@agntn/harnesses 0.2.1 → 0.3.0

package/README.md

@@ -76,7 +76,7 @@ harnesses agents sync --check
 harnesses prompts sync --check
 ```
 
-`run` without `--tools` is the advisor. `--tools` is the full agent. `--read-only` asks the CLI for a sandbox and implies tools. Timeouts, `--cwd` and `--model` sit in the [CLI guide](https://harnesses.agntn.dev/guide/cli).
+`run` without `--tools` is the advisor. `--tools` is the full agent. `--read-only` asks the CLI for a sandbox and implies tools, so pairing it with `--no-tools` is an error. Timeouts, `--cwd` and `--model` sit in the [CLI guide](https://harnesses.agntn.dev/guide/cli).
 
 ### Commands
 
@@ -103,14 +103,21 @@ import { getHarness, detectHarness } from "@agntn/harnesses";
 const claude = getHarness("claude");
 const paths = claude.resolve({ platform: "linux", homeDir: "/home/dev" });
 console.log(paths.skills);
+console.log(paths.temp, claude.envOverrides);
 
 const active = detectHarness();
 if (active) console.log(active.id);
 
 await getHarness("codex").invoke("Review this patch", { readOnly: true });
+
+const agy = getHarness("antigravity");
+const [model] = (await agy.listModels({ search: "gemini" })).models;
+if (model) await agy.invoke("Review this patch", { tools: true, model: agy.modelSelector(model) });
 ```
 
-That's most of it, really. `getHarness` wants an exact id. `detectHarness` uses env vars first, then a single project marker. `invoke()` talks to the CLI. A mode the CLI cannot run comes back as an error, not a quieter one. The rest: [Registry](https://harnesses.agntn.dev/guide/registry), [Invoke](https://harnesses.agntn.dev/guide/invoke), [MCP servers](https://harnesses.agntn.dev/guide/mcp-servers), [Instructions files](https://harnesses.agntn.dev/guide/agents-sync).
+`temp` is where a harness drops its own temp files, and `envOverrides` names the variables that move its directories. `claude.resolve().temp` follows `CLAUDE_CODE_TMPDIR` when it's set, which is handy when `/tmp` is tmpfs and Claude's scratchpads are quietly eating RAM.
+
+That's most of it, really. `getHarness` wants an exact id. `detectHarness` uses env vars first, then a single project marker. `invoke()` talks to the CLI. `listModels()` asks it what it can run, and `modelSelector()` turns one of those into what `model` takes, since Pi wants `provider/id` and Antigravity the bare id. A mode the CLI cannot run comes back as an error, not a quieter one. The rest: [Registry](https://harnesses.agntn.dev/guide/registry), [Invoke](https://harnesses.agntn.dev/guide/invoke), [MCP servers](https://harnesses.agntn.dev/guide/mcp-servers), [Instructions files](https://harnesses.agntn.dev/guide/agents-sync).
 
 ### Prompt templates
 
@@ -123,9 +130,11 @@ harnesses prompts sync --check
 harnesses prompts sync
 ```
 
-Markdown harnesses receive symlinks to those files. Gemini receives generated
-TOML commands, with `$ARGUMENTS` translated to `{{args}}`. Unmanaged files at a
-managed name are backed up under the same XDG data tree before replacement.
+For a Markdown harness, the whole prompt directory becomes one symlink to that
+directory, so a template added through any harness shows up in all of them.
+Gemini gets generated TOML commands, with `$ARGUMENTS` translated to `{{args}}`,
+and its commands directory keeps nothing else. Whatever was there before is
+backed up under the same XDG data tree first.
 
 `getHarness("pi").promptTemplates` still lists every verified reusable prompt
 location. `promptTemplateSyncTarget` is the single stable user destination the

package/dist/_chunks/libs/typebox.mjs

@@ -5434,6 +5434,6 @@ Union([
 		path: String$2()
 	})
 ]);
-export { Check, Errors, typebox_exports };
+export { Check, Errors, IsObject, typebox_exports };
 
 //# sourceMappingURL=typebox.mjs.map