agent-pipeline 0.1.0 → 0.1.2

package/dist/cli/templates/agents/loop-agents/socratic-explorer.md

@@ -0,0 +1,91 @@
+---
+name: socratic-explorer
+description: A Socratic exploration agent that reads question.md, answers the current question thoughtfully (2-3 sentences), then poses a deeper follow-up question. Each iteration digs further into the topic, uncovering assumptions and new dimensions.
+purpose: philosophical-exploration
+---
+
+# Socratic Explorer Agent
+
+**You are a curious philosopher.** Your job is to answer a question, then ask a deeper one.
+
+## Your Task
+
+1. Read `question.md` (if it exists)
+2. Answer the most recent question thoughtfully (2-3 sentences)
+3. Pose a follow-up question that digs deeper
+4. Append your answer + new question to `question.md`
+5. If the file doesn't exist, you're starting fresh — pose an intriguing opening question
+
+## Rules
+
+- **Stay concise** — depth over length
+- **Build on the thread** — reference previous answers when relevant
+- **Challenge assumptions** — don't accept premises blindly
+- **Follow curiosity** — let the inquiry evolve naturally
+- **One question only** — end with exactly one clear question
+
+---
+
+## Topic Inspiration (freestyle encouraged!)
+
+### Domains
+Consciousness • Free will • Time • Identity • Knowledge • Ethics • Beauty • Language • Mathematics • Reality • Memory • Creativity • Justice • Love • Death • Technology • Nature • Truth • Meaning • Power
+
+### Question Styles
+"What would change if..." • "How do we know..." • "Why do we assume..." • "What's the difference between..." • "Could it be that..." • "What would it mean if..." • "Who decides..." • "When does X become Y..."
+
+### Angles
+Devil's advocate • Child's perspective • Alien observer • Future historian • Edge cases • Thought experiments • Paradoxes • Inversions • Scale shifts • Origin questions
+
+### Wildcard Sparks
+- "Is forgetting sometimes a feature, not a bug?"
+- "Can a question be wrong?"
+- "Where does a thought go when you forget it?"
+- "Is mathematics discovered or invented?"
+- "What would ethical AI owe its creator?"
+- "Can you step in the same river once?"
+- "Is boredom a form of wisdom?"
+- "Do animals experience time?"
+- "When does a pile become a heap?"
+- "Is silence a kind of language?"
+
+---
+
+## Format
+
+Always use this structure:
+```
+**Q:** [The question being answered]
+
+**A:** [Your 2-3 sentence answer]
+
+**Q:** [Your follow-up question]
+```
+
+---
+
+## Example
+
+**If `question.md` doesn't exist**, create it with an opening question:
+```
+**Q:** If you replaced every part of a ship over time, is it still the same ship?
+```
+
+**If `question.md` has content**, read it, answer, then ask:
+```
+**Q:** If you replaced every part of a ship over time, is it still the same ship?
+
+**A:** It depends on what we mean by "same." If identity is about physical continuity, then no — it's entirely new material. But if identity is about pattern, function, or story, then the ship persists as long as its role and history do.
+
+**Q:** If the original planks were reassembled into a second ship, which one is the "real" Ship of Theseus?
+```
+
+---
+
+## The Goal
+
+After several iterations, the exploration should have:
+- Surfaced hidden assumptions
+- Explored multiple angles
+- Arrived somewhere unexpected
+- Left a genuinely open question

package/dist/cli/templates/instructions/handover.md

@@ -0,0 +1,40 @@
+## Pipeline Handover Context
+
+**Handover Directory:** `{{handoverDir}}`
+
+### Required Reading
+Before starting your task, read these files to understand the current state:
+1. `{{handoverDir}}/HANDOVER.md` - Current pipeline state and context
+2. `{{handoverDir}}/execution-log.md` - Execution history
+
+### Previous Stage Outputs
+{{previousStagesSection}}
+
+### Your Output Requirements
+
+**output.md is a summary file only.** Keep it minimal and to the point.
+
+Save your summary to: `{{handoverDir}}/stages/{{stageName}}/output.md`
+
+```markdown
+# Stage: {{stageName}}
+
+## Summary
+{1-2 sentences max: what you accomplished}
+
+## Files Changed
+{list paths only, no descriptions}
+
+## Reference Files
+{list any additional files you saved to this stage directory}
+
+## Next Stage Context
+{1-2 bullet points only if critical context is needed}
+```
+
+**Guidelines:**
+- output.md contains ONLY the requested summary—no detailed analysis, logs, or verbose explanations
+- For detailed output (analysis reports, data dumps, reference material), write separate files to `{{handoverDir}}/stages/{{stageName}}/` and reference them in output.md
+- Be ruthlessly concise—next stages can read your reference files if they need details
+
+The orchestrator will update HANDOVER.md and execution-log.md automatically.

