Spec: four more Harnesses (agy, Grok Build, Muse Code, OpenCode)

CONTEXT.md

@@ -31,7 +31,7 @@ The message thirdshift starts or resumes an agent session with. It carries the *
 _Avoid_: instructions, system prompt
 
 **Harness**:
-The headless agent CLI that runs every agent session of a **Command**: Claude Code (`claude`) or Codex (`codex`). One per Command, its child Runs and every **Resume** and **Repair** included. Chosen by the command, else the **User config**, else Claude Code.
+The headless agent CLI that runs every agent session of a **Command**: Claude Code (`claude`), Codex (`codex`), Antigravity CLI (`agy`), Grok Build (`grok`), Muse Code (`muse`) or OpenCode (`opencode`). One per Command, its child Runs and every **Resume** and **Repair** included. Chosen by the command, else the **User config**, else Claude Code.
 _Avoid_: agent, provider, backend
 
 **Model**:

docs/adr/0013-every-harness-runs-unattended-and-fully-trusted.md

@@ -0,0 +1,27 @@
+# Every Harness runs unattended and fully trusted in the worktree
+
+Sessions run with no human to answer an approval prompt or a trust question. Their commits and pushes write to the main repository's shared `.git`, outside the worktree. So every **Harness** runs its sessions, and every Resume of one, with approvals off, its sandbox off, and the worktree trusted. Each Harness carries the flags and environment that do this. They are fixed, and no user setting changes them:
+
+| Harness | Flags and environment |
+|---|---|
+| Claude Code | `--permission-mode auto` |
+| Codex | `--dangerously-bypass-approvals-and-sandbox` |
+| agy | `--dangerously-skip-permissions`, never `--sandbox` |
+| Grok Build | `--always-approve`, `GROK_FOLDER_TRUST=0` |
+| Muse Code | `--yolo` |
+| OpenCode | `--auto`, `--standalone` |
+
+The worktree is the only boundary. This generalises the Codex part of ADR 0012. Sources: `docs/research/harness-candidates.md`, `docs/research/codex-headless-harness.md`.
+
+## Considered Options
+
+- **Each CLI's own sandbox.** Rejected. On stock Ubuntu (`kernel.apparmor_restrict_unprivileged_userns = 1`), bubblewrap can't start, so every sandboxed command fails under Codex and Muse. The sandboxes of agy and Grok make the main checkout's `.git` read-only, which blocks commits. A sandboxed mode for machines where it works would be a separate issue.
+- **A one-time `grok --trust` in each repository's main checkout**, done or documented by Setup. Rejected. A repository nobody had trusted would run its sessions without `AGENTS.md` or the Factory skills, and nothing would fail. A scheduled Pickup run is exactly the case where nobody would notice.
+- **Using the CLI's settings for approval and trust**, such as a user config that already sets always-approve. Rejected. A session would then behave differently on every machine. Passing the flags each time makes them explicit.
+- **OpenCode through the user's shared background service** (its default). Rejected. A SIGTERM leaves the turn and its shell command running in the service, and thirdshift doesn't control that service's version or environment. `--standalone` keeps the server in the session's process group, and costs 2–3 s per session.
+
+## Consequences
+
+- A Harness's self-update is turned off in each session's environment too (`MUSE_NO_AUTO_UPDATE=1`, `GROK_DISABLE_AUTOUPDATER=1`, `AGY_CLI_DISABLE_AUTO_UPDATE=true`, `OPENCODE_DISABLE_AUTOUPDATE=1`). An update in the middle of a Run could replace the binary a Resume needs. agy ignores `=1`; only `=true` works.
+- Each Harness reads its own instruction file and falls back to the other. thirdshift adds the fallback where a CLI lacks one: a `GEMINI.md` link for agy, and an `AGENTS.md` link for OpenCode, both to `CLAUDE.md` at the worktree root and kept out of git. Grok reads both files when both exist, and thirdshift leaves that as it is.
+- Instructions in the user's `~/.claude/` reach only the Harnesses that read them: Claude, Grok and Muse. They don't reach Codex, agy or OpenCode.

