fix(codex): drop transient rs_tmp reasoning replay state - #33146
Merged
Conversation
Contributor
🔎 Lint report:
|
This was referenced May 27, 2026
alt-glitch
added
type/bug
teknium1
pushed a commit
that referenced
this pull request
May 27, 2026
reasoning.encrypted_content is sealed to the Responses endpoint that minted it. When a session switches model providers mid-conversation — say the user runs /model gpt-5.5 after several turns on grok-4.3, or vice versa — the persisted codex_reasoning_items carry blobs the new endpoint cannot decrypt, and every subsequent turn fails with HTTP 400 invalid_encrypted_content. This is the cross-issuer prevention layer. Pairs with: * PR #33035 — runtime recovery when the HTTP 400 fires anyway * PR #33146 — prevention for transient rs_tmp_* items Stamps each reasoning item with the issuer kind that minted it (codex_backend / xai_responses / github_responses / other:<url>) at normalize time, then drops items at replay time when the active endpoint differs from the stamp. Unstamped (legacy) items pass through for backwards compatibility. Cherry-picked from @chaconne67's PR #31629. Conflict against current main (#33035's replay_encrypted_reasoning parameter) resolved as 'keep both' — the two guards compose: replay_encrypted_reasoning=False is the session-wide kill switch, current_issuer_kind is the per-item filter that runs only when replay is still enabled.
1 task
mathias3
pushed a commit
to mathias3/hermes-agent
that referenced
this pull request
May 28, 2026
reasoning.encrypted_content is sealed to the Responses endpoint that minted it. When a session switches model providers mid-conversation — say the user runs /model gpt-5.5 after several turns on grok-4.3, or vice versa — the persisted codex_reasoning_items carry blobs the new endpoint cannot decrypt, and every subsequent turn fails with HTTP 400 invalid_encrypted_content. This is the cross-issuer prevention layer. Pairs with: * PR NousResearch#33035 — runtime recovery when the HTTP 400 fires anyway * PR NousResearch#33146 — prevention for transient rs_tmp_* items Stamps each reasoning item with the issuer kind that minted it (codex_backend / xai_responses / github_responses / other:<url>) at normalize time, then drops items at replay time when the active endpoint differs from the stamp. Unstamped (legacy) items pass through for backwards compatibility. Cherry-picked from @chaconne67's PR NousResearch#31629. Conflict against current main (NousResearch#33035's replay_encrypted_reasoning parameter) resolved as 'keep both' — the two guards compose: replay_encrypted_reasoning=False is the session-wide kill switch, current_issuer_kind is the per-item filter that runs only when replay is still enabled.
Bryce-huang
pushed a commit
to wbkunlun/hermes-agent
that referenced
this pull request
May 29, 2026
reasoning.encrypted_content is sealed to the Responses endpoint that minted it. When a session switches model providers mid-conversation — say the user runs /model gpt-5.5 after several turns on grok-4.3, or vice versa — the persisted codex_reasoning_items carry blobs the new endpoint cannot decrypt, and every subsequent turn fails with HTTP 400 invalid_encrypted_content. This is the cross-issuer prevention layer. Pairs with: * PR NousResearch#33035 — runtime recovery when the HTTP 400 fires anyway * PR NousResearch#33146 — prevention for transient rs_tmp_* items Stamps each reasoning item with the issuer kind that minted it (codex_backend / xai_responses / github_responses / other:<url>) at normalize time, then drops items at replay time when the active endpoint differs from the stamp. Unstamped (legacy) items pass through for backwards compatibility. Cherry-picked from @chaconne67's PR NousResearch#31629. Conflict against current main (NousResearch#33035's replay_encrypted_reasoning parameter) resolved as 'keep both' — the two guards compose: replay_encrypted_reasoning=False is the session-wide kill switch, current_issuer_kind is the per-item filter that runs only when replay is still enabled. #AI commit#
gweeteve
pushed a commit
to gweeteve/hermes-agent
that referenced
this pull request
Jun 2, 2026
reasoning.encrypted_content is sealed to the Responses endpoint that minted it. When a session switches model providers mid-conversation — say the user runs /model gpt-5.5 after several turns on grok-4.3, or vice versa — the persisted codex_reasoning_items carry blobs the new endpoint cannot decrypt, and every subsequent turn fails with HTTP 400 invalid_encrypted_content. This is the cross-issuer prevention layer. Pairs with: * PR NousResearch#33035 — runtime recovery when the HTTP 400 fires anyway * PR NousResearch#33146 — prevention for transient rs_tmp_* items Stamps each reasoning item with the issuer kind that minted it (codex_backend / xai_responses / github_responses / other:<url>) at normalize time, then drops items at replay time when the active endpoint differs from the stamp. Unstamped (legacy) items pass through for backwards compatibility. Cherry-picked from @chaconne67's PR NousResearch#31629. Conflict against current main (NousResearch#33035's replay_encrypted_reasoning parameter) resolved as 'keep both' — the two guards compose: replay_encrypted_reasoning=False is the session-wide kill switch, current_issuer_kind is the per-item filter that runs only when replay is still enabled.
