * fix(build): correct node require typing * docs(security): clarify localhost shared-auth trust model * refactor: move browser runtime seams behind plugin metadata * test: speed up provider policy and auth suites * Memory: move dreaming trail to dreams.md (openclaw#61537) * Memory: move dreaming trail to dreams.md * docs(changelog): add dreams.md entry --------- Co-authored-by: Vincent Koc <vincentkoc@ieee.org> * fix(ci): stabilize ui i18n and gateway watch checks * docs(providers): add generation setup pages * docs(providers): link generation guides * feat: add qa channel foundation * refactor: hide qa channels with exposure metadata * feat: add qa lab extension * chore: polish qa lab follow-ups * feat(qa): recreate qa lab docker stack * fix(qa): restore embedded control ui gateway startup * fix(qa): stabilize docker gateway bootstrap * feat(qa): add repo-backed qa suite runner * fix(qa): stabilize hermetic suite runtime * fix(matrix): split partial and quiet preview streaming (openclaw#61450) Merged via squash. Prepared head SHA: 6a0d7d1 Co-authored-by: gumadeiras <5599352+gumadeiras@users.noreply.github.com> Reviewed-by: @gumadeiras * docs(providers): unify qwen docs * fix(matrix): honor canonical private-network opt-in * fix(matrix): restore cli metadata registrar * test(matrix): isolate migration snapshot seam * fix: prevent duplicate gateway watchers * feat(memory-core): add REM preview and safe promotion replay (openclaw#61540) * memory: add REM preview and safe promotion replay thanks @mbelinky * changelog: note REM preview and promotion replay --------- Co-authored-by: Vignesh <mailvgnsh@gmail.com> * test: fix abort cascade and workspace edit inputs * refactor: harden plugin metadata and browser sdk seams * fix(memory-core): preserve dated DREAMS trail * docs(memory): point dreaming trail docs to dreams.md * fix(memory): standardize DREAMS trail path * fix(google): restore gemini cli provider contract * test(contracts): drop removed claude cli auth export * test(config): align markdown tables with active registry * style(tests): normalize registry mock wrapping * fix: normalize video provider durations * fix: harden video provider transports * fix: honor discord allowlisted channels for native commands * fix: bootstrap pnpm for git updates * docs: add tahoe release-to-dev smoke lane * test: isolate openclaw plugin context coverage * test: stabilize subagent persistence registry coverage * test: isolate gateway tool coverage * fix: surface normalized video durations * fix(google): restore forward-compat provider hooks * test(config): fix markdown table mock typing * test: drop redundant openai extra params coverage * refactor: dedupe discord native command auth * docs: add discord native command changelog note * fix(video): queue fal provider jobs * feat(agents): track video generation tasks * fix(discord): short-circuit bound thread self-loop drops * refactor: harden plugin metadata and bundled channel entry seams * test: fold xai extra params coverage into hot lane * fix: ignore unsupported image generation overrides * docs: document channel persisted auth metadata * test(live): prefer google models over big-pickle * Lobster: run workflows in process (openclaw#61523) * Lobster: run workflows in process * docs: note in-process lobster runtime * docs: add lobster changelog attribution * Lobster: add managed TaskFlow mode (openclaw#61555) * test: split inline provider model coverage * docs: update Lobster in-process mode and REM preview tooling * test: speed up nodes camera coverage * fix: defer plugin sync after git switch * test: optimize macos release-to-dev smoke lane * fix(openai): avoid em dashes in gpt-5 overlay (openclaw#61560) * feat(agents): detach video generation completion * feat(video): add runway provider * docs(video): document runway support * fix: clarify dirty dev update error * fix: ignore unsupported video generation overrides * refactor: add metadata-first channel configured-state probes * fix(video): guard active async generation tasks * docs(providers): surface new video provider pages * feat(qa): add live suite runner and harness * feat(qa): improve qa lab debugger ui * fix: restore pnpm check type safety * test: trim slow agent web and lifecycle coverage * fix: restore green checks * fix(qa): stop embedded control ui reload loop * test: reset guest git root before dev update * test: speed up openai tool id preservation replay coverage * fix: restore qa lab config typing * matrix: align bundled channel metadata * docs: note Matrix persisted auth detection * docs: add changelog note for qa lab config fix * refactor(video): share async task status helpers * memory-core: checkpoint mode-first dreaming refactor * Dreaming: simplify sweep flow and add diary surface * docs: rewrite video generation docs for readability * docs(faq): add gpt-5.4 fast mode entry * feat(memory): add Bedrock embedding provider for memory search (openclaw#61547) * feat(memory): add Bedrock embedding provider for memory search Add Amazon Bedrock as a native embedding provider for memory search. Supports Titan Embed Text v1/v2 and Cohere Embed models via AWS SDK. - New embeddings-bedrock.ts: BedrockRuntimeClient + InvokeModel - Auth via AWS default credential chain (same as Bedrock inference) - Auto-selected in 'auto' mode when AWS credentials are detected - Titan V2: configurable dimensions (256/512/1024), normalization - Cohere: native batch support with search_query/search_document types - 16 new tests covering all model types, auth detection, edge cases Closes openclaw#26289 * fix(memory): harden bedrock embedding selection --------- Co-authored-by: Vincent Koc <vincentkoc@ieee.org> * docs(openai): clarify gpt-5.4 fast mode * test: speed up models config env provider coverage * test: speed up sanitize session history policy smoke * build: refresh lockfile for control ui deps * refactor: narrow bundled channel entry surfaces * test: speed up sanitize session history coverage * fix: skip old-process config writes after git switch * fix(update): bootstrap pnpm for dev preflight * fix(memory-qmd): restore qmd compatibility defaults * test: speed up image tool auth-heavy coverage * test: seed channel setup contract registry in helper tests * Dreaming: update multiphase stats and UI polish * test: add irc runtime api smoke coverage * feat(bedrock-mantle): add IAM credential auth via @aws/bedrock-token-… (openclaw#61563) * feat(bedrock-mantle): add IAM credential auth via @aws/bedrock-token-generator Mantle previously required a manually-created API key (AWS_BEARER_TOKEN_BEDROCK). This adds automatic bearer token generation from IAM credentials using the official @aws/bedrock-token-generator package. Auth priority: 1. Explicit AWS_BEARER_TOKEN_BEDROCK env var (manual API key from Console) 2. IAM credentials via getTokenProvider() → Bearer token (instance roles, SSO profiles, access keys, EKS IRSA, ECS task roles) Token is cached in memory (1hr TTL, generated with 2hr validity) and in process.env.AWS_BEARER_TOKEN_BEDROCK for downstream sync reads. Falls back gracefully when package is not installed or credentials are unavailable — Mantle provider simply not registered. Closes openclaw#45152 * fix(bedrock-mantle): harden IAM auth --------- Co-authored-by: Vincent Koc <vincentkoc@ieee.org> * refactor(update): extract package manager bootstrap logic * feat: add comfy workflow media support * fix: stabilize line and feishu ci shards * feat: add music generation tooling * chore: remove stray finder metadata * docs: document music generation async flow * fix(memory-qmd): streamline compatibility coverage * test: speed up dispatch-from-config thread fallback coverage * docs: improve music generation docs * docs: reorder changelog highlights * fix: skip stale post-switch update follow-ups * test: harden macos release-to-dev smoke verification * fix: route comfy music through shared tool * refactor: remove comfy music tool shim * Gateway: bound websocket shutdown close (openclaw#61565) Merged via squash. Prepared head SHA: 9040dd5 Co-authored-by: mbelinky <132747814+mbelinky@users.noreply.github.com> Reviewed-by: @mbelinky * Docs: clarify Matrix quiet push rules * memory: chunk daily dreaming ingestion (openclaw#61583) Merged via squash. Prepared head SHA: 88816a0 Co-authored-by: mbelinky <17249097+mbelinky@users.noreply.github.com> Co-authored-by: mbelinky <132747814+mbelinky@users.noreply.github.com> Reviewed-by: @mbelinky * fix: stop old cli after package-to-git switch * fix(gateway): accept music generation internal events * docs: update unreleased provider notes * fix(agents): keep large read tool results visible * feat: add vydra media provider * fix(agents): ignore unsupported music generation hints * fix(agents): preserve latest read output during compaction * docs: update changelog for read visibility fixes * test: fix current-main