openclaw - ✅(Solved) Fix Isolated session outputs forwarded without main agent awareness [1 pull requests, 8 comments, 4 participants]

sabo961 · 2026-03-22T07:33:03Z

[openclaw] PR 52170: fix cron : write delivered cron output back into main session transcript - Repository: openclaw/openclaw - Author: sparkyrider - State: cl… # PR #52170: fix(cron): write delivered cron output back into main session transcript - Repository: openclaw/openclaw - Author: sparkyrider - State: closed | merged: False - Link: https://github.com/openclaw/openclaw/pull/52170 ## Description (problem / solution / changelog) ## Summary ### **tldr:** - Cron job outputs delivered to channels were never saved in the agent's session history, so the agent couldn't recall what it sent. This fix writes delivered cron output back into the correct transcript with proper routing and deduplication. - **Problem:** When an isolated cron job delivers a message to a channel (e.g. Telegram/Discord), the main agent has no record of it. Users can ask: - "what did you send this morning?" and the agent has no idea — it forwarded the content but never wrote it to its own transcript. - **Why it matters:** The agent acts as a blind pipe — receive, forward, forget. Users expect the agent to recall its own deliveries, especially when asking follow-up questions about content it just sent. - **What changed:** - **Bug fix:** After a successful direct isolated cron announce delivery, OpenClaw now makes a best-effort `chat.inject` write into the routed main or thread transcript using a stable idempotency key, so the agent can recall what it already sent. New session entries are seeded with delivery context from the main session to avoid creating routing-empty store entries. The `chat.inject` handler now returns success for idempotent dedup replays instead of a spurious error. - **Refactor:** Session key routing was unified so awareness injection, delivery target resolution, and store writes all use the runtime session key consistently instead of pulling from different sources (`job.sessionKey` vs the runtime param): - `resolveCronDeliveryContext` (`run.ts`) — changed from `params.job.sessionKey` to `params.sessionKey` so delivery target resolution uses the correct session context for hook dispatches and shared-contract runs. - `dispatchCronDelivery` params (`delivery-dispatch.ts`) — added a required `sessionKey` field to `DispatchCronDeliveryParams` so the runtime session key flows through the delivery dispatch path. - `resolveDeliveryTarget` (`delivery-target.ts`) — changed to resolve the session store from the session key's agent ID instead of always using the runner's `agentId`, and added `canonicalizeMainSessionAlias` so thread/routed keys are properly canonicalized before store lookup. - **Scope boundary:** This does not resend cron output to users, does not switch cron over to the subagent announce flow, and does not change the existing no-fallback-main-summary behavior. ## Change Type (select all) - [x] Bug fix - [ ] Feature - [x] Refactor required for the fix - [ ] Docs - [ ] Security hardening - [ ] Chore/infra ## Scope (select all touched areas) - [x] Gateway / orchestration - [ ] Skills / tool execution - [ ] Auth / tokens - [x] Memory / storage - [ ] Integrations - [x] API / contracts - [ ] UI / DX - [ ] CI/CD / infra ## Linked Issue/PR - Fixes #52136 ## User-visible / Behavior Changes Successful direct isolated cron announce deliveries now attempt to append a labeled `Cron delivery` assistant entry into the routed main or thread transcript so later follow-up turns can recall what cron already sent externally. ## 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` ## Repro + Verification ### Environment - OS: macOS - Runtime/container: local repo checkout on `main` plus the fix branch - Model/provider: N/A - Integration/channel: isolated cron direct announce delivery with routed main or thread transcript awareness - Relevant config: test harness session store with `agent:main:main` and optional thread session entries ### Steps 1. Configure an isolated cron `agentTurn` job with `delivery.mode="announce"`. 2. Let the isolated run produce a normal text response and deliver it through the direct outbound path. 3. Inspect the routed main or thread transcript after delivery. ### Expected - After a successful direct delivery, OpenClaw writes a `Cron delivery` transcript entry into the routed session. - Replayed `chat.inject` writes for the same run/session combination are deduped by idempotency key. ### Actual Before This Change - Direct delivery completed, but the routed transcript did not receive the delivered content. ## Evidence - [x] Failing test/log before plus passing after - [ ] Trace/log snippets - [ ] Screenshot/recording - [ ] Perf numbers Regression coverage added for: - main-session awareness injection - thread-session awareness injection - idempotent transcript injection behavior through `chat.inject` - runtime session key routing for hook dispat

openclaw2026-03-22 07:33:03

ON THIS PAGE

Recommended Tools

×6

Utilities matched from this issue’s tags and category — try them while you read without losing context.

GitHub issue graph ai analysis

Paste a GitHub issue URL. We fetch that issue, discover linked issues from bodies/comments/timeline, collect linked pull requests, and produce a structured English report.

The report is written in English Markdown for sharing and archival.

