Skip to main content

​
Overview

Workflows are slash commands that trigger predefined sequences of actions. They’re the backbone of Athena’s session management and deep reasoning capabilities.
Workflows live in .agent/workflows/ as Markdown files. Each workflow is a reusable pattern that can be invoked with a simple slash command.

​
Quick Reference

CommandPurposeComplexity
/startBoot session, load identityLow
/endClose session, commit to memoryLow
/tutorialGuided first-session walkthroughLow
/saveMid-session checkpointLow

​
Session Management

​
/start - Boot Sequence

Purpose: Beginning of every session. Loads identity, recalls context from previous sessions.
  1. Load Core_Identity.md (~8K tokens)
  2. Load Memory Bank files (~10K tokens total)
  3. Recall last session summary
  4. Initialize session log
  5. Print active context and current mandates

​
/end - Session Close

Purpose: End of every session. Commits insights to memory, updates indexes.
1

Finalize Session Log

Write key decisions and insights to session log file
2

Update Memory Bank

Update activeContext.md with session outcomes
3

Harvest Check

Scan for reusable patterns worth documenting
4

Rebuild Indexes

Regenerate TAG_INDEX.md and PROTOCOL_SUMMARIES.md
5

Git Commit

Commit session changes to version control

​
/save - Mid-Session Checkpoint

When to Use: Mid-session when you want to checkpoint without closing. Use before risky experiments.
Always checkpoint before:
  • Destructive operations
  • Large refactors
  • Experimental changes
  • Architecture modifications

​
Reasoning Workflows

​
Depth Comparison

WorkflowDepthToken BudgetUse Case
NormalStandard~500 tokensMost queries
/thinkHigh~2000 tokensImportant decisions, complex problems
/ultrathinkMaximum~5000+ tokensLife-altering decisions, multi-stakeholder analysis
Rule: Default to normal mode. Escalate to /think for $10K+ decisions. Use /ultrathink only for maximum-depth analysis.

​
/think - Deep Reasoning

Phases:
  1. Context Loading: Pull relevant protocols and case studies
  2. Deep Analysis: Multi-perspective reasoning
  3. Synthesis: Structured recommendations
  4. Quality Check: Verify against constraints

​
/ultrathink - Maximum Depth

When to Use: Maximum-depth analysis for critical decisions.
Combine /think /search /research for nuclear-level investigation:
  • DeepCode: Planning and structured analysis
  • VectorRAG: Historical context from 850+ documents
  • DSPy: Optimized reasoning chains
  • Web Search: Real-time external validation
  • Multi-Source Research: Exhaustive investigation

​
Planning Workflows

​
/brief - Pre-Prompt Clarification

Purpose: Clarify requirements before complex tasks to reduce wasted tokens. Variants:
  • /brief — Core brief (default)
  • /brief ++ — Expanded fields for complex work
  • /brief build — Technical implementation tasks
  • /brief research — Investigation/analysis tasks
  • /brief interview — Iterative Q&A to extract requirements
Example: 
/brief build a dashboard for tracking trading performance

​
/plan - Structured Task Planning

For “Heavy” tasks (new features, refactors, architecture changes):
1

Enter PLANNING mode

Activate structured planning workflow
2

Generate implementation plan

Create detailed plan with pre-mortem analysis
3

Review with user

Present plan for approval before execution
4

Track progress

Maintain progress in task.md

​
Research Workflows

​
Search Depth Comparison

WorkflowSearchesSources ReadDepth
/search2-30-2Medium
/research5-10+3-10+Maximum
Purpose: Fast web search with citations for fact-checking and current information. Output:
  • 2-3 search queries executed
  • Top results summarized
  • Citations included

​
/research - Exhaustive Investigation

Purpose: Multi-source investigation for comprehensive analysis. Process:
  1. Generate 5-10 search queries
  2. Read 3-10+ full sources
  3. Cross-reference findings
  4. Synthesize comprehensive report
  5. Include full bibliography

​
Maintenance Workflows

​
/refactor - Full Workspace Optimization

Purpose: Maintain workspace integrity through periodic optimization. Phases:
1

Diagnostics

Scan workspace for issues (orphans, broken links, duplicates)
2

Pre-remediation Checkpoint

Create backup before modifications
3

Fix Orphans & Broken Links

Repair or remove orphaned files and broken references
4

Optimization Pass

Consolidate duplicates, clean up tech debt
5

Supabase Sync

Re-index all documents in VectorRAG
6

Cache Refresh

Clear and rebuild internal caches
7

Index Regeneration

Rebuild TAG_INDEX and PROTOCOL_SUMMARIES
8

Commit

Commit all changes with detailed log
Frequency: Run monthly or after major architecture changes.

​
/vibe - Vibe Coding Mode

Philosophy: Ship at 70% confidence, iterate based on feedback. Principles:
  • No over-engineering
  • Rapid iteration
  • User feedback drives refinement
  • Progress over perfection
SEO Checklist (mandatory before deploy):
  • Keyword in H1/Title
  • Internal linking
  • Descriptive URL slug
  • Meta description

​
/deploy - Public Repo Sync

Purpose: Sanitized public repository sync (removes private data). Safety Checks:
  1. Scan for credentials and secrets
  2. Remove personal information
  3. Clean sensitive paths
  4. Validate public-safe content
  5. Sync to public repo

​
Creating Custom Workflows

Workflows live in .agent/workflows/ as Markdown files: 
---
description: Short description for the workflow index
created: YYYY-MM-DD
last_updated: YYYY-MM-DD
---

# /command-name — Title

## Behavior

What happens when invoked...

## Phases

Step-by-step execution...

## Tagging

#workflow #automation

​
Conventions

  • Use // turbo annotation above steps that are safe to auto-run
  • Reference other workflows with relative links
  • Include rollback instructions for destructive operations
  • Add clear exit conditions
  • Document expected token usage

​
Best Practices

Choose the Right Depth

Don’t /ultrathink on simple queries. Reserve deep reasoning for important decisions.

Checkpoint Often

Use /save before risky experiments. Prevention beats recovery.

Brief First

For complex tasks, /brief reduces wasted tokens and improves output quality.

End Sessions Properly

/end commits insights to long-term memory. Never skip it.

Combine Strategically

/think /search /research for maximum coverage on critical decisions.

Maintain Regularly

Run /refactor monthly to keep workspace healthy.

​
Next Steps

Architecture

Understand the overall system design

Memory System

Learn how memory persists across sessions

Protocols

Explore reusable thinking patterns

Build docs developers (and LLMs) love

Get started for freeTalk to us