prep blockers (openclaw#61582) Merged via squash. Prepared head SHA: 49f7b12 Reviewed-by: @mbelinky * test: use explicit node entrypoint in macos update smoke * fix: exit after package-to-git handoff * fix: prune staged feishu sdk types from npm pack * fix(qa): harden new scenario suite * fix(agents): prefer overflow compaction for fresh reads * perf(auto-reply): lazy-load TTS helpers on demand * test(plugin-sdk): tighten ACP command dispatch guards * docs(web): clarify control ui language picker * test(auto-reply): split ACP and reply-dispatch regressions * memory: trim generic daily chunk headings (openclaw#61597) * memory: trim generic daily chunk headings * docs: tag dreaming heading cleanup changelog * docs: attribute dreaming heading cleanup changelog * fix(cli): narrow post-update root * fix(ui): localize control ui strings * Lobster: harden embedded runtime integration (openclaw#61566) Merged via squash. Prepared head SHA: a6f4830 Co-authored-by: mbelinky <132747814+mbelinky@users.noreply.github.com> Reviewed-by: @mbelinky * fix(matrix): reuse raw default account key during onboarding promotion * fix: unblock comfy live plugin loading * fix(agents): extend subagent announce timeout * fix(agents): carry async media wake attachments structurally * fix(tasks): hide internal completion wake rows * test(auto-reply): isolate reply abort dispatch seams * test: fix reply dispatch mock contract * fix(ui): localize more control ui strings * fix: deliver async media generation results directly * perf(test): trim send-policy and abort hot paths * perf(agents): isolate subagent announce origin helper * fix(discord): raise default media cap * Matrix: recover from pinned dispatcher runtime failures (openclaw#61595) Merged via squash. Prepared head SHA: f9a2d9b Co-authored-by: gumadeiras <5599352+gumadeiras@users.noreply.github.com> Co-authored-by: gumadeiras <5599352+gumadeiras@users.noreply.github.com> Reviewed-by: @gumadeiras * fix: harden async media completion delivery * fix: gate async media direct delivery behind config * docs: add changelog note for async media delivery flag * perf(test): trim announce and sessions tool imports * fix: resolve global bundled plugin facade fallback (openclaw#61297) (thanks @openperf) * fix(gateway): resolve globally-installed bundled plugins in facade-runtime * fix: resolve global bundled plugin facade fallback (openclaw#61297) (thanks @openperf) --------- Co-authored-by: Ayaan Zaidi <hi@obviy.us> * chore: prepare 2026.4.6-beta.1 release * style: trim facade fallback comment noise * test: stabilize browser and provider ci shards * fix: restore latest-main ci gates * (chore): delete dream-diary-preview file * perf(test): trim runReplyAgent misc mock imports * fix(ci): harden control ui locale refresh rebases * Matrix: clear undici test override after transport test * chore(ui): refresh zh-CN control ui locale * chore(ui): refresh pt-BR control ui locale * chore(ui): refresh zh-TW control ui locale * chore(ui): refresh de control ui locale * fix: support corepack cmd shim on windows * test: add windows dev-update smoke lanes * chore(ui): refresh es control ui locale * chore(ui): refresh ja-JP control ui locale * chore(ui): refresh ko control ui locale * chore(ui): refresh fr control ui locale * test: capture windows npm debug tails in smoke logs * chore(ui): refresh tr control ui locale * chore(ui): refresh uk control ui locale * chore(ui): refresh id control ui locale * chore(ui): refresh pl control ui locale * fix: restore plugin boundary and ui locale ci gates * fix(ci): stabilize control ui locale checks * chore: release 2026.4.5 * perf(test): split subagent command coverage * fix(ci): patch main regression surfaces * fix: install bun in npm release preflight * test: fix subagent command result assertions * perf(test): split allowlist and models command coverage * fix(openai): allow qa image generation mock routing * feat(qa): execute ten new repo-backed scenarios * fix(matrix): harden startup auth bootstrap (openclaw#61383) Merged via squash. Prepared head SHA: d8011a9 Co-authored-by: gumadeiras <5599352+gumadeiras@users.noreply.github.com> Co-authored-by: gumadeiras <5599352+gumadeiras@users.noreply.github.com> Reviewed-by: @gumadeiras * Docs: clarify Matrix autoJoin invite scope * fix(discord): narrow binding runtime imports * fix: stabilize contract loader seams * test: tighten allowlist fixture typing * fix(qa): support image