alt-glitch
pushed a commit
that referenced
this pull request
Jun 14, 2026
reasoning.encrypted_content is sealed to the Responses endpoint that minted it. When a session switches model providers mid-conversation — say the user runs /model gpt-5.5 after several turns on grok-4.3, or vice versa — the persisted codex_reasoning_items carry blobs the new endpoint cannot decrypt, and every subsequent turn fails with HTTP 400 invalid_encrypted_content. This is the cross-issuer prevention layer. Pairs with: * PR #33035 — runtime recovery when the HTTP 400 fires anyway * PR #33146 — prevention for transient rs_tmp_* items Stamps each reasoning item with the issuer kind that minted it (codex_backend / xai_responses / github_responses / other:<url>) at normalize time, then drops items at replay time when the active endpoint differs from the stamp. Unstamped (legacy) items pass through for backwards compatibility. Cherry-picked from @chaconne67's PR #31629. Conflict against current main (#33035's replay_encrypted_reasoning parameter) resolved as 'keep both' — the two guards compose: replay_encrypted_reasoning=False is the session-wide kill switch, current_issuer_kind is the per-item filter that runs only when replay is still enabled.
T02200059
pushed a commit
to T02200059/hermes-agent
that referenced
this pull request
Jun 18, 2026
reasoning.encrypted_content is sealed to the Responses endpoint that minted it. When a session switches model providers mid-conversation — say the user runs /model gpt-5.5 after several turns on grok-4.3, or vice versa — the persisted codex_reasoning_items carry blobs the new endpoint cannot decrypt, and every subsequent turn fails with HTTP 400 invalid_encrypted_content. This is the cross-issuer prevention layer. Pairs with: * PR NousResearch#33035 — runtime recovery when the HTTP 400 fires anyway * PR NousResearch#33146 — prevention for transient rs_tmp_* items Stamps each reasoning item with the issuer kind that minted it (codex_backend / xai_responses / github_responses / other:<url>) at normalize time, then drops items at replay time when the active endpoint differs from the stamp. Unstamped (legacy) items pass through for backwards compatibility. Cherry-picked from @chaconne67's PR NousResearch#31629. Conflict against current main (NousResearch#33035's replay_encrypted_reasoning parameter) resolved as 'keep both' — the two guards compose: replay_encrypted_reasoning=False is the session-wide kill switch, current_issuer_kind is the per-item filter that runs only when replay is still enabled.
waefrebeorn
pushed a commit
to waefrebeorn/slermes
that referenced
this pull request
Jul 2, 2026
reasoning.encrypted_content is sealed to the Responses endpoint that minted it. When a session switches model providers mid-conversation — say the user runs /model gpt-5.5 after several turns on grok-4.3, or vice versa — the persisted codex_reasoning_items carry blobs the new endpoint cannot decrypt, and every subsequent turn fails with HTTP 400 invalid_encrypted_content. This is the cross-issuer prevention layer. Pairs with: * PR NousResearch#33035 — runtime recovery when the HTTP 400 fires anyway * PR NousResearch#33146 — prevention for transient rs_tmp_* items Stamps each reasoning item with the issuer kind that minted it (codex_backend / xai_responses / github_responses / other:<url>) at normalize time, then drops items at replay time when the active endpoint differs from the stamp. Unstamped (legacy) items pass through for backwards compatibility. Cherry-picked from @chaconne67's PR NousResearch#31629. Conflict against current main (NousResearch#33035's replay_encrypted_reasoning parameter) resolved as 'keep both' — the two guards compose: replay_encrypted_reasoning=False is the session-wide kill switch, current_issuer_kind is the per-item filter that runs only when replay is still enabled.
