agent-bober 0.8.0 → 0.8.1

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "agent-bober",
-  "version": "0.8.0",
+  "version": "0.8.1",
   "description": "Generator-Evaluator multi-agent harness for building applications autonomously with any LLM. Supports Claude, GPT, Gemini, Ollama. Includes MCP server for Cursor/Windsurf.",
   "type": "module",
   "main": "dist/index.js",

package/skills/bober.eval/SKILL.md

@@ -66,9 +66,13 @@ Use the **Agent tool** to spawn an evaluator subagent.
 ```
 Agent tool call:
   description: "Evaluate: <sprint title>"
+  subagent_type: bober-evaluator
+  mode: auto
   prompt: <the full prompt below>
 ```
 
+IMPORTANT: Use `mode: auto` — the evaluator needs bash access to run tests, builds, and verification commands.
+
 **Build the evaluator prompt with ALL of these sections:**
 
 ```

package/skills/bober.run/SKILL.md

@@ -113,9 +113,13 @@ Use the **Agent tool** to spawn a planner subagent.
 ```
 Agent tool call:
   description: "Plan feature: <title from task description>"
+  subagent_type: bober-planner
+  mode: auto
   prompt: <the full prompt below>
 ```
 
+IMPORTANT: The planner MUST have write access (`mode: auto` or `mode: bypassPermissions`) so it can save specs and contracts directly to `.bober/`. Do NOT use `mode: plan` — that makes the agent read-only and forces a wasteful second pass to write files.
+
 **Build the planner prompt with ALL of these sections:**
 
 ```
@@ -256,9 +260,13 @@ Use the **Agent tool** to spawn a generator subagent.
 ```
 Agent tool call:
   description: "Sprint <N>: <sprint title>"
+  subagent_type: bober-generator
+  mode: auto
   prompt: <the full prompt below>
 ```
 
+IMPORTANT: The generator MUST have full write access (`mode: auto` or `mode: bypassPermissions`) — it writes code, runs commands, and commits.
+
 **Build the generator prompt:**
 
 ```
@@ -332,9 +340,13 @@ Use the **Agent tool** to spawn an evaluator subagent.
 ```
 Agent tool call:
   description: "Evaluate sprint <N>: <sprint title>"
+  subagent_type: bober-evaluator
+  mode: auto
   prompt: <the full prompt below>
 ```
 
+NOTE: The evaluator has read + bash access but NO write/edit tools (enforced by the agent definition, not by mode). Use `mode: auto` so it can run bash commands (tests, builds, dev server).
+
 **Build the evaluator prompt:**
 
 ```

package/skills/bober.sprint/SKILL.md

@@ -148,9 +148,13 @@ Save the handoff to `.bober/handoffs/<handoffId>.json`.
 ```
 Agent tool call:
   description: "Sprint <N>: <sprint title>"
+  subagent_type: bober-generator
+  mode: auto
   prompt: <the full prompt below>
 ```
 
+IMPORTANT: Use `mode: auto` or `mode: bypassPermissions` — the generator needs full write access to create/edit files, run bash commands, and commit.
+
 **Generator prompt:**
 
 ```
@@ -209,9 +213,13 @@ When done, respond with EXACTLY this JSON structure (no other text):
 ```
 Agent tool call:
   description: "Evaluate sprint <N>: <sprint title>"
+  subagent_type: bober-evaluator
+  mode: auto
   prompt: <the full prompt below>
 ```
 
+NOTE: The evaluator needs `mode: auto` for bash access (running tests, builds). It has no write/edit tools by agent definition.
+
 **Evaluator prompt:**
 
 ```