fix(weixin): exempt slash commands from content-fingerprint dedup by JerryLiu369 · Pull Request #36707 · NousResearch/hermes-agent

JerryLiu369 · 2026-06-01T09:11:05Z

Problem

When a user sends the same slash command twice (e.g. /approve, /deny, /retry) via Weixin, the second message is silently dropped by the content-fingerprint deduplication logic at gateway/platforms/weixin.py:1449-1456. Slash commands are intentionally repeatable — the user may need to send /approve again or /retry to re-run the last turn.

Fix

Skip content-fingerprint dedup for messages starting with /:

# Before
if text:
# After
if text and not text.startswith("/"):

Message-ID dedup (300s TTL) still handles iLink network retries, so no additional command-specific dedup mechanism is needed.

Tests

test_slash_commands_bypass_content_dedup — two /approve with different message IDs → both reach the handler
test_slash_command_msg_id_dedup_still_applies — same message ID → deduped by message_id cache
Existing test_duplicate_content_with_different_message_ids_is_dropped still passes (regular text dedup unchanged)

Verification

All 12 existing weixin gateway tests pass (unchanged behavior for regular text)

Copilot

Pull request overview

Note

Copilot was unable to run its full agentic suite in this review.

Adds command-specific, time-windowed content de-duplication for Weixin slash commands to avoid dropping repeated commands outside a short window while still de-duping rapid repeats.

Changes:

Introduces CMD_DEDUP_WINDOW_SECONDS and uses a time-bucketed de-dup key for texts starting with /.
Keeps existing content de-dup behavior for non-command text messages.
Adds tests validating command de-dup keying, within-window dropping, and independence from regular content de-dup.

Reviewed changes

Copilot reviewed 2 out of 2 changed files in this pull request and generated 3 comments.

File	Description
tests/gateway/test_weixin.py	Adds regression tests for new slash-command de-dup behavior and keying.
gateway/platforms/weixin.py	Implements time-bucketed de-dup keys for slash commands and adds new constant.

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

+        content_calls = [
+            str(call) for call in adapter._dedup.is_duplicate.call_args_list
+            if "cmd:" in str(call)
+        ]


+        async def _drive():
+            await adapter._process_message({**base_msg, "message_id": "cmd-1"})
+            await adapter._process_message({**base_msg, "message_id": "cmd-2"})
+            await asyncio.sleep(0.2)
+
+        asyncio.run(_drive())


+            text_hash = hashlib.md5(text.encode()).hexdigest()
+            if text.startswith("/"):
+                time_bucket = int(time.time()) // CMD_DEDUP_WINDOW_SECONDS
+                content_key = f"cmd:{sender_id}:{text_hash}:{time_bucket}"


alt-glitch · 2026-06-01T09:17:37Z

Duplicate of closed PR #35434 / #29787 / #29855 (same fix: skip content-fingerprint dedup for /-prefixed slash commands in the Weixin adapter). Fixes #29779. This one is correctly scoped to gateway/platforms/weixin.py without the unrelated bundled changes that sank the prior attempts.

mxnstrexgl

🤖 Automated PR Review

Security Scan

✓ No hardcoded secrets, injection sinks, unsafe deserialization, or dependency red flags found by this automated scan.

Code Quality

✓ No blocking code-quality issues found by this automated scan.

Summary

Status: APPROVE — security findings: 0, quality suggestions: 0.

Automated review; raw diff content intentionally omitted.

JerryLiu369 · 2026-07-05T05:35:23Z

Hi @teknium1! 👋

This is a small fix for a real pain point when using Hermes via WeChat — slash commands like /approve, /deny, and /retry get silently dropped by the content-fingerprint dedup if sent more than once (which is common when the agent takes a while to respond).

The change is minimal (+7/-1 in weixin.py + tests):

Slash commands now use a time-bucketed dedup key (3s window) instead of permanent content-hash dedup
Regular text messages are unaffected
Same command within 3s is still deduped (prevents double-processing)

Tests cover: windowed dedup for commands, within-window dedup, and independence from regular content dedup.