GitHub issue URL

Helpful · Quick feedback

GitHub stats

openclaw/openclaw#52136•Fetched 2026-04-08 01:15:12

View on GitHub

Comments

Participants

Timeline

Reactions

Author

Participants

Timeline (top)

commented ×8mentioned ×5subscribed ×5cross-referenced ×3

Error Message

Agent acts as a pipe: receives → forwards → forgets
No internalization of content it delivers
Agent cannot answer questions about information it just sent
User has to point agent to its own output

RAW_BUFFERClick to expand / collapse

Problem

When an isolated cron job (e.g. bmad-daily) generates output that gets delivered to a channel (Telegram/Discord), the main agent forwards it without ever reading or internalizing the content.

Example

bmad-daily cron runs at 08:00, scrapes BMad changelog, writes to knowledge/bmad/README.md
Output is delivered to Telegram as announcement
User asks main agent: "What's new in BMad v4.30.1?"
Main agent has no idea — it forwarded that exact info hours ago but never read it

Observed behavior

Agent acts as a pipe: receives → forwards → forgets
No internalization of content it delivers
Agent cannot answer questions about information it just sent
User has to point agent to its own output

Expected behavior

Some awareness of content passing through — at minimum, main session should be able to recall "I sent something about BMad v4.30.1 this morning" even if it doesn't have full context.

Possible causes (speculation)

Isolated sessions have no memory bridge to main session
Delivery happens at transport layer, bypasses agent cognition
Main agent context doesn't include recent deliveries from its cron jobs

Environment

OpenClaw 2026.3.13
Cron with sessionTarget: isolated + delivery.mode: announce

extent analysis

Fix Plan

To address the issue, we need to implement a mechanism for the main agent to internalize the content it delivers. Here are the steps:

Modify the cron job to store the output in a shared knowledge base or database.
Update the main agent to periodically scan the knowledge base for new information.
Implement a caching mechanism to store recently delivered content.

Example Code

# cron job (bmad-daily) writes to knowledge base
import json

def scrape_bmad_changelog():
    # scrape BMad changelog
    data = {'version': '4.30.1', 'changes': ['change1', 'change2']}
    with open('knowledge/bmad/README.md', 'w') as f:
        json.dump(data, f)

# main agent periodically scans knowledge base
import os
import json

def update_knowledge_base():
    knowledge_base = 'knowledge/bmad/README.md'
    if os.path.exists(knowledge_base):
        with open(knowledge_base, 'r') as f:
            data = json.load(f)
            # store data in main agent's cache
            cache['bmad'] = data

# main agent recalls recently delivered content
def recall_recent_deliveries():
    if 'bmad' in cache:
        return cache['bmad']
    return None

Verification

To verify that the fix worked, test the following scenarios:

Run the bmad-daily cron job and verify that the main agent can recall the delivered content.
Ask the main agent questions about the delivered content and verify that it can provide accurate answers.

Extra Tips

Consider implementing a more robust caching mechanism, such as a Redis or Memcached store, to handle larger volumes of data.
Ensure that the knowledge base and cache are properly synchronized to avoid data inconsistencies.

Vote matrix · Quick signals

Works

Did the solution work? Tap to confirm.

Easy Fix

Was it a quick fix?

Time Saver

Did it save you time?

Blocking

Was it severely blocking?

Common Issue

Are others likely hitting this too?

Flaky / Intermittent

Is it intermittent?

Verified / Reproducible

Can you reproduce it reliably?

FAQ

Expected behavior

Some awareness of content passing through — at minimum, main session should be able to recall "I sent something about BMad v4.30.1 this morning" even if it doesn't have full context.

#SSR setup #ISR setup #authentication setup #request error #file not found

Still need to ship something?

×6

Another batch ranked right after the header list — different links, same matching logic.

openclaw - ✅(Solved) Fix Isolated session outputs forwarded without main agent awareness [1 pull requests, 8 comments, 4 participants]

Recommended Tools

GitHub issue graph ai analysis

Error Message

Fix Action

Fixed

PR fix notes

PR #52170: fix(cron): write delivered cron output back into main session transcript

Description (problem / solution / changelog)

Summary

tldr: - Cron job outputs delivered to channels were never saved in the agent's session history, so the agent couldn't recall what it sent. This fix writes delivered cron output back into the correct transcript with proper routing and deduplication.

Change Type (select all)

Scope (select all touched areas)

Linked Issue/PR

User-visible / Behavior Changes

Security Impact (required)

Repro + Verification

Environment

Steps

Expected

Actual Before This Change

Evidence

Human Verification (required)

Review Conversations

Compatibility / Migration

Failure Recovery (if this breaks)

Risks and Mitigations

Changed files

Problem

Example

Observed behavior

Expected behavior

Possible causes (speculation)

Environment

extent analysis

Fix Plan

Example Code

Verification

Extra Tips

FAQ

Expected behavior

Still need to ship something?

RELATED_DISCOVERY

TRENDING