@runtypelabs/cli 2.55.4 → 2.56.0

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
Files changed (3) hide show
  1. package/README.md +55 -0
  2. package/dist/index.js +9384 -4317
  3. package/package.json +8 -5
package/README.md CHANGED
@@ -47,6 +47,60 @@ an intentionally pinned global version), or
47
47
  `RUNTYPE_DISABLE_UPDATE_CHECK=true` to keep the active managed version without checking for more
48
48
  updates.
49
49
 
50
+ ## Connect a local coding harness
51
+
52
+ The `runtype harness` commands register and run an installed coding harness as an
53
+ external simulation target. You do not need to clone this repository or start a
54
+ separate device listener.
55
+
56
+ ```bash
57
+ pnpm add --global @runtypelabs/cli
58
+ runtype auth login
59
+ runtype harness doctor --harness claude --mcp-config ./mcp.json
60
+ runtype harness connect --harness claude --mcp-config ./mcp.json
61
+ ```
62
+
63
+ Use a CLI release and API deployment that both support signed harness telemetry.
64
+ The host validates telemetry access before starting each model turn. An older API
65
+ cannot run this host. `doctor` checks installation, authentication, and MCP discovery
66
+ without a model call. `connect` prints the registration, device, agent, and target IDs
67
+ and stays in the foreground.
68
+
69
+ An MCP configuration names the HTTP servers the harness may use:
70
+
71
+ ```json
72
+ {
73
+ "mcpServers": {
74
+ "fixture": { "url": "http://127.0.0.1:4555/mcp" }
75
+ }
76
+ }
77
+ ```
78
+
79
+ Use HTTPS for remote servers. The `controlled-mcp` profile permits only the configured
80
+ MCP tools. The `workspace` profile is unavailable. External tool writes are real,
81
+ and the simulation reports isolation as `unverifiable`. Start with read-only tasks
82
+ and a fixture server.
83
+
84
+ ```bash
85
+ runtype harness status YOUR_REGISTRATION_ID --json
86
+ runtype harness disconnect YOUR_REGISTRATION_ID
87
+ runtype harness connect --registration YOUR_REGISTRATION_ID
88
+ runtype harness remove YOUR_REGISTRATION_ID --yes
89
+ ```
90
+
91
+ Disconnect stops the local host and preserves registration. Remove deletes its
92
+ owned target and agent and revokes its paired device. Use the same Runtype API,
93
+ organization, and user that created the registration.
94
+
95
+ The default capture policy is `redacted`. Select `--data-policy metrics` to omit
96
+ content or `--data-policy full` to retain allowed content. The stricter host or
97
+ organization policy wins. Known secrets remain scrubbed in every mode. Reports
98
+ show per-turn coverage, delivery loss, provenance, and trace links. Harness costs
99
+ are estimates and do not affect platform billing or simulation spend caps.
100
+
101
+ See [Run simulations with a local coding harness](https://docs.runtype.com/developer-guides/guides/local-coding-harnesses)
102
+ for exact versions, key permissions, ACP setup, and retention limits.
103
+
50
104
  ## Start Here
51
105
 
52
106
  The easiest way to try Runtype starting at the CLI is to run a Marathon task. If you are not logged in yet, the CLI will guide you through login or signup first.
@@ -719,6 +773,7 @@ Each entity the TypeScript SDK converges has an `ensure` and a `pull` subcommand
719
773
  | `tools ensure` / `tools pull` | `defineTool` input |
720
774
  | `products ensure` / `products pull` | `defineProduct` input |
721
775
  | `surfaces ensure` / `surfaces pull --product` | `defineSurface` input |
776
+ | `eval ensure` / `eval pull` | `defineEval` input |
722
777
  | `products ensure-fpo` / `products pull-fpo` | Full Product Object |
723
778
 
724
779
  `agents skills` manages Runtype Agent Skills, the context bundles deployed agents