fix(agents): isolated cron busts prompt prefix cache via per-run session id by openperf · Pull Request #96686 · openclaw/openclaw

openperf · 2026-06-25T06:42:24Z

Summary

Problem: isolated cron runs use a fresh run-scoped session key (agent:<agent>:cron:<job>:run:<runId>). buildRuntimeLine rendered that key and the matching session id into the prompt-facing Runtime: line. Transports that strip the explicit cache-boundary marker and rely on automatic literal-prefix caching therefore received different prompt bytes for every run, invalidating reuse before the tool catalog.
Root cause: fix: isolate cron context-engine session keys #72292 threaded the run-scoped key into embedded cron execution to prevent cross-run context leakage. fix #92453: add session identity to runtime prompt #92468 later exposed the execution key and id in Runtime:. Their interaction made prompt identity volatile. fix(cron): refuse keyless implicit isolated cron delivery inherited from shared agent-main bucket #91685 changed delivery-target resolution, but did not introduce or render the run-scoped key. The issue reporter's measured 2026.6.8/2026.6.9 boundary is useful runtime evidence, but that exact version boundary is not source-verifiable because v2026.6.8 already contains both source changes.
Fix: normalize only the model-visible identity. parseCronRunScopeSuffix recognizes the terminal canonical cron-run shape, buildRuntimeLine renders the stable job key, and a sessionId is omitted only when it duplicates the parsed run id. Execution, transcripts, routing, delivery, registry, and cleanup remain run-scoped.

Scope and behavior

Two different runs of one cron job now render byte-identical runtime lines.
A distinct rotated or stable sessionId is preserved.
Non-cron keys, malformed/nonterminal :run: segments, and channel-owned :run: data remain unchanged.
Structural markers remain case-insensitive while the stored key's original casing is preserved.
The parser is intentionally terminal-only; descendant session keys are outside this fix's scope.
Config, plugin SDK, and gateway protocol surfaces are unchanged.

The tradeoff is deliberate: the model sees stable cron-job identity instead of the concrete run key, while every operational identity path still uses the concrete run-scoped key.

Proof

Exact reviewed head: a32fa8912f9bffd23310f851ab41141d1a6e925f

Sanitized untrusted-source AWS Crabbox proof (cbx_b3d9e9da2799, run run_a702d09110a3, Linux/Node 24) ran:

pnpm test \
  src/agents/system-prompt.test.ts \
  src/routing/session-key.test.ts \
  src/cron/isolated-agent/run.session-key-isolation.test.ts \
  src/agents/embedded-agent-runner/prompt-cache-observability.test.ts \
  packages/ai/src/providers/openai-completions.test.ts \
  packages/ai/src/providers/openai-responses-shared.test.ts \
  src/agents/openai-transport-stream.test.ts

Result: 7 files across 4 Vitest shards, 579 tests passed. Fresh maintainer autoreview found no actionable findings. Exact-head hosted CI is required before merge; an Opengrep installer network-fetch failure was rerun successfully.

Not independently tested: live metered-provider token telemetry. The cache-stability mechanism is covered deterministically at the renderer, cron execution, observability, and affected transport boundaries.

Release note

Fix isolated cron prompt-cache stability by keeping per-run UUIDs out of the prompt-facing runtime identity.

Change type

Bug fix

Scope

Gateway / orchestration

Fixes #96677

clawsweeper · 2026-06-25T06:43:36Z

Codex review: needs maintainer review before merge. Reviewed June 25, 2026, 4:11 AM ET / 08:11 UTC.

Summary
The PR adds cron run-scope parsing, renders the stable cron base key in buildRuntimeLine, omits duplicated per-run sessionId, and adds focused parser/runtime-line tests.

PR surface: Source +38, Tests +74. Total +112 across 4 files.

Reproducibility: yes. at source level: current main constructs a :run: isolated-cron session key, passes it into embedded execution, and renders it directly in the Runtime prompt line. I did not independently reproduce the live provider token jump.

Review metrics: 1 noteworthy metric.

Runtime prompt identity surface: 1 normalized, 1 conditionally omitted. The diff changes model-visible session state in the Runtime line, so maintainers should notice the identity tradeoff before merge.

