hive-forge: require HIVE_FORGE_URL, give agent login shells the forge env

hive-forge/README.md

@@ -7,7 +7,7 @@ consistent error handling, exit codes, and JSON shapes. This is the
 **only supported path** to the forge from inside an agent container;
 never `curl` it directly.
 
-Reads credentials from the environment (`HIVE_FORGE_URL`,
+Reads credentials from the environment (`HIVE_FORGE_URL`, required — no default,
 `HYPERHIVE_STATE_DIR`); `-f/--forge <label>` retargets a
 swarm-UI-linked external forge account instead. The active repo
 resolves `-r/--repo` > the `origin` remote of the cwd's git checkout >

hive-forge/src/client.rs

@@ -17,9 +17,6 @@ use reqwest::header::{
 use serde::{Deserialize, Serialize};
 use serde_json::Value;
 
-/// Default Forgejo URL when `HIVE_FORGE_URL` is unset.
-const DEFAULT_URL: &str = "http://localhost:3000";
-
 /// Bound on reaching the forge.
 const HTTP_CONNECT_TIMEOUT: std::time::Duration = std::time::Duration::from_secs(5);
 /// Whole-request budget for a small JSON call (an `/api/v1` GET, or the
@@ -395,9 +392,22 @@ pub fn index(n: u64) -> Result<i64> {
     i64::try_from(n).with_context(|| format!("number {n} out of range"))
 }
 
+/// The internal forge's base URL from the `HIVE_FORGE_URL` value, which has
+/// no default: inside an agent container `localhost` is the agent itself.
+fn forge_url(value: Option<String>) -> Result<String> {
+    match value {
+        Some(url) if !url.trim().is_empty() => Ok(url),
+        _ => bail!(
+            "hive-forge: HIVE_FORGE_URL is unset or empty — the forge URL comes from the \
+             hive config (services.hyperhive.swarm.forge.domain) and is set in every \
+             agent container; export it, or use -f/--forge <label> for a linked external forge"
+        ),
+    }
+}
+
 /// Resolve the `(base_url, token)` pair the client authenticates with.
 /// `None` (the default) resolves the internal forge exactly as before:
-/// `HIVE_FORGE_URL` (default [`DEFAULT_URL`]) + `read_token()`.
+/// `HIVE_FORGE_URL` (required, see [`forge_url`]) + `read_token()`.
 /// `Some(label)` (from `-f/--forge <label>`) instead resolves an
 /// external forge account linked in the swarm UI: the token comes from
 /// `${HYPERHIVE_STATE_DIR}/forge-<label>-token` and the base URL from
@@ -411,7 +421,7 @@ pub fn index(n: u64) -> Result<i64> {
 /// nonsense/traversing path and surfacing a confusing file error later.
 pub(crate) fn resolve_credentials(forge_label: Option<&str>) -> Result<(String, String)> {
     let Some(label) = forge_label else {
-        let base = std::env::var("HIVE_FORGE_URL").unwrap_or_else(|_| DEFAULT_URL.to_owned());
+        let base = forge_url(std::env::var("HIVE_FORGE_URL").ok())?;
         let token = read_token().context("read forge-token")?;
         return Ok((base, token));
     };
@@ -595,7 +605,26 @@ fn check_status(resp: Response, op: &str) -> Result<Response> {
 
 #[cfg(test)]
 mod tests {
-    use super::{WebStatusError, disposition_filename, index, owner_repo_from_git_url, split_repo};
+    use super::{
+        WebStatusError, disposition_filename, forge_url, index, owner_repo_from_git_url, split_repo,
+    };
+
+    #[test]
+    fn forge_url_is_the_value_as_is() {
+        assert_eq!(
+            forge_url(Some("http://forge.t.local".to_owned())).unwrap(),
+            "http://forge.t.local"
+        );
+    }
+
+    #[test]
+    fn forge_url_unset_or_blank_names_the_variable() {
+        for v in [None, Some(String::new()), Some("  ".to_owned())] {
+            let err = forge_url(v).unwrap_err().to_string();
+            assert!(err.contains("HIVE_FORGE_URL is unset or empty"), "{err}");
+            assert!(err.contains("hive config"), "{err}");
+        }
+    }
 
     #[test]
     fn disposition_filename_reads_forgejos_real_header() {

hive-forge/src/main.rs

@@ -1,7 +1,7 @@
 //! `hive-forge` — typed CLI wrapper around the in-cluster Forgejo's
 //! REST API. Reads credentials from the environment:
 //!
-//!   `HIVE_FORGE_URL`      — base URL, e.g. `http://localhost:3000`
+//!   `HIVE_FORGE_URL`      — base URL (required), e.g. `http://forge.example.org`
 //!   `HYPERHIVE_STATE_DIR` — state dir; `forge-token` lives here
 //!   `HIVE_FORGE_DEADLINE` — optional; whole seconds one invocation may run
 //!                           before it exits with code 75 (default 900)

hive-forge/tests/forge_url_required.rs

@@ -0,0 +1,57 @@
+//! `HIVE_FORGE_URL` has no default: unset or empty is a clear error naming
+//! the variable, and a set value is the URL the client dials.
+
+use std::process::{Command, Output};
+
+const UNSET_MSG: &str = "HIVE_FORGE_URL is unset or empty";
+
+fn scratch_state_dir(tag: &str) -> std::path::PathBuf {
+    let ts = std::time::SystemTime::now()
+        .duration_since(std::time::UNIX_EPOCH)
+        .map_or(0, |d| d.as_nanos());
+    let dir = std::env::temp_dir().join(format!("hive-forge-url-required-{tag}-{ts}"));
+    std::fs::create_dir_all(&dir).expect("create scratch state dir");
+    std::fs::write(dir.join("forge-token"), "test-token\n").expect("seed forge-token");
+    dir
+}
+
+fn run(tag: &str, url: Option<&str>) -> Output {
+    let mut cmd = Command::new(env!("CARGO_BIN_EXE_hive-forge"));
+    cmd.args(["-r", "o/r", "pr", "edit", "7", "--title", "t"])
+        .env("HYPERHIVE_STATE_DIR", scratch_state_dir(tag))
+        .env_remove("HIVE_FORGE_TOKEN_FILE")
+        .env_remove("HIVE_LABEL");
+    match url {
+        Some(u) => cmd.env("HIVE_FORGE_URL", u),
+        None => cmd.env_remove("HIVE_FORGE_URL"),
+    };
+    cmd.output().expect("run hive-forge")
+}
+
+#[test]
+fn unset_url_is_a_clear_error() {
+    let out = run("unset", None);
+    let stderr = String::from_utf8_lossy(&out.stderr);
+    assert!(!out.status.success());
+    assert!(stderr.contains(UNSET_MSG), "stderr: {stderr}");
+    assert!(!stderr.contains("localhost"), "stderr: {stderr}");
+}
+
+#[test]
+fn empty_url_is_a_clear_error() {
+    let out = run("empty", Some(""));
+    let stderr = String::from_utf8_lossy(&out.stderr);
+    assert!(!out.status.success());
+    assert!(stderr.contains(UNSET_MSG), "stderr: {stderr}");
+}
+
+/// Presence control: a set value passes the URL check and the client dials it
+/// (nothing listens on port 1, so the failure is a connection error).
+#[test]
+fn set_url_is_used_as_is() {
+    let out = run("set", Some("http://127.0.0.1:1"));
+    let stderr = String::from_utf8_lossy(&out.stderr);
+    assert!(!out.status.success());
+    assert!(!stderr.contains(UNSET_MSG), "stderr: {stderr}");
+    assert!(stderr.contains("127.0.0.1"), "stderr: {stderr}");
+}

hive-meta-flake/src/lib.rs

@@ -208,9 +208,9 @@ pub fn render_flake(
 ) -> Result<String> {
     let mut out = String::new();
     push_inputs(&mut out, hyperhive_url, agents)?;
-    push_mk_agent_open(&mut out);
+    push_mk_agent_open(&mut out, vars);
     push_agent_service_options(&mut out, vars, github_disabled);
-    push_agent_environment(&mut out, vars);
+    push_agent_environment(&mut out);
     push_agent_entries(&mut out, agents);
     Ok(out)
 }
@@ -267,8 +267,10 @@ fn push_inputs(out: &mut String, hyperhive_url: &str, agents: &[AgentSpec]) -> R
 }
 
