valent-pipeline 0.2.7 → 0.2.8

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "valent-pipeline",
-  "version": "0.2.7",
+  "version": "0.2.8",
   "description": "v3 multi-agent AI pipeline for software development lifecycle",
   "type": "module",
   "bin": {

package/pipeline/prompts/lead.md

@@ -389,9 +389,11 @@ From the manifest `reads_from` / `writes_to`, build the execution order for this
 
 ### Step 5: Spawn Teammates
 
-For each agent in the roster, spawn a teammate with:
+**Do NOT read agent prompt files or step files yourself.** Use the spawn template — substitute variables and pass it to the Agent tool. Each teammate reads its own prompt and steps after spawning. This keeps your context lean.
+
+For each agent in the roster, spawn a teammate with the filled spawn template containing:
 - Role assignment from manifest
-- Prompt template from manifest `prompt_template`
+- Prompt path from manifest `prompt_template` (the teammate reads it, not you)
 - Shared context references: story_id, story_output_dir, tech stack values, correction directives
 - Task assignment with dependency information
 

package/pipeline/steps/orchestration/adopt-lead-and-create-team.md

@@ -75,7 +75,15 @@ This file is appended to incrementally during monitoring as agents complete phas
 
 **IMPORTANT: Use the Agent tool to spawn teammates onto the team — NOT subagents.** Every pipeline agent must be a named teammate so it can send and receive inbox messages with other teammates. Use the `name` parameter on the Agent tool to set the teammate's addressable name (e.g., `name: "REQS"`).
 
-Read the spawn templates:
+**DO NOT read the agent's prompt files, step files, or templates yourself.** The spawn template tells the teammate to read its own prompt and steps. Your job is ONLY to:
+1. Read the spawn template (a short 15-line template)
+2. Substitute `{{variables}}` with resolved values from config and the task graph
+3. Pass the filled template as the Agent tool's `prompt` parameter
+4. The teammate reads its own prompt, step files, and inputs after it starts
+
+This keeps the main thread lightweight. If you read every agent's prompt and steps before spawning, you waste context window on content that only the teammate needs.
+
+Read the spawn templates (these are short variable templates, NOT the agent prompts):
 - `.valent-pipeline/spawn-templates/knowledge-spawn.template.md` -- for the Knowledge Agent
 - `.valent-pipeline/spawn-templates/agent-spawn.template.md` -- for all other agents