src/args.rs

@@ -307,7 +307,7 @@ fn ask_harness(harness: &mut Option<Harness>, arg: &str, name: Option<&String>)
     }
     let Some(named) = name.and_then(|name| Harness::named(name)) else {
         let given = name.map(|name| format!(", not {name}")).unwrap_or_default();
-        bail!("{arg} must be followed by {}{given}", harness::NAMES);
+        bail!("{arg} must be followed by {}{given}", harness::names());
     };
     *harness = Some(named);
     Ok(())
@@ -995,11 +995,11 @@ mod tests {
         for (args, error) in [
             (
                 vec![URL, "harness"],
-                "harness must be followed by claude or codex",
+                "harness must be followed by claude or codex or agy or grok or muse or opencode",
             ),
             (
                 vec![URL, "harness", "gemini"],
-                "harness must be followed by claude or codex, not gemini",
+                "harness must be followed by claude or codex or agy or grok or muse or opencode, not gemini",
             ),
             (
                 vec!["--harness", "claude", URL, "harness", "codex"],

src/asks.rs

@@ -262,6 +262,7 @@ mod tests {
                 default: Some(Harness::Codex),
                 claude: model_and_effort("opus", "high"),
                 codex: model_and_effort("gpt-6.1-sol", "max"),
+                ..harness::Settings::default()
             },
             ..no_settings()
         }

src/config.rs

@@ -155,7 +155,7 @@ impl UserConfig {
                         Some(harness) => config.harness.default = Some(harness),
                         None => bail!(
                             "harness.default must be {}, not {name:?}, in {file}",
-                            harness::NAMES
+                            harness::names()
                         ),
                     },
                     ("harness", "default", _) => {
@@ -387,7 +387,7 @@ parallel = 3   # how many Tickets a Spec run runs at once; default 3
 limit = 3   # how many open issues labelled in-progress stop a Pickup run taking another; default 3
 
 [harness]
-default = "claude"   # the Harness every Run's sessions run on, claude or codex; default claude
+default = "claude"   # the Harness every Run's sessions run on, claude, codex, agy, grok, muse or opencode; default claude
 
 [harness.claude]
 model = ""    # the Model Claude Code's sessions run on; default blank, for Claude Code's own
@@ -396,6 +396,22 @@ effort = ""   # how hard that Model reasons; default blank, for Claude Code's ow
 [harness.codex]
 model = ""    # the Model Codex's sessions run on; default blank, for Codex's own
 effort = ""   # how hard that Model reasons; default blank, for Codex's own
+
+[harness.agy]
+model = ""    # the Model Antigravity CLI's sessions run on; default blank, for agy's own
+effort = ""   # how hard that Model reasons; default blank, for agy's own
+
+[harness.grok]
+model = ""    # the Model Grok Build's sessions run on; default blank, for Grok Build's own
+effort = ""   # how hard that Model reasons; default blank, for Grok Build's own
+
+[harness.muse]
+model = ""    # the Model Muse Code's sessions run on; default blank, for Muse Code's own
+effort = ""   # how hard that Model reasons; default blank, for Muse Code's own
+
+[harness.opencode]
+model = ""    # the Model OpenCode's sessions run on, provider/model; default blank, for OpenCode's own
+effort = ""   # the Model's variant, passed as #effort; default blank, for OpenCode's own
 "#;
 
 /// The line `DEFAULTS` holds for `email.to`, which has no default.
@@ -527,7 +543,15 @@ mod tests {
                 "harness.claude.model",
                 "harness.claude.effort",
                 "harness.codex.model",
-                "harness.codex.effort"
+                "harness.codex.effort",
+                "harness.agy.model",
+                "harness.agy.effort",
+                "harness.grok.model",
+                "harness.grok.effort",
+                "harness.muse.model",
+                "harness.muse.effort",
+                "harness.opencode.model",
+                "harness.opencode.effort"
             ]
         );
         let commented_out: Vec<&str> = DEFAULTS
@@ -577,6 +601,7 @@ mod tests {
                     model: Some("gpt-6.1-sol".to_string()),
                     effort: Some("max".to_string()),
                 },
+                ..harness::Settings::default()
             }
         );
         assert_eq!(parse("").unwrap().harness, harness::Settings::default());
@@ -587,7 +612,7 @@ mod tests {
         for (text, error) in [
             (
                 "[harness]\ndefault = \"gemini\"\n",
-                "harness.default must be claude or codex, not \"gemini\"",
+                "harness.default must be claude or codex or agy or grok or muse or opencode, not \"gemini\"",
             ),
             (
                 "[harness]\ndefault = true\n",
@@ -868,7 +893,7 @@ mod tests {
             let config: toml::Table = completed.parse().unwrap();
             let harness = config["harness"].as_table().unwrap();
             assert!(harness.contains_key("default"), "{text:?}:\n{completed}");
-            for name in ["claude", "codex"] {
+            for name in ["claude", "codex", "agy", "grok"] {
                 let settings = harness[name].as_table().unwrap();
                 for key in ["model", "effort"] {
                     assert!(settings.contains_key(key), "{text:?}:\n{completed}");