fix: agent instructions were not being included when ~/.codex/instructions.md was empty (#908)

I had seen issues where `codex-rs` would not always write files without
me pressuring it to do so, and between that and the report of
https://github.com/openai/codex/issues/900, I decided to look into this
further. I found two serious issues with agent instructions:

(1) We were only sending agent instructions on the first turn, but
looking at the TypeScript code, we should be sending them on every turn.

(2) There was a serious issue where the agent instructions were
frequently lost:

* The TypeScript CLI appears to keep writing `~/.codex/instructions.md`:
55142e3e6c/codex-cli/src/utils/config.ts (L586)
* If `instructions.md` is present, the Rust CLI uses the contents of it
INSTEAD OF the default prompt, even if `instructions.md` is empty:
55142e3e6c/codex-rs/core/src/config.rs (L202-L203)

The combination of these two things means that I have been using
`codex-rs` without these key instructions:
https://github.com/openai/codex/blob/main/codex-rs/core/prompt.md

Looking at the TypeScript code, it appears we should be concatenating
these three items every time (if they exist):

* `prompt.md`
* `~/.codex/instructions.md`
* nearest `AGENTS.md`

This PR fixes things so that:

* `Config.instructions` is `None` if `instructions.md` is empty
* `Payload.instructions` is now `&'a str` instead of `Option<&'a
String>` because we should always have _something_ to send
* `Prompt` now has a `get_full_instructions()` helper that returns a
`Cow<str>` that will always include the agent instructions first.

codex-rs/core/src/chat_completions.rs

@@ -38,9 +38,8 @@ pub(crate) async fn stream_chat_completions(
     // Build messages array
     let mut messages = Vec::<serde_json::Value>::new();
 
-    if let Some(instr) = &prompt.instructions {
-        messages.push(json!({"role": "system", "content": instr}));
-    }
+    let full_instructions = prompt.get_full_instructions();
+    messages.push(json!({"role": "system", "content": full_instructions}));
 
     for item in &prompt.input {
         if let ResponseItem::Message { role, content } = item {

codex-rs/core/src/client.rs

@@ -166,9 +166,10 @@ impl ModelClient {
 
         debug!("tools_json: {}", serde_json::to_string_pretty(&tools_json)?);
 
+        let full_instructions = prompt.get_full_instructions();
         let payload = Payload {
             model: &self.model,
-            instructions: prompt.instructions.as_ref(),
+            instructions: &full_instructions,
             input: &prompt.input,
             tools: &tools_json,
             tool_choice: "auto",

codex-rs/core/src/client_common.rs

@@ -2,12 +2,17 @@ use crate::error::Result;
 use crate::models::ResponseItem;
 use futures::Stream;
 use serde::Serialize;
+use std::borrow::Cow;
 use std::collections::HashMap;
 use std::pin::Pin;
 use std::task::Context;
 use std::task::Poll;
 use tokio::sync::mpsc;
 
+/// The `instructions` field in the payload sent to a model should always start
+/// with this content.
+const BASE_INSTRUCTIONS: &str = include_str!("../prompt.md");
+
 /// API request payload for a single model turn.
 #[derive(Default, Debug, Clone)]
 pub struct Prompt {
@@ -15,7 +20,8 @@ pub struct Prompt {
     pub input: Vec<ResponseItem>,
     /// Optional previous response ID (when storage is enabled).
     pub prev_id: Option<String>,
-    /// Optional initial instructions (only sent on first turn).
+    /// Optional instructions from the user to amend to the built-in agent
+    /// instructions.
     pub instructions: Option<String>,
     /// Whether to store response on server side (disable_response_storage = !store).
     pub store: bool,
@@ -26,6 +32,18 @@ pub struct Prompt {
     pub extra_tools: HashMap<String, mcp_types::Tool>,
 }
 
+impl Prompt {
+    pub(crate) fn get_full_instructions(&self) -> Cow<str> {
+        match &self.instructions {
+            Some(instructions) => {
+                let instructions = format!("{BASE_INSTRUCTIONS}\n{instructions}");
+                Cow::Owned(instructions)
+            }
+            None => Cow::Borrowed(BASE_INSTRUCTIONS),
+        }
+    }
+}
+
 #[derive(Debug)]
 pub enum ResponseEvent {
     OutputItemDone(ResponseItem),
@@ -54,8 +72,7 @@ pub(crate) enum Summary {
 #[derive(Debug, Serialize)]
 pub(crate) struct Payload<'a> {
     pub(crate) model: &'a str,
-    #[serde(skip_serializing_if = "Option::is_none")]
-    pub(crate) instructions: Option<&'a String>,
+    pub(crate) instructions: &'a str,
     // TODO(mbolin): ResponseItem::Other should not be serialized. Currently,
     // we code defensively to avoid this case, but perhaps we should use a
     // separate enum for serialization.

codex-rs/core/src/config.rs

@@ -10,11 +10,6 @@ use serde::Deserialize;
 use std::collections::HashMap;
 use std::path::PathBuf;
 
-/// Embedded fallback instructions that mirror the TypeScript CLI’s default
-/// system prompt. These are compiled into the binary so a clean install behaves
-/// correctly even if the user has not created `~/.codex/instructions.md`.
-const EMBEDDED_INSTRUCTIONS: &str = include_str!("../prompt.md");
-
 /// Maximum number of bytes of the documentation that will be embedded. Larger
 /// files are *silently truncated* to this size so we do not take up too much of
 /// the context window.
@@ -42,7 +37,7 @@ pub struct Config {
     /// who have opted into Zero Data Retention (ZDR).
     pub disable_response_storage: bool,
 
-    /// System instructions.
+    /// User-provided instructions from instructions.md.
     pub instructions: Option<String>,
 
     /// Optional external notifier command. When set, Codex will spawn this
@@ -198,9 +193,7 @@ impl Config {
         cfg: ConfigToml,
         overrides: ConfigOverrides,
     ) -> std::io::Result<Self> {
-        // Instructions: user-provided instructions.md > embedded default.
-        let instructions =
-            Self::load_instructions().or_else(|| Some(EMBEDDED_INSTRUCTIONS.to_string()));
+        let instructions = Self::load_instructions();
 
         // Destructure ConfigOverrides fully to ensure all overrides are applied.
         let ConfigOverrides {
@@ -289,7 +282,14 @@ impl Config {
     fn load_instructions() -> Option<String> {
         let mut p = codex_dir().ok()?;
         p.push("instructions.md");
-        std::fs::read_to_string(&p).ok()
+        std::fs::read_to_string(&p).ok().and_then(|s| {
+            let s = s.trim();
+            if s.is_empty() {
+                None
+            } else {
+                Some(s.to_string())
+            }
+        })
     }
 
     /// Meant to be used exclusively for tests: `load_with_overrides()` should