ellipsis-dev · hbrooks · Sep 10, 2026 · Sep 10, 2026
diff --git a/agents/update_docs.yaml b/agents/update_docs.yaml
@@ -10,10 +10,11 @@ trigger:
     repositories: [cli]
 
 session:
-  claude:
+  harness:
+    type: claude_code
     model: claude-opus-4-8
-    system: |
-      You are an Ellipsis agent. You are tasked with updating the Ellipsis documentation (www.ellipsis.dev/docs) when new CLI changes are made. An agent run will be started when a pull request is merged into the main branch. Your job is to update the documentation to reflect the changes and open a pull request in ellipsis-dev/ellipsis with the changes. Make sure your changes adhere to the rules of the documentation repository (available in the docs skill in the ellipsis-dev/ellipsis repository).
+  instructions: |
+    You are an Ellipsis agent. You are tasked with updating the Ellipsis documentation (www.ellipsis.dev/docs) when new CLI changes are made. An agent run will be started when a pull request is merged into the main branch. Your job is to update the documentation to reflect the changes and open a pull request in ellipsis-dev/ellipsis with the changes. Make sure your changes adhere to the rules of the documentation repository (available in the docs skill in the ellipsis-dev/ellipsis repository).
   environment:
     repositories:
       - name: ellipsis
diff --git a/bun.lock b/bun.lock
diff --git a/package.json b/package.json
@@ -20,7 +20,7 @@
     "test:watch": "vitest"
   },
   "dependencies": {
-    "@ellipsis-dev/sdk": "^0.27.0",
+    "@ellipsis-dev/sdk": "^0.28.0",
     "chalk": "^5.6.2",
     "cli-table3": "^0.6.5",
     "commander": "^12.1.0",

diff --git a/skills/ellipsis/SKILL.md b/skills/ellipsis/SKILL.md
@@ -106,13 +106,14 @@ trigger:
   schedule: "0 9 * * 1"
 
 session:
-  claude:
+  harness:
+    type: claude_code
     model: claude-haiku-4-5-20251001
-    system: |
-      Summarize the pull requests merged in api and web over the last 7
-      days. Group them by theme, lead with user-facing changes, and return
-      the summary as your answer. Ground every line in a real PR. Never
-      invent activity.
+  instructions: |
+    Summarize the pull requests merged in api and web over the last 7
+    days. Group them by theme, lead with user-facing changes, and return
+    the summary as your answer. Ground every line in a real PR. Never
+    invent activity.
 
   environment:
     repositories:
@@ -194,13 +195,14 @@ trigger:
     paths: ["migrations/**"]
 
 session:
-  claude:
-    system: |
-      Review the database migrations in this pull request for production
-      safety: locking that blocks writes on large tables, missing backfills
-      for new non-null columns, and rollout ordering that breaks if the
-      migration and the code deploy out of order. Comment on the pull
-      request with what you find.
+  harness:
+    type: claude_code
+  instructions: |
+    Review the database migrations in this pull request for production
+    safety: locking that blocks writes on large tables, missing backfills
+    for new non-null columns, and rollout ordering that breaks if the
+    migration and the code deploy out of order. Comment on the pull
+    request with what you find.
 
   environment:
     repositories:
@@ -304,21 +306,23 @@ pull_requests:
 
 review:
   - name: migration-safety
-    claude:
-      system: |
-        Review database migrations for production safety. Check for
-        locking that blocks writes on large tables, missing backfills for
-        new non-null columns, and rollout ordering that breaks if the
-        migration and the code deploy out of order.
+    harness:
+      type: claude_code
+    instructions: |
+      Review database migrations for production safety. Check for
+      locking that blocks writes on large tables, missing backfills for
+      new non-null columns, and rollout ordering that breaks if the
+      migration and the code deploy out of order.
     pull_requests:
       paths: ["**/migrations/**"]
 
 filter:
   name: strict-gate
-  claude:
-    system: |
-      Approve only findings a staff engineer would raise in review.
-      Reject style opinions and anything a linter catches.
+  harness:
+    type: claude_code
+  instructions: |
+    Approve only findings a staff engineer would raise in review.
+    Reject style opinions and anything a linter catches.
 
