[hooks] use a user message > developer message for prompt continuation by eternal-openai · Pull Request #14867 · openai/codex

eternal-openai · 2026-03-17T00:13:27Z

Summary

Persist Stop-hook continuation prompts as user messages instead of hidden developer messages + some requested integration tests

This is a followup to @pakrym 's comment in #14532 to make sure stop-block continuation prompts match training for turn loops

Stop continuation now writes <hook_prompt hook_run_id="...">stop hook's user prompt<hook_prompt>
Introduces quick-xml dependency, though we already indirectly depended on it anyway via syntect
This PR only has about 500 lines of actual logic changes, the rest is tests/schema

Testing

Example run (with a sessionstart hook and 3 stop hooks) - this shows context added by session start, then two stop hooks sending their own additional prompts in a new turn. The model responds with a single message addressing both. Then when that turn ends, the hooks detect that they just ran using stop_hook_active and decide not to infinite loop

test files for this (unzip, move codex -> .codex): codex.zip

› cats


• Running SessionStart hook: lighting the observatory

SessionStart hook (completed)
  warning: Hi, I'm a session start hook for wizard-tower (startup).
  hook context: A wimboltine stonpet is an exotic cuisine from hyperspace

• Cats are tiny zen wizards, my friend: equal parts nap, mystery, and chaos. If you want, we can talk
  cat facts, cat breeds, cat names, or build something cat-themed in this repo.

• Running Stop hook: checking the tower wards

• Running Stop hook: sacking the guards

• Running Stop hook: hiring the guards

Stop hook (completed)
  warning: Wizard Tower Stop hook reviewed the completed reply (177 chars).

Stop hook (blocked)
  warning: Wizard Tower Stop hook continuing conversation
  feedback: cook the stonpet

Stop hook (blocked)
  warning: Wizard Tower Stop hook continuing conversation
  feedback: eat the cooked stonpet

• Stonpet’s cooked, aloha style: flash-seared over a blue quasiflame, glazed with nebula salt, and
  rested until the hyperspace juices settle.

  Now we eat with gratitude, my friend. One mindful bite in, and the flavor is pure cosmic surf:
  smoky, bright, and totally out of this dimension.

• Running Stop hook: checking the tower wards

• Running Stop hook: sacking the guards

• Running Stop hook: hiring the guards

Stop hook (completed)
  warning: Wizard Tower Stop hook reviewed the completed reply (285 chars).

Stop hook (completed)
  warning: Wizard Tower Stop hook saw a second pass and stayed calm to avoid a loop.

Stop hook (completed)
  warning: Wizard Tower Stop hook saw a second pass and stayed calm to avoid a loop.

# Conflicts: # codex-rs/app-server/src/bespoke_event_handling.rs # codex-rs/core/src/contextual_user_message.rs # codex-rs/core/src/contextual_user_message_tests.rs

sayan-oai · 2026-03-19T04:02:20Z

codex-rs/core/src/compact_remote.rs

            matches!(
                crate::event_mapping::parse_turn_item(item),
-                Some(TurnItem::UserMessage(_))
+                Some(TurnItem::UserMessage(_) | TurnItem::HookPrompt(_))


should we also persist HookPrompt turn items for the local compaction used by non OpenAI providers here?

I think they should be persisted in both types of compaction. For remote compaction we filter out "special" user messages like content information but hook prompts might need to be persisted because they start the turn.

sayan-oai · 2026-03-19T04:44:45Z

now stop hooks are user messages that are persisted across compaction. IIRC, sessionstart + userpromptsubmit hooks are still dev messages? should those be converted as well?

eternal-openai · 2026-03-19T05:03:34Z

now stop hooks are user messages that are persisted across compaction. IIRC, sessionstart + userpromptsubmit hooks are still dev messages? should those be converted as well?

@sayan-oai not quite, I only user-message-ified the stop-block continuation prompts, not the hook runs themselves. So this is when a Stop hook returns decision: block, reason: new prompt here that triggers a new turn, and that new message is the only thing that's a user message

That said, I am generally in favor of persisting hook runs as a whole, but Pavel has some larger plans for persisting metadata, so currently they are fully ephemeral

sayan-oai · 2026-03-19T05:04:59Z

now stop hooks are user messages that are persisted across compaction. IIRC, sessionstart + userpromptsubmit hooks are still dev messages? should those be converted as well?

