fix: guard empty streams + parse OpenRouter output-cap errors#40589
Merged
Conversation
…ut-cap errors
Two isolated reliability fixes:
- chat_completion_helpers: raise on a zero-chunk stream (no finish_reason,
no content/reasoning/tool_calls) so retry handles it instead of
fabricating a successful empty turn.
- model_metadata: parse the OpenRouter/Nous output-cap error phrasing
("maximum context length is N ... (A of text input, B of tool input,
C in the output)") so parse_available_output_tokens_from_error returns
a real cap and the caller stops looping on it.
Salvaged from #40405 (@ashishpatel26) — took the two stream/error-parsing
fixes. The PR also bundled compression-state changes (on_session_start
clearing _previous_summary; cron session-id prefix preservation, #38788);
those touch the compression hot path and are split out for separate review.
Co-authored-by: ashishpatel26 <ashishpatel26@users.noreply.github.com>
4 tasks
Contributor
🔎 Lint report:
|
| Rule | Count |
|---|---|
unresolved-import |
1 |
First entries
tests/test_output_cap_parsing.py:1: [unresolved-import] unresolved-import: Cannot resolve imported module `pytest`
✅ Fixed issues: none
Unchanged: 5167 pre-existing issues carried over.
Diagnostics are surfaced as warnings — this check never fails the build.
agogo233
added a commit
to agogo233/hermes-agent
that referenced
this pull request
Jun 8, 2026
* upstream/main: (430 commits) fix(yuanbao): bound ws.close() so an idle server can't stall shutdown ~5s (NousResearch#40607) docs: add Urdu translation of README (NousResearch#40578) fix(hindsight): send only new-turn delta on append retains instead of whole session (NousResearch#40605) feat(gateway): render terminal tool calls as native bash code blocks on markdown platforms (NousResearch#41215) feat(desktop): stop the chat viewport from following streaming output (NousResearch#41414) chore(release): map AlchemistChaos co-author email for NousResearch#40135 salvage fix(desktop): recover chat after sleep/wake by revalidating a stale remote backend fix(web): make _has_env config-aware so SEARXNG_URL auto-detect honors Hermes config fix(web): honor Hermes config-aware SEARXNG_URL lookup install.sh: hint at root-owned npm cache when desktop npm install fails (NousResearch#39688) fix(tools): percent-encode non-ascii URL components fix(skills): browse shows full catalog, not first 5000 (NousResearch#41413) feat(desktop+gateway): remote media relay — attach images/PDFs and display gateway images over the network feat(desktop): full tool-backend config (pickers + per-backend settings) in Settings (NousResearch#41232) hardening(api-server): scan cron prompts on REST create/update for parity with the agent tool fix: skip MCP preflight content-type probe on reconnect when already ready (NousResearch#40604) fix(kanban): sweep deferred scratch parent on non-scratch child completion + tests fix: defer scratch workspace cleanup when task has active children (NousResearch#33774) feat(onboarding): opt-in structured profile-build path on first contact (NousResearch#41114) feat(compression): temporal anchoring in compaction summaries (NousResearch#41102) test(discord): align clarify/model-picker tests with fail-closed component auth (NousResearch#41338) chore(release): map Dusk1e and LaPhilosophie for approval fail-closed salvage (NousResearch#33844, NousResearch#33866, NousResearch#30964) fix(discord): fail closed for component button auth when no allowlist set fix(feishu): fail closed for update prompt card actions fix(slack): re-check gateway auth on approval and slash-confirm buttons fix: guard int(os.getenv()) casts against malformed env vars (NousResearch#40598) fix: respect Honcho env var fallback in doctor and honcho status chore(release): add synapsesx to AUTHOR_MAP for NousResearch#40495 salvage fix(research): keep tool_call/tool_response pairs intact when compressing trajectories fix(simplex): accept display name in SIMPLEX_ALLOWED_USERS fix(desktop): make the running-turn timer per-session (NousResearch#41182) test(approval): regression for shell-escape denylist bypass (NousResearch#36846, NousResearch#36847) fix(security): strip shell escapes in denylist normalizer; fail-closed on missing approval module fix(stream+output-cap): guard empty streams and parse OpenRouter output-cap errors (NousResearch#40589) fix(desktop): bootstrap falls back to installed agent install.sh on GitHub 404 feat(dashboard): change UI font from the theme picker, independent of theme (NousResearch#41145) fix(cli): return bool (not None) when a destructive-slash confirmation is cancelled (NousResearch#40583) fix(desktop): preserve configured base_url on same-provider model switch (NousResearch#41121) fix(desktop): stop bare-URL autolinker swallowing trailing emphasis asterisks (NousResearch#41093) fix(cron): bound the desktop run-history query to one job (NousResearch#41088) fix(desktop): scope in-session /model switch per-session, stop process-env leak (NousResearch#41120) chore: map bmoore210 author email for PR NousResearch#40550 salvage fix(desktop): scope session list to active profile + longer timeout fix: harden gateway startup and turn persistence fix(computer_use): honor custom vision routing fix(aux): honor model.default_headers on auxiliary client too (NousResearch#40033) fix(agent): honor model.default_headers for custom OpenAI-compatible providers (NousResearch#40033) docs(i18n): port deep-audit corrections to zh-Hans mirror (NousResearch#41104) fix(compression): don't overwrite the -1 post-compression sentinel in preflight seed (NousResearch#36718) chore(release): map singhsanidhya741@gmail.com to sanidhyasin (NousResearch#41094) ...
changman
pushed a commit
to changman/hermes-agent
that referenced
this pull request
Jun 10, 2026
…ut-cap errors (NousResearch#40589) Two isolated reliability fixes: - chat_completion_helpers: raise on a zero-chunk stream (no finish_reason, no content/reasoning/tool_calls) so retry handles it instead of fabricating a successful empty turn. - model_metadata: parse the OpenRouter/Nous output-cap error phrasing ("maximum context length is N ... (A of text input, B of tool input, C in the output)") so parse_available_output_tokens_from_error returns a real cap and the caller stops looping on it. Salvaged from NousResearch#40405 (@ashishpatel26) — took the two stream/error-parsing fixes. The PR also bundled compression-state changes (on_session_start clearing _previous_summary; cron session-id prefix preservation, NousResearch#38788); those touch the compression hot path and are split out for separate review. Co-authored-by: ashishpatel26 <ashishpatel26@users.noreply.github.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
Two isolated reliability fixes carved out of #40405:
agent/chat_completion_helpers.py): a stream that yields nothing usable (nofinish_reason, no content/reasoning/tool_calls) is a provider/upstream error or malformed SSE, not a legitimate empty completion. Raise so the retry machinery handles it instead of fabricating a successful empty turn.agent/model_metadata.py):parse_available_output_tokens_from_errornow recognizes the"maximum context length is N … (A of text input, B of tool input, C in the output)"phrasing and computes the available output cap, so the caller stops looping on that 400.Scope note
The original PR (#40405) also bundled two compression-state changes (
on_session_startclearing_previous_summary, and cron session-id prefix preservation through compression, citing #38788). Those touch the compression hot path and are a different concern — split out for separate review rather than landing alongside the stream/error-parsing fixes.Validation
tests/test_output_cap_parsing.py4 passed (incl. Anthropic-format regression + no-room → None). py_compile OK on both files.Salvaged from #40405 (@ashishpatel26), credited.