Stored data model
Persistent data-model change detected: persistent cache schema: src/agents/system-prompt.test.ts, serialized state: src/routing/session-key.test.ts, serialized state: src/sessions/session-key-utils.ts. Confirm migration or upgrade compatibility proof before merge.

Root-cause cluster
Relationship: fixed_by_candidate
Canonical: #96677
Summary: This PR is the candidate fix for the open isolated-cron Runtime-line prompt-cache regression; older prompt-cache and cron-isolation items are related context, not replacements.

Members:

canonical: [Bug]: 2026.6.9 re-busts prompt prefix caching — per-run :run:<UUID> injected into the system-prompt "Runtime:" line (regression of #43148) #96677 - The PR uses closing syntax for this issue and directly targets its reported isolated-cron Runtime-line cache instability.
partial_overlap: [Bug]: Changing system prompt causes cache invalidations #43148 - This earlier closed issue established the broader prompt-cache stability class, but it does not track this later isolated-cron run-id regression.
partial_overlap: fix: stabilize isolated cron prompt cache affinity #87309 - This merged PR added stable isolated-cron promptCacheKey affinity, but the current issue shows Runtime prompt text can still carry volatile run identity.
partial_overlap: fix(cron): refuse keyless implicit isolated cron delivery inherited from shared agent-main bucket #91685 - This merged PR changed adjacent isolated-cron delivery/session behavior cited as root-cause context, but it is not the remaining prompt-rendering fix.
partial_overlap: Agent runtime header lacks session identity, causing misleading self-references #92453 - This closed issue introduced the Runtime session identity surface whose prompt-cache interaction is being adjusted here.

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

Merge readiness
Overall: 🦞 diamond lobster
Proof: 🌊 off-meta tidepool
Patch quality: 🦞 diamond lobster
Result: ready for maintainer review.

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

Risk before merge

[P1] The PR intentionally changes model-visible Runtime session identity for isolated cron runs: session= becomes the stable cron base key and duplicated per-run sessionId= is omitted while routing, transcript, cleanup, and delivery state remain run-scoped.
[P1] I did not independently run a live metered-provider token trace; source inspection and the PR tests prove byte-stable Runtime rendering at the fix boundary, while the exact billing delta remains reporter-provided.

Maintainer options:

Accept Prompt-Facing Stable Cron Identity (recommended)
Merge with the explicit maintainer decision that only Runtime prompt identity is normalized while actual isolated-cron run state remains per-run.
Request A Redacted Live Cache Trace
Ask for before/after isolated-cron provider output showing the Runtime line and cache or token counts if billing proof is required before merge.
Pause For Broader Prompt-Volatility Design
Defer this PR only if maintainers prefer a larger cache-boundary redesign, with the current linked regression remaining open meanwhile.

Next step before merge

[P2] No repair lane is needed; maintainers should review the model-visible session identity tradeoff and either land the PR or request live cache/token proof.

Security
Cleared: The diff only changes TypeScript prompt-rendering/session-key helper logic and focused tests; no dependency, workflow, package, secret, or supply-chain surface changed.

Review details

Best possible solution:

Land this prompt-rendering fix after maintainers accept the prompt-facing cron identity tradeoff, while keeping actual isolated-cron routing, transcript, cleanup, and delivery state per-run.

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

Yes at source level: current main constructs a :run: isolated-cron session key, passes it into embedded execution, and renders it directly in the Runtime prompt line. I did not independently reproduce the live provider token jump.

Is this the best way to solve the issue?

Yes. Prompt-facing normalization in buildRuntimeLine is narrower than reusing one cron session across runs, and the existing promptCacheKey path alone does not remove volatile Runtime prompt bytes.

AGENTS.md: found and applied where relevant.

Codex review notes: model internal, reasoning high; reviewed against 770b19f496c1.

Label changes

Label justifications:

P2: This is a focused isolated-cron prompt-cache cost regression with bounded blast radius and no outage, data loss, or security bypass.
merge-risk: 🚨 session-state: The PR changes model-visible Runtime session identity for cron-scoped agent runs, even though underlying routing and transcript state remain run-scoped.
rating: 🦞 diamond lobster: Overall readiness is 🦞 diamond lobster; proof is 🌊 off-meta tidepool and patch quality is 🦞 diamond lobster.
status: 👀 ready for maintainer look: ClawSweeper has no concrete contributor-facing blocker left for this PR. Not applicable: The external-contributor real-behavior proof gate does not apply to this MEMBER-authored PR; the body reports deterministic render-boundary tests and states no live metered-provider trace was run.