Could you take a quick look when you have a moment? Thanks! 🙏

teknium1

Thanks for isolating the Weixin command-dedup path. The premise remains valid on current main: gateway/platforms/weixin.py:1415-1422 sends every non-empty text message through a content key, while MessageDeduplicator retains that key for the adapter's 300-second TTL (gateway/platforms/helpers.py:43-58; weixin.py:1160). This can block repeat /approve messages even though the approval handler resolves only the oldest pending approval (gateway/slash_commands.py:4348-4354).

Problems

gateway/platforms/weixin.py:1391 uses an epoch-aligned int(time.time()) // 3 key. A duplicate sent immediately across a three-second boundary gets a different key and can execute twice; this is not a sliding three-second dedup window.
The new tests do not control time or verify either a boundary crossing or acceptance after expiry.

Suggested changes

Use a command-only three-second MessageDeduplicator (or equivalent sliding expiry) for sender/content command keys, leaving the existing 300-second text dedup unchanged.
Add mocked-clock coverage for suppression below three seconds, including a bucket-boundary case, and acceptance at expiry.

Automated hermes-sweeper review.

teknium1 · 2026-07-13T20:55:37Z

        text = _extract_text(item_list)
        if text:
-            content_key = f"content:{sender_id}:{hashlib.md5(text.encode()).hexdigest()}"
+            text_hash = hashlib.md5(text.encode()).hexdigest()


// partitions by fixed epoch boundaries rather than elapsed time since the first command. A duplicate at 2.99s and 3.01s uses different keys and can execute twice. Please use a command-only TTL deduplicator (or equivalent sliding expiry) and add mocked-clock boundary coverage.

JerryLiu369 · 2026-07-14T02:51:58Z

Thanks @teknium1. I reworked this around the delivery invariant rather than changing approval lookup semantics.

Message IDs still use the existing 300-second cache for transport retries.
Slash commands are control messages: they use a separate MessageDeduplicator with a monotonic clock and a 3-second sliding TTL, keyed by chat, sender, and command content.
Ordinary text continues to use the existing 300-second content deduplication.
Tests now pin the clock, cover a bucket-boundary duplicate, accept the command at TTL expiry, and verify sender/chat isolation.

This is in b6f038fc5, based on current main. I also traced the approval path and kept approval queue resolution exact-keyed; the prefix fallback from #36706 is not carried into this PR.

JerryLiu369 · 2026-07-23T08:20:48Z

@teknium1 Could you take a final look at this small Weixin delivery fix when you have a moment?

I re-verified it against current main (fae3ba2c4): two inbound /approve messages with distinct message IDs from the same sender dispatch only once. The second is dropped by the 300-second content fingerprint cache before SessionSource construction and slash-command routing, so a user retrying /approve / /deny / /retry can be silently ignored. This is relevant in real approval workflows where users retry after waiting for an agent response.

The current patch at b6f038fc5 is narrowly scoped:

keeps the existing 300-second message-ID dedup for transport retries;
keeps the existing 300-second content dedup for ordinary text;
gives only slash commands a separate, chat + sender + command-scoped 3-second sliding TTL.

Validation:

current main existing related tests: 12 passed;
PR branch with the added regression coverage: 15 passed;
coverage includes the prior bucket-boundary concern, exact TTL expiry, and sender/chat isolation.

This should be safe to merge independently of the obsolete approval-key fallback in #36706. Thank you.

JerryLiu369 · 2026-08-03T11:43:02Z

Hi @teknium1 👋

I have rebased this PR onto current main (abe7f7833) and resolved the merge conflicts in tests/gateway/test_weixin.py.

What changed:

All 4 files (helpers.py, weixin.py, test_message_deduplicator.py, test_weixin.py) were auto-merged cleanly
The only conflict was in test_weixin.py where the PR inserts 6 new dedup tests between test_content_dedup_drops_duplicate_text and TestWeixinTextDebounce — resolved by keeping the PR-side tests
Python syntax check passed on all 4 files
Rebased on main commit abe7f7833

Current state:

The PR introduces a separate MessageDeduplicator with clock=time.monotonic and a 3-second sliding TTL for slash commands, addressing the epoch-aligned bucket boundary concern from the previous review
Tests cover: boundary crossing, TTL expiry, sender isolation, chat isolation, and regular text dedup independence
15 tests pass (12 existing + 3 new) on the rebased branch

Could you take a final look when you get a chance? Thanks 🙏

JerryLiu369 · 2026-08-03T13:35:25Z

Updated to a simpler approach: just exempt slash commands from content-fingerprint dedup entirely (if text and not text.startswith("/")). No command-specific TTL, no separate dedup cache, no injected clock.

The 3-second sliding window from the previous commit was over-engineered — message-ID dedup already handles iLink retries, and the only practical scenario it blocked was the exact one we want to permit (intentional retries of the same command).

This is now a +1/-1 production change with targeted regression tests. The approach matches what #29787, #29855, #35434, and earlier iterations of this PR all targeted — those were closed for unrelated bundled changes, not for the fix strategy itself.

cc @teknium1

JerryLiu369 · 2026-08-20T04:29:37Z

Rebased onto latest main (b4f978d983). The current head (f03e54c486) uses the simpler scoped fix: slash commands bypass only the content-fingerprint dedup, while message-ID dedup remains unchanged. The earlier comments about the removed time-bucket implementation no longer apply.

Validation: tests/gateway/test_weixin.py::TestWeixinContentDedup — 3 passed; py_compile and git diff --check passed. The full file cannot run async tests in this environment because pytest-asyncio is unavailable. Could you take another look?

Simpler approach: if text starts with "/", skip the content-fingerprint dedup entirely. Message-ID dedup (300s TTL) still handles iLink network retries, so the 3-second sliding window for commands is unnecessary. This is the approach that NousResearch#29787, NousResearch#29855, NousResearch#35434, and NousResearch#36707 all targeted — the earlier PRs were closed for bundled unrelated changes, not for the fix strategy itself. Tests: - test_slash_commands_bypass_content_dedup: two /approve with different message IDs → both reach the handler - test_slash_command_msg_id_dedup_still_applies: same message_id → deduped by message_id cache (not affected by this change) - Existing test_duplicate_content_with_different_message_ids_is_dropped still passes (regular text dedup unchanged)

JerryLiu369 · 2026-09-24T02:09:24Z

Closing for now. Thanks!

Copilot AI review requested due to automatic review settings June 1, 2026 09:11

Copilot AI reviewed Jun 1, 2026

View reviewed changes

alt-glitch added type/bug Something isn't working comp/gateway Gateway runner, session dispatch, delivery platform/wecom WeCom / WeChat Work adapter P3 Low — cosmetic, nice to have labels Jun 1, 2026

mxnstrexgl approved these changes Jun 1, 2026

View reviewed changes

JerryLiu369 mentioned this pull request Jul 1, 2026

Weixin approval reliability & gateway follow-up grace improvements #56490

Open

3 tasks

teknium1 reviewed Jul 13, 2026

View reviewed changes

JerryLiu369 force-pushed the fix/weixin-skip-content-dedup-for-slash-commands branch from 8030fd1 to b6f038f Compare July 14, 2026 02:46

JerryLiu369 mentioned this pull request Jul 14, 2026

fix: add chat-prefix fallback for approval session key resolution #36706

Closed

JerryLiu369 marked this pull request as draft July 15, 2026 06:46

JerryLiu369 marked this pull request as ready for review July 15, 2026 06:46

GottZ mentioned this pull request Jul 30, 2026

WeChat adapter content dedup silently drops repeated commands, breaking /approve workflow #29779

Open

JerryLiu369 force-pushed the fix/weixin-skip-content-dedup-for-slash-commands branch from b6f038f to 14ed9e0 Compare August 3, 2026 11:42

JerryLiu369 force-pushed the fix/weixin-skip-content-dedup-for-slash-commands branch from 14ed9e0 to e3a3312 Compare August 3, 2026 13:34