understanding inputs * feat(qa): add attachment understanding scenario * docs(matrix): clarify historyLimit default * feat(memory-wiki): restore llm wiki stack * chore: update appcast for 2026.4.5 * perf(test): split reply command coverage * perf(reply): lazy load compact runtime * refactor(reply): extract subagent text helper * style(reply): normalize subagent import order * fix: restore protocol and extension ci * chore: bump version to 2026.4.6 * fix(config): normalize channel streaming config shape (openclaw#61381) * feat(config): add canonical streaming config helpers * refactor(runtime): prefer canonical streaming accessors * feat(config): normalize preview channel streaming shape * test(config): lock streaming normalization followups * fix(config): polish streaming migration edges * chore(config): refresh streaming baseline hash * docs(memory): add promote-explain and rem-harness CLI reference * build: refresh pnpm lockfile * fix: stop emitting post-background exec updates (openclaw#61627) (thanks @openperf) * fix(exec ): stop emitting tool updates after session is backgrounded When an exec session is backgrounded (background: true), the owning agent run resolves its tool-call promise and may finish. The stdout handler's emitUpdate() closure, however, kept invoking opts.onUpdate(), delivering tool_execution_update events to a listener whose active run had already ended. This surfaced as an unhandled rejection and crashed the gateway process. Guard emitUpdate() with a session.backgrounded || session.exited check so that post-background output is still captured via appendOutput() but no longer forwarded to the (now-stale) agent-loop callback. Fixes openclaw#61592 * style: trim exec backgrounding comments * fix: stop emitting post-background exec updates (openclaw#61627) (thanks @openperf) * fix: place exec changelog entry at end of fixes (openclaw#61627) (thanks @openperf) --------- Co-authored-by: Ayaan Zaidi <hi@obviy.us> * test(memory-core): align dreaming expectations * test(memory-wiki): share plugin test helpers * test(memory-core): share workspace test helper * test(memory-core): reuse narrative workspace helper * test(plugin-sdk): share temp dir test helper * test(plugin-sdk): reuse temp dir helpers in facade tests * test(memory-core): reuse workspace helper in dreaming tests * perf(agents): add continuation-skip context injection (openclaw#61268) * test(agents): cover continuation bootstrap reuse * perf(agents): add continuation-skip context injection * docs(changelog): note context injection reuse * perf(agents): bound continuation bootstrap scan * fix(agents): require full bootstrap proof for continuation skip * fix(agents): decide continuation skip under lock * fix(commands): re-export subagent chat message type * fix(agents): clean continuation rebase leftovers * test(memory-core): reuse workspace helper in temp dir tests * docs: add contextInjection config key to reference * feat(gateway): preserve session history on /new command Backend changes for session sidebar feature: - session.ts: create compound key entry for old session when /new triggered - agent.ts: pass preserveHistory=true to session reset - session-reset-service.ts: add file reuse optimization for preserved sessions - types.ts: add previousSessionKey field to SessionEntry - session-utils.ts: add compound key support in session key resolution This enables the UI to show previous sessions in sidebar while preserving complete chat history for archived sessions. --------- Co-authored-by: Vincent Koc <vincentkoc@ieee.org> Co-authored-by: Peter Steinberger <steipete@gmail.com> Co-authored-by: Gustavo Madeira Santana <gumadeiras@gmail.com> Co-authored-by: Tyler Yust <64381258+tyler6204@users.noreply.github.com> Co-authored-by: Dave Morin <dave@morin.com> Co-authored-by: gumadeiras <5599352+gumadeiras@users.noreply.github.com> Co-authored-by: Mariano <132747814+mbelinky@users.noreply.github.com> Co-authored-by: Vignesh <mailvgnsh@gmail.com> Co-authored-by: Vignesh Natarajan <vignesh.natarajan92@gmail.com> Co-authored-by: wirjo <daniel@wirjo.com> Co-authored-by: Mariano <mbelinky@gmail.com> Co-authored-by: mbelinky <17249097+mbelinky@users.noreply.github.com> Co-authored-by: Chunyue Wang <80630709+openperf@users.noreply.github.com> Co-authored-by: Ayaan Zaidi <hi@obviy.us> Co-authored-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com> Co-authored-by: OpenClaw Agent <caoha@openclaw.ai>