Evidence reviewed

PR surface:

Source +38, Tests +74. Total +112 across 4 files.

View PR surface stats

Area	Files	Added	Removed	Net
Source	2	40	2	+38
Tests	2	74	0	+74
Docs	0	0	0	0
Config	0	0	0	0
Generated	0	0	0	0
Other	0	0	0	0
Total	4	114	2	+112

What I checked:

Repository policy read: Root AGENTS.md was read fully; its review guidance applies because this PR touches agent prompt/session state and requires whole-path session-state review. (AGENTS.md:1, 770b19f496c1)
Scoped policy read: The scoped agents guide was read fully; it did not add a runtime blocker beyond keeping agent tests focused and import-light. (src/agents/AGENTS.md:1, 770b19f496c1)
Current main renders volatile session identity: Current main directly renders runtimeInfo.sessionKey and runtimeInfo.sessionId into the Runtime prompt line, so a run-scoped cron key is model-visible before this PR. (src/agents/system-prompt.ts:1400, 770b19f496c1)
Current main creates run-scoped cron keys: Current main appends :run:${runSessionId} to agent-scoped cron keys for isolated cron execution, matching the reported volatile Runtime field. (src/cron/isolated-agent/run.ts:634, 770b19f496c1)
Prompt cache key is separate from prompt text: The executor already passes a stable promptCacheKey and the run-scoped sessionKey separately, so the prior cache-affinity fix does not remove the volatile Runtime prompt text by itself. (src/cron/isolated-agent/run-executor.ts:361, 770b19f496c1)
Cache boundary checked: The cache boundary is inserted before dynamic sections, while Runtime is appended later; boundary-aware transports can split it, but transports that strip the marker still see Runtime as part of the literal prompt string. (src/agents/system-prompt.ts:1277, 770b19f496c1)

Likely related people:

zhangguiping-xydt: Authored the merged Runtime identity PR that added session= and sessionId= to prompt rendering. (role: introduced Runtime session identity; confidence: high; commits: 78c66742ab97; files: src/agents/system-prompt.ts, src/agents/system-prompt-params.ts, src/agents/embedded-agent-runner/run/attempt.ts)
vincentkoc: History search shows multiple prompt-cache boundary and transport commits that this fix must preserve. (role: prompt-cache area contributor; confidence: high; commits: 64f28906de09, 4798e125f40c; files: src/agents/system-prompt.ts, src/agents/system-prompt-cache-boundary.ts, src/agents/openai-transport-stream.ts)
steipete: Prior review and history point to the isolated-cron fresh-session change that made the cron identity run-scoped. (role: introduced isolated-cron freshness behavior; confidence: high; commits: 91f75a2b33b6; files: src/cron/isolated-agent/run.ts)
ferminquant: Merged prior isolated-cron prompt-cache affinity work that this PR complements rather than replaces. (role: previous prompt-cache affinity contributor; confidence: high; commits: 3f9d2415acc8; files: src/cron/isolated-agent/run-executor.ts, src/cron/isolated-agent/run.session-key-isolation.test.ts, src/agents/openai-transport-stream.ts)
nxmxbbd: Authored the merged adjacent isolated-cron delivery/session PR cited as root-cause context for the run-scoped key. (role: recent adjacent cron delivery contributor; confidence: medium; commits: d4833e27c775; files: src/cron/isolated-agent/run.ts, src/cron/isolated-agent/delivery-target.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.