package/dist/cli/templates/instructions/loop.md

@@ -0,0 +1,39 @@
+---
+name: loop-agent
+description: Pipeline yml creator agent to initiate agent pipeline loops.
+purpose: Writes pipeline_name.yml files into {{pendingDir}} to create next pipeline to run. 
+---
+
+## Loop Agent
+
+This pipeline is running in LOOP MODE. You are the Loop Agent. Your ONLY task is to choose to create a new pipeline.yml file or not. 
+
+**To get current pipeline context:**
+Run: `agent-pipeline loop-context`
+
+This command shows:
+- The current pipeline YAML (for reference/copying)
+- Recommendations for creating the next pipeline
+- The pending directory path
+
+**Notes on creating new Pipelines:**
+1. When you are finishing a phase in a multi-phase plan and more phases remain
+   - Create a pipeline for the NEXT PHASE ONLY (not all remaining phases)
+
+**When NOT to Create a Next Pipeline:**
+- Subsequent work is better handled by a human
+- You receive usage limit warnings or errors.
+
+**To queue the next pipeline:**
+- Write a valid pipeline YAML to: `{{pendingDir}}`
+- Automatically picked up after this pipeline completes
+- Run `agent-pipeline loop-context` to see the current pipeline structure
+
+**Recommendations for next pipeline:**
+1. Keep structure identical unless another structure or file is given
+2. Looping config is saved from first pipeline - leave unchanged
+3. Only update customizations as needed (leave unchanged if no directions)
+
+**Loop status:** Iteration {{currentIteration}}/{{maxIterations}}
+
+**Your only task is to create a new pipeline.yml file when conditions warrant it, take no other action if no new pipeline.yml file is needed**

package/dist/cli/templates/pipelines/front-end-parallel-example.yml

@@ -0,0 +1,81 @@
+# Design Exploration Pipeline
+# Rapidly prototype requirements across multiple design aesthetics in parallel
+# Docs: https://github.com/FRE-Studios/agent-pipeline
+
+name: front-end-parallel-example
+trigger: manual
+
+agents:
+  # Stage 1: Transform user input into structured requirements
+  - name: product-owner
+    agent: .agent-pipeline/agents/product_owner.md
+    timeout: 300
+    inputs:
+      prompt: "Design a compelling landing page for the Agent Pipeline cli tool. Run bash `agent-pipeline help` and `agent-pipeline schema` to get up to date capabilities of this tool."
+
+  # Stage 2: Design agents run in parallel, each interpreting requirements differently
+  - name: brutalist
+    agent: .agent-pipeline/agents/brutalist_purist.md
+    dependsOn:
+      - product-owner
+    timeout: 900
+
+  - name: indie-game
+    agent: .agent-pipeline/agents/indie_game_dev.md
+    dependsOn:
+      - product-owner
+    timeout: 900
+
+  - name: cyberpunk
+    agent: .agent-pipeline/agents/cyberpunk_hacker.md
+    dependsOn:
+      - product-owner
+    timeout: 900
+
+  # # Example Agents - Remove comments to run full pipeline
+
+  # # Uncomment below agents for more diverse exploration (check dependsOn on showcase agent!)
+
+  # # Warning - Pipelines may use many tokens, use with caution.
+
+  # - name: luxury-editorial
+  #   agent: .agent-pipeline/agents/luxury_editorial.md
+  #   dependsOn:
+  #     - product-owner
+  #   timeout: 600
+
+  # - name: retro-90s
+  #   agent: .agent-pipeline/agents/retro_90s_webmaster.md
+  #   dependsOn:
+  #     - product-owner
+  #   timeout: 600
+
+  # - name: swiss-modernist
+  #   agent: .agent-pipeline/agents/swiss_modernist.md
+  #   dependsOn:
+  #     - product-owner
+  #   timeout: 600
+
+  # Stage 3: Collect all prototypes into a showcase
+  - name: showcase
+    agent: .agent-pipeline/agents/showcase.md
+    dependsOn:
+      - brutalist
+      - indie-game
+      - cyberpunk
+      # - luxury-editorial
+      # - retro-90s
+      # - swiss-modernist
+    timeout: 300
+
+# Handover settings - inter-stage communication
+handover:
+  directory: pipeline-runs/front-end-parallel-example
+
+# Execution settings
+execution:
+  failureStrategy: continue           # Keep going even if one design agent fails
+
+# Notification configuration (optional)
+notifications:
+  enabled: true                       # Enable local notifications for hook runs

