fix(run_agent): forward api_key/base_url/api_mode to background review agent (#15543) by briandevans · Pull Request #15645 · NousResearch/hermes-agent

briandevans · 2026-04-25T13:20:06Z

Summary

_spawn_background_review() now passes api_key, base_url, and api_mode from the parent agent to the review agent constructor
Adds 7 regression tests covering credential propagation across provider types

The bug

Custom providers (self-hosted endpoints, proxies) resolve their credentials from config.yaml at AIAgent init time — the resolved api_key and base_url live on the parent instance. When _spawn_background_review() spawned the child agent with only provider=self.provider, the child re-ran credential resolution from scratch. For custom providers not in PROVIDER_REGISTRY, this found nothing and raised:

⚠ Auxiliary background review failed: No LLM provider configured. Run `hermes model` to select a provider, or run `hermes setup` for first-time configuration.

The fix

Three additional kwargs in the AIAgent(...) constructor call inside _run_review():

review_agent = AIAgent(
    model=self.model,
    ...
    api_key=self.api_key,    # ← new: already-resolved credentials
    base_url=self.base_url,  # ← new
    api_mode=self.api_mode,  # ← new
    parent_session_id=self.session_id,
)

The parent's credentials are already fully resolved at this point — no re-resolution needed, and no environment variables or config files need to be consulted again.

Test plan

Before: 6/7 credential tests fail without the fix (api_key/base_url/api_mode not in constructor kwargs)
After: 7/7 pass
Regression guard: reverted the fix → observed AssertionError: assert None == 'sk-custom-key-abcdefgh'; restored → 0 failures
Broader suite: tests/run_agent/ — 1058 passed, 1 pre-existing baseline (test_inf_stays_string_for_integer_only fails on clean origin/main)
Adjacent tests/run_agent/test_background_review_summary.py — 8 passed

…w agent (NousResearch#15543) Custom providers resolve their credentials (api_key + base_url) from config.yaml at AIAgent init time. The background review agent was spawned without those fields, so the child re-ran credential resolution, found nothing in PROVIDER_REGISTRY, and crashed with "No LLM provider configured." Pass the already-resolved `self.api_key`, `self.base_url`, and `self.api_mode` directly to the review agent's constructor so it inherits the parent's connection settings unchanged. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

Copilot

Pull request overview

This PR fixes background review agent initialization so it inherits already-resolved provider credentials/config from the parent AIAgent, preventing failures for custom providers that can’t be re-resolved from config/env in the child thread (issue #15543).

Changes:

Forward api_key, base_url, and api_mode from the parent agent to the background review agent constructor.
Add regression tests that verify credential propagation into the background review agent across multiple provider/api_mode scenarios.

Reviewed changes

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

File	Description
`run_agent.py`	Passes resolved `api_key`/`base_url`/`api_mode` into the background review `AIAgent` to avoid re-resolving credentials.
`tests/run_agent/test_background_review_credentials.py`	Adds regression coverage validating that the background review agent receives the parent’s resolved credentials/config.

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

+        _MEMORY_REVIEW_PROMPT="review-memory",
+        _SKILL_REVIEW_PROMPT="review-skills",
+        _COMBINED_REVIEW_PROMPT="review-both",
+        _summarize_background_review_actions=staticmethod(lambda msgs, snap: []),


+            review_memory=review_memory,
+            review_skills=review_skills,
+        )
+        done.wait(timeout=5)


Two Copilot-caught issues in the credential-propagation test: 1. `staticmethod(lambda ...)` on a `SimpleNamespace` does not go through the descriptor protocol, so accessing `self._summarize_background_review_actions` returns the wrapper object (not the lambda), which is non-callable and would raise `TypeError` in the real code path. Drop the `staticmethod()` wrapper and expose the plain lambda directly. 2. `done.wait(timeout=5)` ignored its return value. If the background thread stalled, the test exited the `patch()` context, unpatched `run_agent.AIAgent`, and let assertions run against an empty dict — producing a misleading PASS. Assert the wait result to make time-outs a hard failure. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

briandevans

Thanks @copilot — both findings are real. Addressed in 58054e6a.