…ion id Isolated cron runs carry a per-run :run:<id> session scope (openclaw#91685) rendered verbatim into the cached system-prompt Runtime line, re-busting byte-exact prefix caching for the tool catalog after it every run (openclaw#96677, openclaw#43148 class). buildRuntimeLine now renders the stable base session key and drops the per-run id the run scope duplicates; parseCronRunScopeSuffix is gated to the isolated-cron key shape so a :run: segment in any other session key is never truncated.

steipete · 2026-07-09T15:50:35Z

Maintainer landing proof for a32fa8912f9bffd23310f851ab41141d1a6e925f:

Rebased onto 2e160718d4b77716ec7675e659619dff893b3fb7; the final merge guard compared current main and found only unrelated drift.
Deep review confirmed prompt-only normalization: operational execution, transcript, routing, delivery, registry, and cleanup identity remain run-scoped.
Review caught and fixed mixed-case canonical marker parsing, added a regression for it, and added explicit coverage that a distinct cron sessionId remains visible.
Fresh autoreview: no actionable findings (0.94 confidence).
Sanitized AWS Crabbox proof: provider aws, lease cbx_b3d9e9da2799, run run_a702d09110a3; 7 focused files across 4 Vitest shards, 579 tests passed.
Exact-head hosted CI is green. The only initial failure was Opengrep's installer failing to fetch its release list; the exact failed job was rerun and passed.
Repository prepare and merge-verification gates passed. Root changelog is release-owned for this file set; release-note context is in the PR body.

Residual evidence limit: no independent live metered-provider token trace; cache stability is proven deterministically at the renderer, cron execution, observability, and affected transport boundaries.

steipete · 2026-07-09T15:51:10Z

Merged via squash.

Prepared head SHA: a32fa8912f9bffd23310f851ab41141d1a6e925f
Landed commit: f6c16d22b478f80b37bc27efc61a95629cc088d7

…ion id (openclaw#96686) * fix(agents): isolated cron busts prompt prefix cache via per-run session id Isolated cron runs carry a per-run :run:<id> session scope (openclaw#91685) rendered verbatim into the cached system-prompt Runtime line, re-busting byte-exact prefix caching for the tool catalog after it every run (openclaw#96677, openclaw#43148 class). buildRuntimeLine now renders the stable base session key and drops the per-run id the run scope duplicates; parseCronRunScopeSuffix is gated to the isolated-cron key shape so a :run: segment in any other session key is never truncated. * fix(agents): preserve mixed-case cron run markers * test(agents): cover rotated cron session identity --------- Co-authored-by: Peter Steinberger <[email protected]>

openclaw-barnacle Bot added agents Agent runtime and tooling size: S labels Jun 25, 2026

openperf force-pushed the fix/96677-cron-prompt-cache-runscope branch from cea99f9 to c8d516b Compare June 25, 2026 07:36

clawsweeper Bot mentioned this pull request Jun 28, 2026

fix(system-prompt): strip volatile :run: suffix from Runtime line (#96677) #97449

Closed

steipete self-assigned this Jul 9, 2026

openperf and others added 3 commits July 9, 2026 08:38

fix(agents): preserve mixed-case cron run markers

cc422ea

test(agents): cover rotated cron session identity

a32fa89

steipete force-pushed the fix/96677-cron-prompt-cache-runscope branch from c8d516b to a32fa89 Compare July 9, 2026 15:40

steipete merged commit f6c16d2 into openclaw:main Jul 9, 2026
103 of 104 checks passed

github-actions Bot mentioned this pull request Jul 9, 2026

📡 Upstream Digest — 2026-07-09 17:48 UTC curtismercier/openclaw-mods#1244

Open

clawsweeper Bot mentioned this pull request Jul 9, 2026

fix(models): preserve static provider catalog metadata #103051

Closed

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

Uh oh!

fix(agents): isolated cron busts prompt prefix cache via per-run session id#96686

fix(agents): isolated cron busts prompt prefix cache via per-run session id#96686
steipete merged 3 commits into
openclaw:mainfrom
openperf:fix/96677-cron-prompt-cache-runscope

openperf commented Jun 25, 2026 •

edited by steipete

Loading

Uh oh!

clawsweeper Bot commented Jun 25, 2026 •

edited

Loading

Uh oh!

steipete commented Jul 9, 2026

Uh oh!

Uh oh!

steipete commented Jul 9, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

Uh oh!

Uh oh!

Conversation

openperf commented Jun 25, 2026 • edited by steipete Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Summary

Scope and behavior

Proof

Release note

Change type

Scope

Uh oh!

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

Uh oh!

Uh oh!

steipete commented Jul 9, 2026

Uh oh!

Uh oh!

steipete commented Jul 9, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

openperf commented Jun 25, 2026 •

edited by steipete

Loading

clawsweeper Bot commented Jun 25, 2026 •

edited

Loading