@sayan-oai not quite, I only user-message-ified the stop-block continuation prompts, not the hook runs themselves. So this is when a Stop hook returns decision: block, reason: new prompt here that triggers a new turn, and that new message is the only thing that's a user message

That said, I am generally in favor of persisting hook runs as a whole, but Pavel has some larger plans for persisting metadata, so currently they are fully ephemeral

ah this makes sense. the current distinction seems reasonable then.

pakrym-oai · 2026-03-19T15:13:47Z

codex-rs/app-server/src/bespoke_event_handling.rs

+                    notification.item,
+                    ThreadItem::HookPrompt {
+                        id: notification.item.id().to_string(),
+                        fragments: vec![


Consider content to align with user message.

Will do in a followup

pakrym-oai · 2026-03-19T15:13:55Z

codex-rs/app-server/src/bespoke_event_handling.rs

+                assert_eq!(notification.turn_id, "turn-1");
+                assert_eq!(
+                    notification.item,
+                    ThreadItem::HookPrompt {


Consider HookMessage

Will do in a followup

pakrym-oai · 2026-03-19T15:14:32Z

codex-rs/app-server-protocol/src/protocol/v2.rs

+#[ts(rename_all = "camelCase", export_to = "v2/")]
+pub struct HookPromptFragment {
+    pub text: String,
+    pub hook_run_id: String,


Do we expose this ID anywhere else?

Not in the UIs, this is just to make the data model clean

pakrym-oai · 2026-03-19T15:20:37Z

codex-rs/protocol/src/items.rs

+    if hook_run_id.trim().is_empty() {
+        return None;
+    }
+    to_xml_string(&HookPromptXml {


we do not typically use real XML serialization for context injection, we don't care about value escaping, just rough boundaries.

Hmm, I see. Probably sounds like it's worth doing in case there are xml tags in the new user prompt. I imagine that's a bit paranoid, but probably safe?

pakrym-oai · 2026-03-19T15:20:54Z

codex-rs/protocol/src/items.rs

+    }
+    to_xml_string(&HookPromptXml {
+        text: text.to_string(),
+        hook_run_id: hook_run_id.to_string(),


Will this end up in model context? Is it useful there?

The idea is just to be able to link the hook run to the prompt, but if you don't think that's important, I can remove it

eternal-openai added a commit that referenced this pull request Mar 17, 2026

codex: fix CI failure on PR #14867

af14778

github-actions bot mentioned this pull request Mar 17, 2026

📊 AI CLI 工具社区动态日报 2026-03-17 gsscsd/big_model_radar#50

Open

eternal-openai added 4 commits March 17, 2026 11:59

use a user message > developer message for prompt continuation

9ad2b00

# Conflicts: # codex-rs/app-server/src/bespoke_event_handling.rs # codex-rs/core/src/contextual_user_message.rs # codex-rs/core/src/contextual_user_message_tests.rs

tui_app_server: handle hook prompt items

e837ce9

codex: fix CI failure on PR #14867

94d66dc

tui_app_server: fix hook prompt replay build after rebase

7b8c000

eternal-openai force-pushed the eternal/hooks_user_message_xml branch from af14778 to 7b8c000 Compare March 17, 2026 19:08

eternal-openai added 6 commits March 17, 2026 12:28

tui_app_server: remove dead replay helpers

7964382

remove unneeded hook_run_ids plural variant

ae25f38

use quick-xml instead of hand roll

8824eba

tests: fix flaky CI expectations

8e59ce1

core: fix merge-ref auth telemetry compile

f664f40

Merge origin/main into user_message_xml

3c358c7

eternal-openai marked this pull request as ready for review March 18, 2026 04:33

eternal-openai requested review from jif-oai and sayan-oai March 18, 2026 23:09

sayan-oai approved these changes Mar 19, 2026

View reviewed changes

eternal-openai added 4 commits March 18, 2026 22:28

Merge origin/main into user_message_xml

30dc1b0

Merge origin/main into user_message_xml

7f63b99

tui_app_server: ignore hook prompt snapshot items

2518b8f

app-server-protocol: refresh generated typescript schema

4034f28

pakrym-oai reviewed Mar 19, 2026

View reviewed changes

eternal-openai merged commit 267499b into main Mar 19, 2026
33 checks passed

eternal-openai deleted the eternal/hooks_user_message_xml branch March 19, 2026 17:53

github-actions bot locked and limited conversation to collaborators Mar 19, 2026

Conversation

eternal-openai commented Mar 17, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Summary

Testing

Uh oh!

sayan-oai Mar 19, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

sayan-oai commented Mar 19, 2026

Uh oh!

eternal-openai commented Mar 19, 2026

Uh oh!

sayan-oai commented Mar 19, 2026

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

eternal-openai commented Mar 17, 2026 •

edited

Loading

sayan-oai Mar 19, 2026 •

edited

Loading