fix(qqbot): add AI Agent identity framing to platform hint (#66959)

[dsitmilis]

Summary

Fixes #66959 — the QQ bot adapter works, but PLATFORM_HINTS["qqbot"] lacked explicit agent-identity framing. Weaker/free models routed via OmniRoute-style aggregators then skip skill_view() and rely only on the short <available_skills> description, e.g. failing to know the default transaction account.

Mirror the CLI hint (You are a CLI AI Agent) by leading the QQ hint with You are a QQ AI Agent and reminding the model to load skill bodies via skill_view.

Changes

• agent/prompt_builder.py: qqbot hint now frames the model as an agent and references skills.
• tests/agent/test_prompt_builder.py: added test_qqbot_hint_frames_agent_identity.

Validation

• python3 -m pytest tests/agent/test_prompt_builder.py → 165 passed (the 2 unrelated failures — test_excludes_disabled_skills, test_walks_parents_inside_git_repo — also fail on clean upstream main and are pre-existing/ordering-related, not introduced by this change).

Closes #66959.

[alt-glitch]

This was generated by AI during triage.

Re-triage correction: the SimpleX repair was split into #66969. The live diff here is QQBot-only, related to #66959 and the distinct volatile-prompt approach in #63633; the claimed weaker-model effect still needs reproduction.

[teknium1]

Thanks for the focused fixes. Both target behaviors are still present on current main, but the combined change needs separation and the SimpleX repair is incomplete.

Problems

• SimplexAdapter.send() still uses the bare DM command at plugins/platforms/simplex/adapter.py:841, but the same defect remains in the standalone cron/proactive fallback: _standalone_send() is used when no live adapter exists (plugins/platforms/simplex/adapter.py:1170-1173) and still emits the bare command at plugins/platforms/simplex/adapter.py:1202. Its current test preserves that form at tests/gateway/test_simplex_plugin.py:352.
• This combines the QQ hint with an unrelated SimpleX delivery repair. The existing member comment identifies fix(agent): reinforce WeChat, WeCom, and QQ platform identity #63633 as a separate QQ prompt-tier approach and requests split scope.
• The added QQ test checks substrings in PLATFORM_HINTS; it does not exercise the platform-hint insertion path in agent/system_prompt.py:417-450.

Suggested changes

• Split the QQ and SimpleX changes.
• For the SimpleX fix, update the standalone fallback and its test to the same structured DM command.
• Add an assembled-system-prompt test for qqbot rather than only constant substring assertions.

Automated hermes-sweeper review.

[dsitmilis]

Thanks for the review. Scope is now split:

• This PR (fix(qqbot): add AI Agent identity framing to platform hint (#66959) #66976) is qqbot-only: the PLATFORM_HINTS["qqbot"] agent-identity framing. Per the suggestion I also added a real insertion-path test (tests/agent/test_system_prompt.py::TestQqbotHint) that builds the system prompt for platform="qqbot" and asserts "You are a QQ AI Agent" is present (not just a constant substring check).
• The SimpleX DM repair was moved out to its own PR fix(simplex): send DMs by numeric contactId via /_send, not bare @<id> #66969, which now also covers the standalone _standalone_send() fallback.

Verified: tests/agent/test_prompt_builder.py + tests/agent/test_system_prompt.py → 175 passed (the 2 unrelated failures in test_prompt_builder.py also fail on clean upstream main).

[alt-glitch]

This was generated by AI during triage.

Re-triage update: the SimpleX change has been split into #66969. This branch is now QQBot-only and remains related to #66959 and the distinct volatile-prompt approach in #63633.

[alt-glitch]

This was generated by AI during triage.

Related to report #66959 and the competing volatile-prompt approach in #63633. This patch changes QQ's platform hint; the reported weaker-model causal effect still needs reproduction.

[GottZ]

This was generated by AI during triage.

Summary

One PR, #66976, directly addresses issue #66959 by adding QQ-specific AI-agent and skill-loading guidance to the platform hint, with tests for both the hint content and its insertion into the assembled system prompt. The diff implements the reported prompt-surface change, while the contributor discussion states that the claimed weaker-model behavioral effect still needs reproduction.

Related pull requests

• fix(qqbot): add AI Agent identity framing to platform hint (#66959) #66976 best fix — (+32/-4) — n/a: The QQBot-only diff adds explicit “QQ AI Agent” framing and directs models to load skill bodies with skill_view(), directly changing the prompt surface identified in fix: qqbot platform hint lacks agent-identity framing, causing weaker models to skip skill_view() #66959; it also tests the real system-prompt insertion path. Consistent with the automated keep_open review, the salvageable focused change should remain open: its earlier split-scope and insertion-test concerns are addressed by the visible diff, while empirical reproduction of the weaker-model effect remains the author action needed.

Duplicates

No listed PR duplicates #66976; #63633 is a related but distinct volatile-system-prompt approach, and the unrelated SimpleX work was split into #66969.

Suggested consolidation

Keep #66976 open with a salvage path focused on its QQ platform-hint change and assembled-system-prompt regression test. Ask the author to provide a reproducible comparison demonstrating the claimed weaker-model skill_view() behavior; there are no listed duplicate PRs to close.

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
    I66959(["issue #66959 (open)"])
    P66976["PR #66976 (open)"]
    P66976 -->|best fix| I66959
    class I66959 open
    class P66976 open
    class P66976 best
    class P66976 target
    click I66959 "https://github.com/NousResearch/hermes-agent/issues/66959"
    click P66976 "https://github.com/NousResearch/hermes-agent/pull/66976"

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 1 pull request and 1 issue in this complex. Each diff was read against this issue; Assessment working set: 3 kB of PR diffs, 3 kB of issue/PR text, 2 kB of discussion (5 comments), 2 verify verdicts. verdicts reflect diff content, not PR titles. Part of an automated triage batch.