 /// Closes `inputs`, opens `outputs` and the `mkAgent` function up to the
-/// start of the per-agent module.
-fn push_mk_agent_open(out: &mut String) {
+/// start of the per-agent module. Binds `agentEnv`, the one attrset the
+/// container-wide, login-shell and harness-service environments all take.
+fn push_mk_agent_open(out: &mut String, vars: &[(&'static str, String)]) {
+    use std::fmt::Write as _;
     out.push_str(
         "  };\n  outputs =\n    { self, hyperhive, ... }@inputs:\n    let\n      mkAgent = { name, port, archetype ? null }:\n",
     );
@@ -286,12 +288,27 @@ fn push_mk_agent_open(out: &mut String) {
               ).hyperhiveArchetypes.${archetype.name}.module
                 or (throw "agent ${name}: this hive's ${archetype.source} input has no archetype ${archetype.name}; update that input");
           service = "hive-agent";
+          # One attrset for `environment.variables` (login shells),
+          # `systemd.globalEnvironment` (every unit) and the harness service,
+          # so the three cannot disagree. The forwarded vars come from the host.
+          agentEnv = {
+            HIVE_LABEL = name;
+            HYPERHIVE_STATE_DIR = "/agents/${name}/state";
+            HYPERHIVE_HARNESS_DIR = "/agents/${name}/harness";
+"#,
+    );
+    for (var, val) in vars {
+        let escaped = val.replace('\\', "\\\\").replace('"', "\\\"");
+        let _ = writeln!(out, "            {var} = \"{escaped}\";");
+    }
+    out.push_str(
+        r"          };
         in
         base.extendModules {
           modules = [
             agentModule
             {
-"#,
+",
     );
 }
 
@@ -339,8 +356,7 @@ fn push_agent_service_options(
 
 /// The rest of the per-agent module: user, hostname, git identity, and the
 /// container-wide and harness-service environments.
-fn push_agent_environment(out: &mut String, vars: &[(&'static str, String)]) {
-    use std::fmt::Write as _;
+fn push_agent_environment(out: &mut String) {
     out.push_str(
         r#"              # The harness service inside the container runs as a
               # non-root unix user named after the agent (`damocles`,
@@ -362,58 +378,15 @@ fn push_agent_environment(out: &mut String, vars: &[(&'static str, String)]) {
                 name = name;
                 email = "${name}@hyperhive.local";
               };
-              # Container-wide env: every service + co-process daemon can
-              # resolve the agent's durable state dir without hard-coding it.
-              # `environment.variables` only writes /etc/environment (login
-              # shells); `systemd.globalEnvironment` is the analogue for
-              # systemd units so forge-avatar-sync /
-              # hive-matrix-daemon etc. can read `$HYPERHIVE_STATE_DIR`
-              # without each service having to redeclare it.
-              environment.variables = {
-                HIVE_LABEL = name;
-                HYPERHIVE_STATE_DIR = "/agents/${name}/state";
-                HYPERHIVE_HARNESS_DIR = "/agents/${name}/harness";
-              };
-              systemd.globalEnvironment = {
-                HIVE_LABEL = name;
-                HYPERHIVE_STATE_DIR = "/agents/${name}/state";
-                HYPERHIVE_HARNESS_DIR = "/agents/${name}/harness";
-"#,
-    );
-    // Forwarded vars (HIVE_FORGE_URL etc.) also go into globalEnvironment,
-    // not just the harness service env below, so EVERY service + shell in
-    // the container inherits them — crucially the bash-task runner (where
-    // `hive-forge` + `git` actually run), plus the matrix daemon, forge-avatar-sync
-    // and interactive shells. Scoped to the harness service alone they were
-    // invisible to bash tasks: harmless in shared netns (the localhost
-    // default works) but broken under isolation, where the in-cluster
-    // `forge.<domain>` URL is the only reachable path.
-    for (var, val) in vars {
-        let escaped = val.replace('\\', "\\\\").replace('"', "\\\"");
-        let _ = writeln!(out, "                {var} = \"{escaped}\";");
-    }
-    out.push_str(
-        r"              };
-              systemd.services.${service}.environment = {
+              # Container-wide env, all from `agentEnv`: every unit and daemon
+              # (the bash-task runner where `hive-forge` and `git` run included)
+              # resolves the state dir and the forwarded service URLs without
+              # redeclaring them. `environment.variables` is what login shells
+              # read; `systemd.globalEnvironment` is the analogue for units.
+              environment.variables = agentEnv;
+              systemd.globalEnvironment = agentEnv;
+              systemd.services.${service}.environment = agentEnv // {
                 HIVE_PORT = toString port;
-                HIVE_LABEL = name;
-",
-    );
-    // Forwarded env vars, as the caller passed them. We copy whatever's set into
-    // each sub-agent's harness service env so the in-container surfaces
-    // (`identity.rs`, `forge_notify`) see a consistent view across the
-    // whole hive. Absent host-side env (option not set) → skip emission
-    // → in-container accessors fall back to None / defaults gracefully.
-    //
-    // - HIVE_FORGE_URL: agents poll this for Forgejo notifications.
-    // - HYPERHIVE_SWARM_NAME: human display name for the swarm.
-    for (var, val) in vars {
-        let escaped = val.replace('\\', "\\\\").replace('"', "\\\"");
-        let _ = writeln!(out, "                {var} = \"{escaped}\";");
-    }
-    out.push_str(
-        r#"                HYPERHIVE_STATE_DIR = "/agents/${name}/state";
-                HYPERHIVE_HARNESS_DIR = "/agents/${name}/harness";
               };
             }
             # Tells the harness the agent has no config repo of its own.
@@ -580,6 +553,36 @@ mod tests {
         );
     }
 
