fix(plugins): register entry-point plugins declared as module:function

[mannnrachman]

What does this PR do?

Makes pip-installed plugins with module:function entry points actually register.

importlib.metadata.EntryPoint.load() resolves the module:function form to the referenced attribute (typically the register callable itself), not the module. PluginManager._load_plugin() then looked for a .register attribute on that function object, found none, and logged Plugin '<name>' has no register() function — once per discovery pass, which for a gateway with an every-minute cron tick means a warning per minute, forever. Such plugins could never register at all.

The loader now detects a non-module callable returned by ep.load() and uses it directly as the register function. LoadedPlugin.module is resolved from sys.modules via the callable's __module__ so attribution and hermes plugins list reporting stay intact. Bare module entry points and directory plugins are unaffected.

Related Issue

Fixes #72052

Type of Change

• 🐛 Bug fix (non-breaking change that fixes an issue)
• ✨ New feature (non-breaking change that adds functionality)
• 🔒 Security fix
• 📝 Documentation update
• ✅ Tests (adding or improving test coverage)
• ♻️ Refactor (no behavior change)
• 🎯 New skill (bundled or hub)

Changes Made

• hermes_cli/plugins.py
  • _load_plugin(): if the entry-point load result is a callable that is not a module, treat it as the register function directly and resolve the owning module from sys.modules for LoadedPlugin.module.
  • _load_entrypoint_module(): widen the return annotation and document that ep.load() returns the module for bare module entry points, or the referenced attribute for the module:function form.
• tests/hermes_cli/test_plugins.py
  • New regression test test_entry_point_function_form_registers whose mocked ep.load() returns the function — mirroring real importlib behavior. The pre-existing entry-point test masked this bug because its mock returned the module even though its declared entry-point value was module:function.

How to Test

1. Install any pip plugin whose entry point is declared name = package.module:register (e.g. mnemosyne-hermes), enable it via hermes plugins enable <name>, and confirm it registers instead of warning has no register() function on every discovery pass.
2. Run the focused suite:

scripts/run_tests.sh tests/hermes_cli/test_plugins.py

3. Run the cross-platform check:

python3 scripts/check-windows-footguns.py --diff origin/main

Validation Results

• tests/hermes_cli/test_plugins.py: 117 passed, 0 failed (Python 3.11.15, Ubuntu 22.04 x86_64).
• The new regression test fails on unpatched main with exactly the reported error (AssertionError: no register() function) and passes with the fix.
• Windows-footgun diff check: passed (0 findings).
• The full repository suite was not run locally; hosted CI remains authoritative for the complete matrix.

Checklist

Code

• I've read the Contributing Guide
• My commit messages follow Conventional Commits (fix(scope):, feat(scope):, etc.)
• I searched open and merged issues/PRs; no open PR addresses Entry-point plugins declared as module:function never register — loader calls getattr(register, 'register') and warns every discovery pass #72052
• My PR contains only changes related to this fix
• I've run pytest tests/ -q and all tests pass — focused suite above is green; see the transparency note in Validation Results
• I've added tests for my changes
• I've tested on Ubuntu 22.04 x86_64 with Python 3.11.15

Documentation & Housekeeping

• Documentation N/A — no user-facing command or configuration key was added
• cli-config.yaml.example N/A — no configuration key was added or changed
• CONTRIBUTING.md / AGENTS.md N/A — no architecture or contributor workflow changed
• Cross-platform impact considered; scripts/check-windows-footguns.py --diff origin/main passes
• Tool descriptions/schemas N/A — no tool API changed

Screenshots / Logs

No UI changes. Loader behavior before/after on a module:function entry point:

before: WARNING hermes_cli.plugins: Plugin 'fn_plugin' has no register() function   (every discovery pass)
after : registered — enabled=True, error=None, register(ctx) called once

🤖 Generated with Claude Code

[mannnrachman]

The check-windows-footguns.py --all CI failure is pre-existing on main, not introduced here: a2c42be missed one bare write_text at scripts/tool_search_livetest2.py:190, so the --all gate currently fails for every PR. Fix submitted separately as #72097; this PR's own diff passes --diff clean.

[teknium1]

Thanks for the focused fix. The premise remains valid on current main: hermes_cli/plugins.py:1786 receives the entry-point result and hermes_cli/plugins.py:1791 only looks for .register, while the documented provider entry-point form is acme_hermes_plugin:register at website/docs/developer-guide/model-provider-plugin.md:254-258.

Problems

• The regression-test hunk needs relocation during salvage. Its anchor, test_entry_points_scanned, was removed from current main by 39975613b13b418e0eceda178434d7be90ad4f91; current tests/hermes_cli/test_plugins.py therefore has no equivalent entry-point loader test to patch beside.

Suggested changes

• Preserve the proposed callable-return regression under the remaining TestPluginDiscovery coverage when applying the production change.

This is an automated hermes-sweeper review.

[GottZ]

This was generated by AI during triage.

Summary

One PR addresses issue #72052. #72094 directly fixes the callable-versus-module mismatch by treating the callable returned for a module:function entry point as the registration function while retaining its owning module for attribution.

Related pull requests

• fix(plugins): register entry-point plugins declared as module:function #72094 best fix — (+66/-3) — keep open with a salvage path: The production diff fixes the reported root cause in hermes_cli/plugins.py by invoking a non-module callable returned by EntryPoint.load() directly, and the test models that exact resolution behavior. Consistent with the automated keep_open review on fix(plugins): register entry-point plugins declared as module:function #72094, preserve this production change and relocate the regression test under the remaining TestPluginDiscovery coverage because commit 3997561 removed its current test_entry_points_scanned anchor from main.

Suggested consolidation

Keep #72094 open with a salvage path: retain its callable-return loader fix and its assertions for successful registration, absence of a loader error, and owning-module attribution, but rebase and relocate the regression coverage to the current TestPluginDiscovery structure. It is the only PR for #72052, so there are no 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
    I72052(["issue #72052 (open)"])
    P72094["PR #72094 (open)"]
    P72094 -->|best fix| I72052
    class I72052 open
    class P72094 open
    class P72094 best
    class P72094 target
    click I72052 "https://github.com/NousResearch/hermes-agent/issues/72052"
    click P72094 "https://github.com/NousResearch/hermes-agent/pull/72094"

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: 4 kB of PR diffs, 8 kB of issue/PR text, <1 kB of discussion (1 comments), 2 verify verdicts. verdicts reflect diff content, not PR titles. Part of an automated triage batch.