santhreal
pushed a commit
to santhreal/hermes-agent
that referenced
this pull request
Jul 13, 2026
reasoning.encrypted_content is sealed to the Responses endpoint that minted it. When a session switches model providers mid-conversation — say the user runs /model gpt-5.5 after several turns on grok-4.3, or vice versa — the persisted codex_reasoning_items carry blobs the new endpoint cannot decrypt, and every subsequent turn fails with HTTP 400 invalid_encrypted_content. This is the cross-issuer prevention layer. Pairs with: * PR NousResearch#33035 — runtime recovery when the HTTP 400 fires anyway * PR NousResearch#33146 — prevention for transient rs_tmp_* items Stamps each reasoning item with the issuer kind that minted it (codex_backend / xai_responses / github_responses / other:<url>) at normalize time, then drops items at replay time when the active endpoint differs from the stamp. Unstamped (legacy) items pass through for backwards compatibility. Cherry-picked from @chaconne67's PR NousResearch#31629. Conflict against current main (NousResearch#33035's replay_encrypted_reasoning parameter) resolved as 'keep both' — the two guards compose: replay_encrypted_reasoning=False is the session-wide kill switch, current_issuer_kind is the per-item filter that runs only when replay is still enabled.
donbowman
pushed a commit
to donbowman/hermes-agent
that referenced
this pull request
Jul 13, 2026
reasoning.encrypted_content is sealed to the Responses endpoint that minted it. When a session switches model providers mid-conversation — say the user runs /model gpt-5.5 after several turns on grok-4.3, or vice versa — the persisted codex_reasoning_items carry blobs the new endpoint cannot decrypt, and every subsequent turn fails with HTTP 400 invalid_encrypted_content. This is the cross-issuer prevention layer. Pairs with: * PR NousResearch#33035 — runtime recovery when the HTTP 400 fires anyway * PR NousResearch#33146 — prevention for transient rs_tmp_* items Stamps each reasoning item with the issuer kind that minted it (codex_backend / xai_responses / github_responses / other:<url>) at normalize time, then drops items at replay time when the active endpoint differs from the stamp. Unstamped (legacy) items pass through for backwards compatibility. Cherry-picked from @chaconne67's PR NousResearch#31629. Conflict against current main (NousResearch#33035's replay_encrypted_reasoning parameter) resolved as 'keep both' — the two guards compose: replay_encrypted_reasoning=False is the session-wide kill switch, current_issuer_kind is the per-item filter that runs only when replay is still enabled.
Gravezzz
pushed a commit
to Gravezzz/hermes-agent
that referenced
this pull request
Jul 21, 2026
reasoning.encrypted_content is sealed to the Responses endpoint that minted it. When a session switches model providers mid-conversation — say the user runs /model gpt-5.5 after several turns on grok-4.3, or vice versa — the persisted codex_reasoning_items carry blobs the new endpoint cannot decrypt, and every subsequent turn fails with HTTP 400 invalid_encrypted_content. This is the cross-issuer prevention layer. Pairs with: * PR NousResearch#33035 — runtime recovery when the HTTP 400 fires anyway * PR NousResearch#33146 — prevention for transient rs_tmp_* items Stamps each reasoning item with the issuer kind that minted it (codex_backend / xai_responses / github_responses / other:<url>) at normalize time, then drops items at replay time when the active endpoint differs from the stamp. Unstamped (legacy) items pass through for backwards compatibility. Cherry-picked from @chaconne67's PR NousResearch#31629. Conflict against current main (NousResearch#33035's replay_encrypted_reasoning parameter) resolved as 'keep both' — the two guards compose: replay_encrypted_reasoning=False is the session-wide kill switch, current_issuer_kind is the per-item filter that runs only when replay is still enabled.
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
3 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.
Salvage of #11038 (@kpadilha) onto current main — prevention layer for encrypted-content corruption.
Summary
Stops Hermes from persisting transient Codex reasoning replay items whose ids use the
rs_tmp_*prefix. These temporary blobs were getting written into conversation history and replayed on subsequent turns, where they surface as HTTP 400invalid_encrypted_contentfailures.This is a prevention fix — complements the existing recovery path in PR #33035 (
_disable_codex_reasoning_replay+ retry). With both in place: prevent at write time (this PR), recover at read time (#33035).Changes
agent/codex_responses_adapter.py:_normalize_codex_responseskips reasoning items whose id starts withrs_tmp_before they get persisted. Addssaw_reasoning_itemtracking so the finish_reason heuristic still treats reasoning-only turns as "incomplete" even when the only reasoning item was a transient one that got dropped.tests/agent/test_codex_responses_adapter.py: new test file with two assertions — dropsrs_tmp_*items, keeps stablers_*items.Validation
tests/agent/test_codex_responses_adapter.py+tests/run_agent/test_run_agent_codex_responses.py→ 70/70 passingAttribution
Cherry-picked from @kpadilha's #11038. AUTHOR_MAP updated in follow-up commit. Original PR was opened May 13; cherry-pick onto current main was clean (no conflicts).
Infographic