+    /// Login shells, every unit and the harness service take the forwarded
+    /// vars from the one `agentEnv` attrset, so none can lack a value another has.
+    #[test]
+    fn forwarded_vars_reach_login_shells_units_and_the_harness_from_one_attrset() {
+        let vars = [
+            ("HIVE_FORGE_URL", "http://forge.t.local".to_owned()),
+            ("HYPERHIVE_SWARM_NAME", "t".to_owned()),
+        ];
+        let archetype = ArchetypeRef {
+            source: "hyperhive".to_owned(),
+            name: "reviewer".to_owned(),
+        };
+        let flake = render_flake("u", &[spec("rev", Some(archetype), None)], &vars, false).unwrap();
+        assert_eq!(
+            flake
+                .matches("HIVE_FORGE_URL = \"http://forge.t.local\";")
+                .count(),
+            1,
+            "the value is written once, in agentEnv"
+        );
+        assert!(flake.contains("agentEnv = {\n            HIVE_LABEL = name;"));
+        for sink in [
+            "environment.variables = agentEnv;",
+            "systemd.globalEnvironment = agentEnv;",
+            "systemd.services.${service}.environment = agentEnv // {",
+        ] {
+            assert!(flake.contains(sink), "missing `{sink}`");
+        }
+    }
+
     #[test]
     fn a_rendered_flake_names_the_commit_each_input_is_pinned_to() {
         let flake = rendered();

hive-meta-flake/testdata/meta-flake.nix.in

@@ -21,6 +21,14 @@
               ).hyperhiveArchetypes.${archetype.name}.module
                 or (throw "agent ${name}: this hive's ${archetype.source} input has no archetype ${archetype.name}; update that input");
           service = "hive-agent";
+          # One attrset for `environment.variables` (login shells),
+          # `systemd.globalEnvironment` (every unit) and the harness service,
+          # so the three cannot disagree. The forwarded vars come from the host.
+          agentEnv = {
+            HIVE_LABEL = name;
+            HYPERHIVE_STATE_DIR = "/agents/${name}/state";
+            HYPERHIVE_HARNESS_DIR = "/agents/${name}/harness";
+          };
         in
         base.extendModules {
           modules = [
@@ -46,28 +54,15 @@
                 name = name;
                 email = "${name}@hyperhive.local";
               };
-              # Container-wide env: every service + co-process daemon can
-              # resolve the agent's durable state dir without hard-coding it.
-              # `environment.variables` only writes /etc/environment (login
-              # shells); `systemd.globalEnvironment` is the analogue for
-              # systemd units so forge-avatar-sync /
-              # hive-matrix-daemon etc. can read `$HYPERHIVE_STATE_DIR`
-              # without each service having to redeclare it.
-              environment.variables = {
-                HIVE_LABEL = name;
-                HYPERHIVE_STATE_DIR = "/agents/${name}/state";
-                HYPERHIVE_HARNESS_DIR = "/agents/${name}/harness";
-              };
-              systemd.globalEnvironment = {
-                HIVE_LABEL = name;
-                HYPERHIVE_STATE_DIR = "/agents/${name}/state";
-                HYPERHIVE_HARNESS_DIR = "/agents/${name}/harness";
-              };
-              systemd.services.${service}.environment = {
+              # Container-wide env, all from `agentEnv`: every unit and daemon
+              # (the bash-task runner where `hive-forge` and `git` run included)
+              # resolves the state dir and the forwarded service URLs without
+              # redeclaring them. `environment.variables` is what login shells
+              # read; `systemd.globalEnvironment` is the analogue for units.
+              environment.variables = agentEnv;
+              systemd.globalEnvironment = agentEnv;
+              systemd.services.${service}.environment = agentEnv // {
                 HIVE_PORT = toString port;
-                HIVE_LABEL = name;
-                HYPERHIVE_STATE_DIR = "/agents/${name}/state";
-                HYPERHIVE_HARNESS_DIR = "/agents/${name}/harness";
               };
             }
             # Tells the harness the agent has no config repo of its own.