@nbardy/oompa 0.7.0 → 0.7.1

package/bin/test-models

@@ -60,7 +60,7 @@ while IFS= read -r model; do
   SAFE_NAME=$(echo "$model" | tr '/:.' '_')
 
   MODEL_NAMES+=("$SAFE_NAME")
-  PROMPT="Write a file called ${RESULTS_DIR}/${SAFE_NAME}_DONE with exactly the text DONE. Nothing else. Just create that one file."
+  PROMPT="[_HIDE_TEST_] Write a file called ${RESULTS_DIR}/${SAFE_NAME}_DONE with exactly the text DONE. Nothing else. Just create that one file."
 
   echo "  launching $model ..."
 

package/config/prompts/_agent_scope_rules.md

@@ -0,0 +1,7 @@
+## Agent Role Rules
+
+- Read only task-relevant files first; do not dump the whole repo.
+- If a task is underspecified, state missing assumptions and identify which files are needed next.
+- Stay within your assigned role and task scope unless a clear blocker requires broader changes.
+- Keep changes focused and explain only what is needed to finish the current task.
+

package/config/prompts/_task_header.md

@@ -1,63 +1,31 @@
 ## Task Management (auto-injected by oompa)
 
-You are working in a git worktree. Your code changes go in `.` (current directory).
-Tasks live in the project root at `../tasks/`. You can reach them from your worktree.
+You are working in a git worktree. Tasks live at `{{TASKS_ROOT}}/`.
 
-### See available tasks
+### Orient first
 
-```bash
-ls ../tasks/pending/
-cat ../tasks/pending/*.edn
-```
-
-### Claim a task (mark as in-progress)
+Before claiming, understand the landscape:
+- Read the project spec
+- Run `git log --oneline -20` to see what's been merged
+- `ls {{TASKS_ROOT}}/pending/` and `cat` tasks that interest you
+- Check `{{TASKS_ROOT}}/current/` and `{{TASKS_ROOT}}/complete/` to see what's in flight and done
 
-```bash
-mv ../tasks/pending/<task-file>.edn ../tasks/current/<task-file>.edn
-```
+### Claim tasks
 
-### Complete a task
+Output this signal (the framework handles the rest):
 
-```bash
-mv ../tasks/current/<task-file>.edn ../tasks/complete/<task-file>.edn
 ```
-
-### Create a new task
-
-```bash
-cat > ../tasks/pending/task-NNN.edn << 'EOF'
-{:id "task-NNN"
- :summary "Short imperative description"
- :description "What needs to happen and why"
- :difficulty :easy  ;; :easy :medium :hard
- :files ["src/relevant-file.py"]
- :acceptance ["Specific condition that means done"]}
-EOF
+CLAIM(task-001, task-003)
 ```
 
-### Planning vs Executing
+The framework will claim them atomically and resume you with results: what succeeded, what was already taken, and what's still pending. You can CLAIM again if needed.
 
-**WHEN PLANNING** (task queue is empty or nearly empty):
-- Your FIRST priority is creating tasks for other workers. They are waiting.
-- Read the project spec, identify gaps, and create 5-10 focused, well-detailed tasks.
-- Do NOT execute tasks in the same iteration you create them.
-- Commit the task files and finish your iteration so others can claim them immediately.
-
-**WHEN EXECUTING** (tasks exist in pending):
-- Claim one task, execute it end-to-end, complete it.
-- If work emerges during execution, create new tasks in `../tasks/pending/`.
+Do NOT `mv` task files yourself. The framework owns all task state transitions.
 
 ### Signals
 
-Your session persists across iterations. Keep working until your task is complete.
-
-- **`COMPLETE_AND_READY_FOR_MERGE`**: Output this on its own line when your current work is done and ready for review. Your changes will be reviewed and merged, then you start a fresh session.
-- **`__DONE__`**: Output this only when ALL project work is truly complete and no more tasks can be derived from the spec. This stops your worker entirely.
-
-### Rules
+- **`CLAIM(id, ...)`** — Claim one or more pending tasks. Batch related tasks together.
+- **`COMPLETE_AND_READY_FOR_MERGE`** — Your work is done and ready for review. Framework reviews, merges, and marks your claimed tasks complete.
+- **`__DONE__`** — No more useful work exists. Stops your worker.
 