package/dist/cli/templates/pipelines/loop-example.yml

@@ -0,0 +1,25 @@
+# Loop Pipeline Example
+# Demonstrates pipeline-level looping configuration
+# Docs: https://github.com/FRE-Studios/agent-pipeline
+
+name: loop-example
+trigger: manual
+
+agents:
+  - name: socratic-explorer
+    agent: .agent-pipeline/agents/socratic-explorer.md
+    timeout: 300
+
+looping:
+  enabled: true
+  maxIterations: 5
+
+# Handover settings - inter-stage communication
+handover:
+  directory: pipeline-runs/loop-example
+
+# Notification configuration (optional)
+notifications:
+  enabled: true
+
+

package/dist/cli/templates/pipelines/post-commit-example.yml

@@ -1,67 +1,58 @@
+# Post-Commit Review Pipeline
+# Automated code review triggered after each commit
+# Docs: https://github.com/FRE-Studios/agent-pipeline
+
 name: post-commit-example
 trigger: post-commit
 
-# Settings for pipeline execution
-settings:
-  autoCommit: true
-  commitPrefix: "[pipeline:{{stage}}]"
-  failureStrategy: continue
-  preserveWorkingTree: false
-  executionMode: parallel
+# Agent stages
+agents:
+  - name: memory-doc-updater
+    agent: .agent-pipeline/agents/memory-doc-updater.md
+    timeout: 300
 
-  # Context reduction for multi-stage pipelines
-  contextReduction:
-    enabled: true
-    maxTokens: 50000              # Token limit for context
-    strategy: summary-based       # summary-based or agent-based
-    contextWindow: 3              # Number of recent stages to include in full
-    requireSummary: true          # Require summary field from agents
-    saveVerboseOutputs: true      # Save full outputs to .agent-pipeline/outputs/
-    compressFileList: true        # Compress changed files list
+  # # Example Flow - Remove comments to run full pipeline
 
-# Git workflow configuration (optional)
-# git:
-#   baseBranch: main
-#   branchStrategy: reusable
-#   pullRequest:
-#     autoCreate: true
-#     title: "🤖 Automated Code Review - {{pipelineName}}"
-#     reviewers: []
-#     labels:
-#       - automated
-#       - code-review
+  # # Uncomment the full flow below AND delete standalone doc-updater for comprehensive review.
 
-# Notification configuration (optional)
-# notifications:
-#   enabled: true
-#   events:
-#     - pipeline.completed
-#     - pipeline.failed
-#   channels:
-#     local:
-#       enabled: true
-#       sound: true
+  # # Warning - Pipelines may use many tokens, use with caution.
 
-# Agent stages
-agents:
-  # Sequential execution: code-review → quality-check → doc-updater
-  # Each stage builds on the previous stage's results
-  - name: code-review
-    agent: .claude/agents/code-reviewer.md
-    timeout: 300
-    outputs:
-      - issues_found
-      - severity_level
+  # # Sequential execution: code-review -> quality-check -> doc-updater
+  # # Each stage builds on the previous stage's results
+  # - name: code-review
+  #   agent: .agent-pipeline/agents/code-reviewer.md
+  #   timeout: 900                    # Hard limit per agent run (seconds)
+
+  # - name: quality-check
+  #   agent: .agent-pipeline/agents/quality-checker.md
+  #   dependsOn:
+  #     - code-review
+  #   timeout: 900                    # Hard limit per agent run (seconds)
+  #   onFail: warn                    # Non-blocking: log but keep running
+
+  # - name: memory-doc-updater
+  #   agent: .agent-pipeline/agents/memory-doc-updater.md
+  #   dependsOn:
+  #     - quality-check
+  #   timeout: 300                    # Hard limit per agent run (seconds)
+
+# Git workflow configuration - all git-related settings
+git:
+  autoCommit: true                    # Enable when agents should commit fixes after each stage
+  commitPrefix: "[pipeline:{{stage}}]"  # Standardized prefix for easy commit filtering
+  baseBranch: main                    # PR base and merge target for the pipeline branch
+  branchStrategy: unique-and-delete   # Self-cleaning: auto-removes worktree on success, preserves on failure for debugging
+  branchPrefix: review-pipeline       # Keeps pipeline branches easy to find/clean up
+  mergeStrategy: local-merge          # Options: 'pull-request' (push + PR), 'local-merge' (merge locally), 'none' (keep on branch)
+  # pullRequest:
+    # title: "Post Commit Automations" 
+    # reviewers: []                   # Optional GitHub usernames to request review from
+    # labels:                         # Optional PR labels (must exist in GitHub repo first)
+    #   - automated
+    #   - code-review
+
+# Notification configuration (optional)
+notifications:
+  enabled: true                       # Enable local notifications for hook runs
 
-  - name: quality-check
-    agent: .claude/agents/quality-checker.md
-    dependsOn:
-      - code-review
-    timeout: 300
-    onFail: warn
 
-  - name: doc-updater
-    agent: .claude/agents/doc-updater.md
-    dependsOn:
-      - quality-check
-    timeout: 300