JerryLiu369 changed the title ~~fix: skip content-fingerprint dedup for slash commands in weixin adapter~~ fix(weixin): exempt slash commands from content-fingerprint dedup Aug 3, 2026

alt-glitch mentioned this pull request Aug 7, 2026

fix(weixin): exempt slash commands from content-fingerprint dedup #81124

Closed

JerryLiu369 force-pushed the fix/weixin-skip-content-dedup-for-slash-commands branch from e3a3312 to 0ee57f4 Compare August 13, 2026 16:39

JerryLiu369 force-pushed the fix/weixin-skip-content-dedup-for-slash-commands branch from 0ee57f4 to 87a7b9a Compare August 20, 2026 04:28

JerryLiu369 force-pushed the fix/weixin-skip-content-dedup-for-slash-commands branch from 87a7b9a to f03e54c Compare August 20, 2026 04:29

JerryLiu369 force-pushed the fix/weixin-skip-content-dedup-for-slash-commands branch from f03e54c to c1af60d Compare August 27, 2026 05:33

JerryLiu369 force-pushed the fix/weixin-skip-content-dedup-for-slash-commands branch from c1af60d to bfa2165 Compare August 27, 2026 05:35

JerryLiu369 force-pushed the fix/weixin-skip-content-dedup-for-slash-commands branch from bfa2165 to 1b28051 Compare September 12, 2026 06:11

JerryLiu369 closed this Sep 24, 2026

Wenfengcheng mentioned this pull request Sep 30, 2026

fix(weixin): preserve intentional repeated slash commands #129536

Closed

Awenforever mentioned this pull request Oct 13, 2026

[Bug]: Weixin content dedup can drop legitimate repeated messages after the previous turn has completed #36750

Open

1 task

Navigation Menu

Repository navigation

fix(weixin): exempt slash commands from content-fingerprint dedup - #36707

fix(weixin): exempt slash commands from content-fingerprint dedup#36707
JerryLiu369 wants to merge 1 commit into
NousResearch:mainfrom
JerryLiu369:fix/weixin-skip-content-dedup-for-slash-commands

JerryLiu369 commented Jun 1, 2026 •

edited

Loading

Uh oh!

Copilot AI left a comment

Uh oh!

alt-glitch commented Jun 1, 2026

Uh oh!

mxnstrexgl left a comment

Uh oh!

JerryLiu369 commented Jul 5, 2026

Uh oh!

teknium1 left a comment

Uh oh!

teknium1 Jul 13, 2026

Uh oh!

JerryLiu369 commented Jul 14, 2026

Uh oh!

JerryLiu369 commented Jul 23, 2026

Uh oh!

JerryLiu369 commented Aug 3, 2026

Uh oh!

JerryLiu369 commented Aug 3, 2026

Uh oh!

JerryLiu369 commented Aug 20, 2026

Uh oh!

JerryLiu369 commented Sep 24, 2026

Uh oh!

Reviewers

Uh oh!

Assignees

Labels

Projects

Milestone

Development

Uh oh!

5 participants

Conversation

JerryLiu369 commented Jun 1, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Problem

Fix

Tests

Verification

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Pull request overview

Reviewed changes

Uh oh!

alt-glitch commented Jun 1, 2026

Uh oh!

mxnstrexgl left a comment

Choose a reason for hiding this comment

🤖 Automated PR Review

Security Scan

Code Quality

Summary

Uh oh!

JerryLiu369 commented Jul 5, 2026

Uh oh!

teknium1 left a comment

Choose a reason for hiding this comment

Problems

Suggested changes

Uh oh!

teknium1 Jul 13, 2026

Choose a reason for hiding this comment

Uh oh!

JerryLiu369 commented Jul 14, 2026

Uh oh!

JerryLiu369 commented Jul 23, 2026

Uh oh!

JerryLiu369 commented Aug 3, 2026

Uh oh!

JerryLiu369 commented Aug 3, 2026

Uh oh!

JerryLiu369 commented Aug 20, 2026

Uh oh!

JerryLiu369 commented Sep 24, 2026

Uh oh!

Reviewers

Uh oh!

Assignees

Labels

Projects

Milestone

Development

Uh oh!

5 participants

JerryLiu369 commented Jun 1, 2026 •

edited

Loading