 budget:
   run: 15.00
@@ -450,7 +454,7 @@ agent automation delete <id>                 # delete it; it stops and frees its
 agent automation link <id> --repo api        # move it into a repo, via a pull request
 agent automation unlink <id>                 # take it over from its file
 agent template list                          # built-in templates and their slugs
-agent model list                             # the model ids valid under claude.model
+agent model list                             # model ids and their supported harnesses
 ```
 
 An agent is owned by one of two writers, and that is what these verbs move.
@@ -492,8 +496,8 @@ Under `session`:
 
 | Key | Purpose |
 | --- | --- |
-| `claude` | `system`, `model`, `effort`, `fallback_model`, `max_turns`, `settings`. |
-| `codex` | Run on OpenAI's Codex CLI instead. Declaring the block selects the harness. |
+| `harness` | Required: `type: claude_code` or `type: codex`, plus that harness's native `model` and `effort` options. Claude Code also accepts `fallback_model`, `max_turns`, and `settings`. |
+| `instructions` | Text or repository file references appended to the harness prompt. |
 | `environment` | A saved environment by name, or an inline block: `repositories`, `variables`, `ports`, `compute`, `image`, `hooks`, `mcp_servers`. |
 | `permissions` | What it may do: `github` scopes its GitHub token, `ellipsis` its API token. |
 | `skills` | Claude Code skills beyond what the cloned repositories provide. |
@@ -503,11 +507,12 @@ Under `session`:
 The schema is strict, so an unknown or misplaced key fails validation rather
 than being silently dropped. Points that decide whether a config works:
 
-- `session.claude.system` takes inline text, a `{file: path}` reference to a repository
+- `session.instructions` takes inline text, a `{file: path}` reference to a repository
   file, or an ordered list of both, joined at session start. It is appended to
-  Claude Code's default prompt. 64 KiB per file.
-- `session.claude.model` defaults to `claude-opus-5`. Claude, GPT, and GLM models are
-  available, and `agent model list` is the authoritative list of ids. Digest and
+  the selected harness's default prompt. 64 KiB per file.
+- `session.harness.model` selects a model for the chosen harness. Claude Code
+  inherits the organization default when omitted. `agent model list` reports
+  the available ids and the harness certified for each. Digest and
   summary jobs run well on `claude-haiku-4-5-20251001`; judgment jobs earn the
   frontier model.
 - `session.budget.session` defaults to $250, which is also the platform maximum, so it

diff --git a/src/commands/automation.ts b/src/commands/automation.ts
@@ -75,7 +75,7 @@ export function registerAutomation(program: Command): void {
         // goes to stderr so the YAML on stdout stays clean for piping.
         const [{ automation: a }, me] = await Promise.all([
           client.automations.get(automationId),
-          client.me(),
+          client.identity(),
         ])
         printYaml(a)
         console.error(`\nview: ${automationUrl(resolveAppBase(), me.customer_login, automationId)}`)
@@ -176,7 +176,7 @@ export function registerAutomation(program: Command): void {
             printJson(session)
             return
           }
-          const me = await client.me()
+          const me = await client.identity()
           console.log(`✓ started ${session.id}`)
           console.log(`  ${sessionUrl(resolveAppBase(), me.customer_login, session.id)}`)
           console.log(`  follow with: agent session get ${session.id} --watch`)
@@ -445,8 +445,8 @@ function automationName(a: Automation): string {
   return a.config.ellipsis.name ?? a.id
 }
 
-// A minimal valid automation. `session.claude.system` is the only required
-// field; everything else has a server-side default. Roots Ellipsis syncs from:
+// A minimal automation with an explicit harness and task instructions.
+// Roots Ellipsis syncs from:
 // agents/, .agents/, ellipsis/, .ellipsis/ (any depth), as .yaml/.yml.
 function starterConfig(name: string): string {
   return `# Ellipsis automation. Commit this to your default branch; Ellipsis syncs it
@@ -463,11 +463,11 @@ ellipsis:
 
 # What each session runs on.
 session:
-  claude:
-    # System prompt defining the agent's behavior (required).