* test(agents): cover continuation bootstrap reuse * perf(agents): add continuation-skip context injection * docs(changelog): note context injection reuse * perf(agents): bound continuation bootstrap scan * fix(agents): require full bootstrap proof for continuation skip * fix(agents): decide continuation skip under lock * fix(commands): re-export subagent chat message type * fix(agents): clean continuation rebase leftovers

vincentkoc self-assigned this Apr 5, 2026

openclaw-barnacle Bot added docs Improvements or additions to documentation agents Agent runtime and tooling size: M maintainer Maintainer-authored PR labels Apr 5, 2026

vincentkoc marked this pull request as ready for review April 5, 2026 09:32

greptile-apps Bot reviewed Apr 5, 2026

View reviewed changes

Comment thread CHANGELOG.md

chatgpt-codex-connector Bot reviewed Apr 5, 2026

View reviewed changes

Comment thread src/agents/bootstrap-files.ts Outdated

chatgpt-codex-connector Bot reviewed Apr 5, 2026

View reviewed changes

Comment thread src/agents/pi-embedded-runner/run/attempt.ts

openclaw-barnacle Bot added size: L and removed size: M labels Apr 5, 2026

chatgpt-codex-connector Bot reviewed Apr 5, 2026

View reviewed changes

Comment thread src/agents/pi-embedded-runner/run/attempt.ts

vincentkoc added 8 commits April 6, 2026 05:21

test(agents): cover continuation bootstrap reuse

ba58046

perf(agents): add continuation-skip context injection

3de19f4

docs(changelog): note context injection reuse

aaa4ae2

perf(agents): bound continuation bootstrap scan

632c8e4

fix(agents): require full bootstrap proof for continuation skip

551efcd

fix(agents): decide continuation skip under lock

6cfca07

fix(commands): re-export subagent chat message type

2be4270

fix(agents): clean continuation rebase leftovers

f1d4b10

vincentkoc force-pushed the issue-9157-context-injection branch from bb7dbf6 to f1d4b10 Compare April 6, 2026 04:27

vincentkoc merged commit 9ba97ce into main Apr 6, 2026
9 checks passed

vincentkoc deleted the issue-9157-context-injection branch April 6, 2026 04:27

chatgpt-codex-connector Bot reviewed Apr 6, 2026

View reviewed changes

github-actions Bot mentioned this pull request Apr 6, 2026

📡 Upstream Digest — 2026-04-06 04:38 UTC curtismercier/openclaw-mods#483

Open

This was referenced Apr 20, 2026

[Bug]: MSTeams DM inbound messages appear duplicated in agent session context #67323

Closed

Umbrella: duplicate transcript, replay, and context assembly across channels #69208

Open

This was referenced Apr 29, 2026

Session context bloat: bootstrap files re-injected every turn, wasting 20-30% tokens #67419

Open

Fix continuation-skip after assistant turns #74813

Draft

Feature: Selective/lazy injection for workspace context files #61347

Closed

clawsweeper Bot mentioned this pull request May 2, 2026

feat(agents): allow per-agent contextInjection override in agents.list[] #76054

Closed

This was referenced May 11, 2026

Feature Request: Smart Context Archival & Per-Channel Session Isolation #80526

Closed

Baseline Context Load: 40K tokens per message regardless of reply length #59427

Closed

Per-agent contextInjection override #76046

Closed

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

perf(agents): add continuation-skip context injection#61268

perf(agents): add continuation-skip context injection#61268
vincentkoc merged 8 commits into
mainfrom
issue-9157-context-injection

vincentkoc commented Apr 5, 2026

Uh oh!

greptile-apps Bot commented Apr 5, 2026

Uh oh!

Uh oh!

chatgpt-codex-connector Bot left a comment

Uh oh!

Uh oh!

chatgpt-codex-connector Bot left a comment

Uh oh!

Uh oh!

chatgpt-codex-connector Bot left a comment

Uh oh!

Uh oh!

Uh oh!

chatgpt-codex-connector Bot left a comment

Uh oh!

chatgpt-codex-connector Bot Apr 6, 2026

Uh oh!

ethangui commented Apr 6, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

Uh oh!

Conversation

vincentkoc commented Apr 5, 2026

Summary

Change Type (select all)

Scope (select all touched areas)

Linked Issue/PR

Root Cause (if applicable)

Regression Test Plan (if applicable)

User-visible / Behavior Changes

Diagram (if applicable)

Security Impact (required)

Repro + Verification

Environment

Steps

Expected

Actual

Evidence

Human Verification (required)

Review Conversations

Compatibility / Migration

Risks and Mitigations

Credits

Uh oh!

greptile-apps Bot commented Apr 5, 2026

Greptile Summary

Confidence Score: 5/5

Uh oh!

Uh oh!

chatgpt-codex-connector Bot left a comment

Choose a reason for hiding this comment

💡 Codex Review

Uh oh!

Uh oh!

chatgpt-codex-connector Bot left a comment

Choose a reason for hiding this comment

💡 Codex Review

Uh oh!

Uh oh!

chatgpt-codex-connector Bot left a comment

Choose a reason for hiding this comment

💡 Codex Review

Uh oh!

Uh oh!

Uh oh!

chatgpt-codex-connector Bot left a comment

Choose a reason for hiding this comment

💡 Codex Review

Uh oh!

chatgpt-codex-connector Bot Apr 6, 2026

Choose a reason for hiding this comment

Uh oh!

ethangui commented Apr 6, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants