fix(gateway): respect own-policy group_policy when resolving unauthorized DM fallback

[Dusk1e]

Summary

This brings the config-driven own-policy gateway adapters into parity with the existing allowlist-aware unauthorized-DM fallback.

Before this change, GatewayRunner._get_unauthorized_dm_behavior() already defaulted to ignore when env-based user/group allowlists were configured, but the config-driven own-policy path only looked at dm_policy. That left a gap for WeCom, Weixin, Yuanbao, and QQBot: operators could restrict group access with group_policy: allowlist or group_policy: disabled, yet unauthorized DMs would still receive pairing codes unless they also set an explicit DM restriction or explicit unauthorized_dm_behavior.

This patch treats restrictive group_policy values on own-policy adapters the same way as the existing allowlist-aware fallback:

• group_policy: allowlist => ignore
• group_policy: disabled => ignore

Explicit overrides still win:

• per-platform unauthorized_dm_behavior
• global unauthorized_dm_behavior
• dm_policy: pairing

Why this is correct

These adapters already enforce group_policy at intake, so a restrictive group config is an operator-declared access restriction, just like the env-based allowlist cases already handled in _get_unauthorized_dm_behavior().

That makes the previous behavior inconsistent and noisy: a stranger DM could still get a pairing code even though the operator had intentionally locked group access down on the same adapter.

Parity / prior art

This is a parity follow-up to the allowlist-aware unauthorized-DM fallback work from #9337.

#9337 taught _get_unauthorized_dm_behavior() to default to ignore when env-based allowlists were present, to avoid pairing spam / unnecessary info leakage to unauthorized contacts. This PR applies the same principle to the config-driven own-policy sibling path.

Related background: #34515 established the own-policy adapter contract and gateway trust path for WeCom / Weixin / Yuanbao / QQBot.

Changes

• Extend _get_unauthorized_dm_behavior() to treat restrictive group_policy values as ignore for adapters that declare enforces_own_access_policy.
• Keep explicit unauthorized_dm_behavior: pair behavior unchanged.
• Add regression coverage for:
  • WeCom / Weixin / Yuanbao / QQBot with group_policy in {"allowlist", "disabled"}
  • explicit per-platform unauthorized_dm_behavior: pair overriding the restrictive default

Tests

Passed:

• tests/gateway/test_config_driven_access_policy.py
• tests/gateway/test_unauthorized_dm_behavior.py
• tests/gateway/test_wecom.py
• tests/gateway/test_weixin.py
• tests/gateway/test_qqbot.py
• tests/test_yuanbao_pipeline.py

Local result:

• 413 passed, 4 warnings

The warnings are pre-existing QQBot test warnings and are not introduced by this change.

[teknium1]

Thanks for the focused parity fix. The premise remains present on current main: GatewayAuthorizationMixin._get_unauthorized_dm_behavior() checks restrictive dm_policy values at gateway/authz_mixin.py:653-669, then falls through to pair at gateway/authz_mixin.py:710; it does not inspect group_policy.

Problems

• The PR targets the pre-refactor location in gateway/run.py; current GatewayRunner now inherits GatewayAuthorizationMixin at gateway/run.py:2775, and GitHub reports this branch as conflicting.
• Current resolution is profile-aware. A direct config.extra read would bypass the live profile adapter, while _adapter_group_policy(..., profile=profile) already provides the required effective-policy lookup at gateway/authz_mixin.py:161-194.

Suggested changes

• Salvage the group-policy branch into gateway/authz_mixin.py:607 using _adapter_enforces_own_access_policy(..., profile=profile) and _adapter_group_policy(..., profile=profile).
• Port the regression tests and add a multiplexed secondary-profile group-policy case; current multiplex coverage covers only DM policy at tests/gateway/test_multiplex_profile_authz.py:130-139.

This is an automated hermes-sweeper review.

[teknium1]

Current main moved this resolver to gateway/authz_mixin.py and made it profile-aware. When salvaging, use _adapter_group_policy(platform, profile=profile) behind _adapter_enforces_own_access_policy(..., profile=profile) rather than reading only config.extra, so multiplexed and environment-resolved adapter policies are honored.