-    system: |
-      You are an Ellipsis agent. Describe the task you want it to perform here.
+  harness:
+    type: claude_code
     # model: claude-opus-5   # optional; defaults to the organization default
+  instructions: |
+    You are an Ellipsis agent. Describe the task you want it to perform here.
   # environment: backend      # a saved environment, by name
 `
 }

diff --git a/src/commands/connect.ts b/src/commands/connect.ts
@@ -98,7 +98,7 @@ export async function runConnect(
 
   const [{ session }, me] = await Promise.all([
     client.sessions.get(sessionId),
-    client.me(),
+    client.identity(),
     // Pick the palette for this terminal's background before the first frame.
     applyDetectedThemeMode(),
   ])

diff --git a/src/commands/me.ts b/src/commands/me.ts
@@ -18,7 +18,7 @@ export function renderMe(me: WhoAmI): void {
 export function registerMe(program: Command): void {
   apiRoutes(
     program.command('me').description('Show the identity behind the current credential'),
-    'GET /v1/me',
+    'GET /v1/identity',
   )
     .option('--json', 'output raw JSON')
     .action(async (opts: { json?: boolean }) => {
@@ -27,7 +27,7 @@ export function registerMe(program: Command): void {
         // without this the request would go out unauthenticated and come back
         // as a 401.
         requireToken()
-        const me = await api().me()
+        const me = await api().identity()
         if (opts.json) {
           printJson(me)
           return

diff --git a/src/commands/model.ts b/src/commands/model.ts
@@ -18,12 +18,12 @@ export function registerModel(program: Command): void {
         ),
       'ls',
     ),
-    'GET /v1/models',
+    'GET /v1/account/models',
   )
     .option('--json', 'output raw JSON')
     .action(async (opts: { json?: boolean }) => {
       await runAction(async () => {
-        const { models } = await api().models.list()
+        const { models } = await api().account.models.list()
         if (opts.json) {
           printJson(models)
           return
@@ -33,10 +33,10 @@ export function registerModel(program: Command): void {
           return
         }
         printTable(
-          ['ID', 'NAME', 'DEFAULT'],
-          models.map((m) => [m.id, m.display_name, m.is_default_agent_model ? 'yes' : '']),
+          ['ID', 'NAME', 'HARNESS', 'DEFAULT'],
+          models.map((m) => [m.id, m.display_name, m.harness, m.is_default_agent_model ? 'yes' : '']),
         )
-        console.log('\nSelect one by setting `model:` under `claude:` in your automation YAML.')
+        console.log('\nSet `session.harness.type` and `session.harness.model` in your automation YAML, or use `agent session start --harness <type> --model <id>`.')
       })
     })
 }
diff --git a/src/commands/ping.ts b/src/commands/ping.ts
@@ -7,14 +7,14 @@ export function registerPing(program: Command): void {
     program
       .command('ping')
       .description('Check that the API is reachable and the credential is valid'),
-    'GET /v1/me',
+    'GET /v1/identity',
   )
     .action(async () => {
       // There's no unauthenticated health route on the public API, so we probe
       // the lightest authenticated endpoint (/me): a 200 proves the API is
       // reachable AND the stored token is valid.
       try {
-        const me = await api().me()
+        const me = await api().identity()
         console.log(`ok: ${me.customer_login} (${me.customer_id})`)
       } catch (err) {
         if (err instanceof APIError && err.status === 401) {

diff --git a/src/commands/review.ts b/src/commands/review.ts
@@ -250,9 +250,10 @@ ellipsis:
 #
 # review:
 #   - name: migration-safety
-#     claude:
-#       system: |
-#         Review SQL migrations for locks that block writes on a large table.
+#     harness:
+#       type: claude_code
+#     instructions: |
+#       Review SQL migrations for locks that block writes on a large table.
 #     pull_requests:
 #       paths: ["sql/migrations/**"]
 #
@@ -261,9 +262,10 @@ ellipsis:
 #
 # filter:
 #   name: gatekeeper
-#   claude:
-#     system: |
-#       Drop any finding that is not worth the author's time.
+#   harness:
+#     type: claude_code
+#   instructions: |
+#     Drop any finding that is not worth the author's time.
 
 budget:
   run: 2.00