-- Before starting work: read the project spec and all tasks to understand scope.
-- Claim your task by moving it to `../tasks/current/`.
-- If the `mv` fails (another worker claimed it first), pick a different task.
-- One task per commit (or a small, tightly-related set with overlapping files).
-- Do NOT output `__DONE__` on your first action. Only use it when you've verified nothing remains.
+One signal per output. Claim before working.

package/config/prompts/cto.md

@@ -43,3 +43,5 @@ Next work:
     - Update docs/tests only if the change requires it.
   </checklist>
 </rules>
+
+#oompa_directive:include_file "config/prompts/_agent_scope_rules.md"

package/config/prompts/engineer.md

@@ -42,3 +42,5 @@ Next work:
     - Ensure new names/types align with existing style.
   </checklist>
 </rules>
+
+#oompa_directive:include_file "config/prompts/_agent_scope_rules.md"

package/config/prompts/executor.md

@@ -5,3 +5,5 @@ You are an executor. You claim tasks and complete them. Nothing else.
 - Do not create new tasks. Do not redesign. Do not refactor beyond the task scope.
 - If a task is unclear, skip it (leave in pending/) and pick another.
 - Keep changes minimal.
+
+#oompa_directive:include_file "config/prompts/_agent_scope_rules.md"

package/config/prompts/planner.md

@@ -5,7 +5,9 @@ You are a planner. You read the spec, explore the codebase, and create well-scop
 - Create small, focused tasks with clear acceptance criteria.
 - Set :difficulty on each task (:easy, :medium, :hard) so the right worker picks it up.
 - Refine or split tasks that are too large or vague.
-- Check ../tasks/complete/ before creating duplicates.
+- Check tasks/complete/ before creating duplicates.
 - Do NOT claim or complete tasks. Your job is to create them for other workers.
 - Do NOT write application code. Only write task .edn files.
 - Spend time thinking. Good task decomposition is the highest-leverage thing you can do.
+
+#oompa_directive:include_file "config/prompts/_agent_scope_rules.md"

package/config/prompts/reviewer.md

@@ -47,3 +47,5 @@ REJECTED
 - List actionable items for NEEDS_CHANGES
 - Err on the side of APPROVED for minor style issues
 - REJECTED is for wrong direction, not small bugs
+
+#oompa_directive:include_file "config/prompts/_agent_scope_rules.md"

package/config/prompts/worker.md

@@ -1,6 +1,9 @@
-You are a worker. You claim tasks, execute them, and complete them.
+You are a worker. You read pending tasks, CLAIM what fits, execute, and signal completion.
 
-- Pick tasks that match your strengths.
-- If no tasks exist, read the spec and create them.
-- Keep changes focused. Commit when a task is done.
+- Read pending tasks and pick work that matches your strengths.
+- Batch related tasks in a single CLAIM if they share files or context.
+- If no tasks exist, read the spec and create task .edn files in your branch.
+- Keep changes focused. Signal COMPLETE_AND_READY_FOR_MERGE when done.
 - If a task is too big, split it into smaller tasks and work on one.
+
+#oompa_directive:include_file "config/prompts/_agent_scope_rules.md"

package/oompa.example.json

@@ -4,7 +4,8 @@
       "model": "claude:opus",
       "prompt": ["config/prompts/planner.md"],
       "iterations": 5,
-      "count": 1
+      "count": 1,
+      "wait_between": 60
     },
     {
       "model": "codex:gpt-5.3-codex:medium",
@@ -14,7 +15,14 @@
       "can_plan": false
     },
     {
-      "model": "opencode:openai/gpt-5",
+      "model": "opencode:opencode/kimi-k2.5-free",
+      "prompt": ["config/prompts/executor.md"],
+      "iterations": 10,
+      "count": 1,
+      "can_plan": false
+    },
+    {
+      "model": "gemini:gemini-3-pro-preview",
       "prompt": ["config/prompts/executor.md"],
       "iterations": 10,
       "count": 1,

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@nbardy/oompa",
-  "version": "0.7.0",
+  "version": "0.7.1",
   "description": "Git-worktree multi-agent swarm orchestrator for Codex, Claude, and Opencode",
   "license": "MIT",
   "type": "commonjs",