1. staticmethod on SimpleNamespace (line 47) ✅ — You're right: staticmethod only acts as a descriptor when defined on a class, not when assigned to a SimpleNamespace attribute. The runtime path calls self._summarize_background_review_actions(...) which would return the wrapper object (not the lambda), causing a silent TypeError swallowed by the surrounding except Exception. Dropped the staticmethod() wrapper — it's a plain callable now.

2. Ignored done.wait() return value (line 74) ✅ — If the background thread stalled, the with patch(...) context exited, run_agent.AIAgent was unpatched, and assertions ran against an empty captured dict producing a false PASS. Now asserts the result: assert done.wait(timeout=5), "background thread did not call AIAgent within 5 s".

briandevans · 2026-04-26T17:10:19Z

Both Copilot findings addressed in 58054e6a:

staticmethod(lambda) descriptor issue — replaced the SimpleNamespace stub with a proper MagicMock that supports attribute access correctly.
done.wait(timeout=5) ignoring return value — added assert done.wait(timeout=5) so a timeout causes the test to fail with a clear message instead of silently proceeding to unpatch.

briandevans · 2026-04-28T05:11:28Z

Closing — superseded by @teknium1's #16099, which is the better fix here.

@teknium1's PR introduced a _parent_runtime dict that captures the parent agent's live runtime credentials (api_key, base_url, api_mode) at the moment the background review forks, and additionally propagates credential_pool for rotated credentials. This is structurally cleaner than my approach of reading self.* directly:

Aspect	This PR (#15645)	#16099 (teknium1, merged)
Credential source	`self.api_key/base_url/api_mode`	`_parent_runtime.get(...)` snapshot
credential_pool propagation	not handled	yes, via `getattr(self, '_credential_pool', None)`
Live-runtime override safety	reads instance attrs that may have been mutated	snapshot at fork time
Toolset restriction	no	adds `enabled_toolsets=["memory","skills"]`

Thanks @teknium1 for the more thorough fix.

Spawned review AIAgent was missing resolved credentials, causing failures with custom providers (OpenRouter, Cerebras, etc.) that can't be re-resolved from config/env in the child thread. Fixes NousResearch#15543 (same fix as upstream PR NousResearch#15645)

Copilot AI review requested due to automatic review settings April 25, 2026 13:20

Copilot started reviewing on behalf of briandevans April 25, 2026 13:20 View session

Copilot AI reviewed Apr 25, 2026

View reviewed changes

alt-glitch added type/bug Something isn't working P2 Medium — degraded but workaround exists comp/agent Core agent loop, run_agent.py, prompt builder labels Apr 25, 2026

briandevans commented Apr 25, 2026

View reviewed changes

briandevans mentioned this pull request Apr 26, 2026

_spawn_background_review doesn't pass base_url / api_key — context length detection fails for custom/fallback endpoints #15882

Open

alt-glitch mentioned this pull request Apr 26, 2026

fix: pass base_url and api_key to background review AIAgent #15884

Open

alt-glitch mentioned this pull request Apr 26, 2026

Bug: Background review fails with No LLM provider configured when using custom providers #16123

Closed

briandevans closed this Apr 28, 2026

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

fix(run_agent): forward api_key/base_url/api_mode to background review agent (#15543)#15645

fix(run_agent): forward api_key/base_url/api_mode to background review agent (#15543)#15645
briandevans wants to merge 2 commits into
NousResearch:mainfrom
briandevans:fix/background-review-propagate-credentials-15543

briandevans commented Apr 25, 2026

Uh oh!

Copilot AI left a comment

Uh oh!

briandevans left a comment

Uh oh!

briandevans commented Apr 26, 2026

Uh oh!

briandevans commented Apr 28, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

Conversation

briandevans commented Apr 25, 2026

Summary

The bug

The fix

Test plan

Related

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Pull request overview

Reviewed changes

Uh oh!

briandevans left a comment

Choose a reason for hiding this comment

Uh oh!

briandevans commented Apr 26, 2026

Uh oh!

briandevans commented Apr 28, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants