fix(discord): reconcile ambiguous auto-thread creation

[izumi0uu]

What does this PR do?

Discord auto-thread creation is a remote side effect: message.create_thread() can commit on Discord and still raise locally if the response is lost. The existing exception path treats that as a definitive failure and immediately creates a second thread from a fallback seed message.

This PR reconciles the original starter-message thread before allowing the fallback. It reuses a matching thread from message.thread, the client cache, or Discord's channel fetch; retries the lookup briefly for delayed visibility; and skips the fallback when reconciliation itself is inconclusive. A fallback seed is still created when Discord confirms that the original thread is absent.

Related Issue

Fixes #73032

Related implementation: #73092 checks the in-memory message.thread path. This PR also covers cache misses, REST lookup, delayed visibility, mismatched cached channels, and lookup failures.

This preserves the original author commit from the closed #48685 and adapts it to current main.

Type of Change

• Bug fix (non-breaking change that fixes an issue)
• Tests (adding or improving test coverage)

Changes Made

• Validate that a reconciled object is a Discord thread whose ID matches the starter message ID.
• Resolve an already-created thread through the message object, client cache, and Discord REST fetch.
• Retry reconciliation briefly before creating a fallback seed.
• Only use the fallback after absence is confirmed; an inconclusive lookup returns failure rather than risking a duplicate remote side effect.
• Cover message, cache, fetch, delayed-visibility, unrelated-channel, confirmed-absence fallback, and inconclusive-lookup paths.

How to Test

1. Simulate a direct create that attaches the original-message thread and then raises ConnectionError.
2. On upstream/main@fae29c841, observe one direct thread plus one fallback thread.
3. On this branch, observe the original thread being reused with no fallback seed.
4. Run:

./.venv/bin/python -m pytest \
  tests/gateway/test_discord_channel_controls.py \
  tests/gateway/test_discord_slash_commands.py \
  tests/gateway/test_discord_double_dispatch.py \
  tests/gateway/test_discord_free_response.py \
  tests/gateway/test_discord_thread_persistence.py \
  -q -o addopts=

./.venv/bin/ruff check \
  plugins/platforms/discord/adapter.py \
  tests/gateway/test_discord_channel_controls.py \
  tests/gateway/test_discord_slash_commands.py

Validation

Baseline harness: direct_creates=1, seed_messages=1, fallback_creates=1
Fixed harness:    direct_creates=1, seed_messages=0, fallback_creates=0
Related tests:    139 passed
Ruff:             passed
git diff --check: passed
Environment:      macOS 26.5.2, Python 3.11.15

Remaining Risk

The issue's duplicate-reply observation is real, but its separate ingress path was not isolated. Current main already has live-event deduplication, thread-starter pre-seeding, recovery deduplication, and persistent processing claims, so this PR intentionally does not change dispatch behavior without evidence. It fixes the deterministically reproduced duplicate-thread trigger.

Checklist

• I've read the Contributing Guide.
• My commit messages follow Conventional Commits.
• I searched existing PRs and documented the relevant overlap above.
• This PR contains only issue-related changes.
• I've added and run focused regression tests.
• I've tested on macOS 26.5.2 with Python 3.11.15.
• Full pytest tests/ -q suite run (focused Discord coverage is reported above).
• Documentation/config/schema changes are N/A.

[alt-glitch]

This was generated by AI during triage.

Related: #73092 handles the in-memory message.thread recovery path for #73032; this PR additionally covers cache misses, REST lookup, delayed visibility, and fail-safe inconclusive lookup. They overlap but are not redundant.

[teknium1]

Thanks for isolating the ambiguous-success create path. The production premise remains valid on current main: plugins/platforms/discord/adapter.py:6580-6590 catches a direct-create exception and immediately posts the fallback seed / creates its separate thread, with no reconciliation step.

Problems

• The test portion needs manual salvage. This PR is based on ef267011, while current main pruned the insertion region in tests/gateway/test_discord_channel_controls.py (the file now reaches its config section at line 219). GitHub reports the PR CONFLICTING/DIRTY; the added tests cannot apply as-is.

Suggested changes

• Preserve the adapter reconciliation logic and relocate its regression coverage into the remaining Discord auto-thread test surface, adapting the current fixture rather than restoring removed tests.

Automated hermes-sweeper review.

[izumi0uu]

Addressed in d55bab7 and 39486ff.
Rebased onto current main, preserved the adapter reconciliation logic, and relocated the regression coverage into the surviving Discord auto-thread test surface in test_discord_slash_commands.py rather than restoring the pruned test_discord_channel_controls.py block.

[GottZ]

This was generated by AI during triage.

Summary

Nine PRs address or reference this Discord auto-thread complex. Their diffs cover the original seed-message fallback, retry/fail-closed routing, orphan-seed cleanup, and, for #73032, reconciliation of an ambiguous direct-create result before any second thread is created.

Related pull requests

• fix(discord): add auto-thread fallback path #8692 [closed] related — (+47/-4) — unrelated: Adds the seed-message fallback after direct creation fails rather than reconciling ambiguous success; although closed, it remains relevant as the implementation merged via fix(discord): auto-thread cluster — free-response + reply + fallback + mention-name-strip #11629 that introduced the fallback path implicated here.
• fix(discord): auto-thread failure must not silently fall back to inline reply (#20243) #20260 [closed] fixes — (+144/-21) — fixes: Retries thread creation and prevents an inline agent reply after definitive failure, but does not reconcile a direct create that committed before raising; although closed, it remains relevant because its work was salvaged into fix(discord): auto-thread failure must not silently fall back to inline reply (#20243) #56108.
• fix(discord): reuse existing auto-thread on create race #48685 [closed] fixes — (+59/-0) — partial fix: Reuses an attached or cache/REST-resolved starter thread before the fallback, but does not validate the result or suppress fallback after an inconclusive lookup; although closed during conflict cleanup, it is the predecessor preserved and expanded by fix(discord): reconcile ambiguous auto-thread creation #73097.
• fix(discord): avoid duplicate auto-thread seed posts #48690 [closed] partial — (+57/-27) — fixes: Reuses an existing starter thread and otherwise removes the seed fallback, preventing a second thread but changing fallback behavior more broadly; closed as a contributor-confirmed duplicate of fix(discord): reuse existing auto-thread on create race #48685.
• fix(discord): delete orphaned auto-thread seed message on fallback failure #52423 best fix — (+161/-0) — best fix: Deletes a seed announcement when fallback thread creation fails and adds focused regression coverage, fixing Discord auto-thread: orphaned “Thread created by Hermes” message + inline reply when thread creation is rate-limited #52422's orphan-message cause without addressing two successful thread creations. Its approved review supports keeping this independent cleanup available for salvage.
• fix(discord): delete orphaned seed message when auto-thread creation fails #52430 [closed] fixes — (+8/-0) — fixes: Implements the same orphan-seed deletion as fix(discord): delete orphaned auto-thread seed message on fallback failure #52423 without regression tests; closed as the contributor-confirmed duplicate of fix(discord): delete orphaned auto-thread seed message on fallback failure #52423.
• fix(discord): auto-thread failure must not silently fall back to inline reply (#20243) #56108 [merged] best fix — (+156/-24) — best fix: The merged implementation retries creation and fails closed before agent execution after definitive failure, resolving [Bug]: Discord auto-thread failure silently falls back to inline channel reply #20243, but it still enters the seed fallback after an ambiguous direct-create exception. It remains relevant as the merged successor to fix(discord): auto-thread failure must not silently fall back to inline reply (#20243) #20260 and the behavioral baseline for [Bug]: Discord auto-thread fallback can create duplicate threads and duplicate agent replies #73032.
• fix(discord): prevent duplicate threads on create_thread exception (#73032) #73092 partial — (+16/-0) — partial fix: Reuses only a pre-populated message.thread, leaving the reproduced cache/REST discovery case uncovered and adding no regression test. Despite its keep_open maintainer-bot verdict, fix(discord): reconcile ambiguous auto-thread creation #73097 supersedes this diff with the same in-memory path plus authoritative reconciliation and tests.
• fix(discord): reconcile ambiguous auto-thread creation #73097 best fix — (+178/-2) — best fix: Reconciles the starter thread through the message, cache, and REST, retries delayed visibility, validates identity/type, and suppresses fallback when lookup remains inconclusive while retaining it after confirmed absence. Consistent with its keep_open review, commits d55bab7 and 39486ff reportedly rebased the branch and the supplied diff shows the requested tests relocated into tests/gateway/test_discord_slash_commands.py.

Duplicates

#20260 was superseded by merged #56108; #48690 duplicates #48685's core recovery mechanism; #52430 duplicates the tested canonical #52423; #73092 is a narrower subset of #73097. The target-fix lineage is #48685 -> #73097.

Suggested consolidation

Keep #73097 open with a salvage path centered on its adapter reconciliation logic and relocated ambiguous-success regression tests; this preserves the Verify-designated best fix without making a merge recommendation. Close #73092 as a duplicate of #73097 despite its keep_open maintainer-bot verdict because its diff covers only message.thread, while #73097 also covers cache/REST lookup, delayed visibility, confirmed absence, and inconclusive lookup; independently keep #52423 open for its #52422 cleanup, with #52430 remaining closed as its duplicate.

Complex graph

flowchart LR
    classDef open fill:#dbeafe,stroke:#1d4ed8,color:#1e3a8a
    classDef merged fill:#dcfce7,stroke:#15803d,color:#14532d
    classDef closed fill:#e5e7eb,stroke:#6b7280,color:#1f2937
    classDef unverified fill:#f3f4f6,stroke:#9ca3af,color:#374151
    classDef best stroke-width:3px,stroke:#b45309
    classDef target stroke-width:3px,stroke:#4338ca
    I73032(["issue #73032 (open)"])
    P73097["PR #73097 (open)"]
    P73097 -->|best fix| I73032
    class I73032 open
    class P73097 open
    class P73097 best
    class P73097 target
    click I73032 "https://github.com/NousResearch/hermes-agent/issues/73032"
    click P73097 "https://github.com/NousResearch/hermes-agent/pull/73097"

Loading

Graph: solid arrow = fixes / best fix, dashed arrow = partial or unverified (see edge label); boxed group = PRs duplicating each other; amber border = best fix; indigo border = target; gray node = closed (state tag in the node label).

Cross-PR triage: Reviewed 9 pull requests and 3 issues in this complex. Each diff was read against this issue; Assessment working set: 60 kB of PR diffs, 33 kB of issue/PR text, 11 kB of discussion (14 comments), 23 verify verdicts. verdicts reflect diff content, not PR titles. Part of an automated triage batch.