[GottZ]

This was generated by AI during triage.

Summary

Seven PRs address related own-policy gateway authorization gaps, but they cover distinct paths: #45444/#45634 establish fail-closed baseline semantics, #37916/#56838/#62340 address Feishu chat-scoped rules, #55875 addresses WhatsApp group authorization, and #34791 addresses unauthorized-DM fallback under restrictive group policy. The diffs show that the remaining fixes must use current, profile-aware effective-policy resolution rather than stale files, adapter-wide flags, or the default adapter registry.

Related pull requests

• #34791 related — (+50/-9) — salvage: The diff correctly makes restrictive own-policy group settings suppress pairing-code replies, but it modifies the pre-refactor gateway/run.py path and reads config.extra directly. Consistent with the keep_open review on #34791, port it to gateway/authz_mixin.py using profile-aware effective group-policy lookup and add a secondary-profile regression.
• #37916 related — (+13/-0) — superseded: The diff adds only enforces_own_access_policy to the obsolete gateway/platforms/feishu.py location and cannot represent Feishu's per-chat group_rules policy. Despite the keep_open review on #37916, the complete diff is neither current-main-applicable nor sufficient; #62340 implements the required exact-rule handling in the bundled adapter and gateway.
• #45444 [closed] related — (+158/-55) — merged via #45634: This closed PR supplied the core fail-closed change, trusting own-policy adapters only for effective allowlist restrictions rather than default open or pairing behavior. It remains relevant as the authored source that #45634 cherry-picked and extended.
• #45634 [merged] related — (+258/-55) — merged reference implementation: It landed #45444's fail-closed authorization model and added the WeCom groups.<group_id>.allow_from sibling case with regression coverage. This is the baseline against which the remaining platform-specific diffs must be integrated.
• #55875 related — (+126/-0) — salvage: The diff correctly separates WhatsApp's chat-scoped group allowlist from its DM/user allowlist, but resolves self.adapters without source.profile and therefore risks authorizing a secondary-profile message from the default profile's adapter. Consistent with the keep_open review on #55875, retain the narrow fix only after switching to the profile-aware adapter resolver and adding an isolation regression.
• #56838 related — (+14/-0) — superseded duplicate of #37916: The diff updates the current Feishu adapter but still adds only a static capability flag plus generic flag coverage; it does not resolve effective group_rules policy for source.chat_id. Despite the keep_open review on #56838, #62340's exact active-rule check addresses the documented open/allowlist inversion that this diff leaves unresolved.
• #62340 related — (+170/-7) — merge candidate: Consistent with the keep_open review on #62340, the diff adds profile-aware lookup and trusts only an exact active Feishu group_rules entry, preserving fail-closed behavior for broad/default open policy and disabled rules. Its complete six-file diff does not add a secondary-profile or multiplex isolation test, so that regression should be added before merge even though the implementation passes source.profile into authorization lookup.

Duplicates

#37916 and #56838 are substantially the same insufficient capability-flag-only Feishu fix; #56838 merely targets the migrated adapter and extends generic flag coverage. #45444 and #45634 contain the same core fail-closed hardening, with #45634 being the merged salvage plus the WeCom per-group follow-up.

Suggested consolidation

Merge #62340 after adding an explicit secondary-profile/multiplex isolation regression; close #37916 and #56838 as superseded duplicates. Keep #45634 as the merged baseline and #45444 closed as its preserved source, then salvage #34791 and #55875 as separate profile-aware follow-ups because their unauthorized-DM and WhatsApp group-chat paths are not implemented by #62340.

Cross-PR triage: Reviewed 7 pull requests and 0 issues in this complex. Each diff was read against this issue; Assessment working set: 67 kB of PR diffs, 12 kB of issue/PR text, 16 kB of discussion (15 comments), 2 verify verdicts. verdicts reflect diff content, not PR titles. Part of an automated triage batch.