Skip to main content

Core Concepts

Understanding the fundamental concepts behind ACE.

Playbooks

A playbook is a structured set of instructions that guides an AI agent on how to perform a specific task. Think of it as a detailed standard operating procedure for your AI.

Anatomy of a Playbook

Playbooks use the ACE bullet format—structured instructions optimized for AI agents and evolution tracking.

Code Review Assistant

Guidelines for reviewing code quality, security, and best practices.

STRATEGIES & INSIGHTS

  • [check-context-first] helpful=5 harmful=0 :: Read the PR description and linked issues before reviewing code to understand the intent and scope of changes.
  • [security-mindset] helpful=4 harmful=0 :: Look for common vulnerabilities: SQL injection, XSS, hardcoded secrets, and improper input validation.

COMMON MISTAKES TO AVOID

  • [avoid-nitpicking] helpful=3 harmful=0 :: Focus on substantive issues over style preferences. Save formatting debates for linter configuration.
  • [explain-why] helpful=3 harmful=0 :: Don't just say "this is wrong"—explain why and suggest a better approach.

PROBLEM-SOLVING HEURISTICS

  • [test-coverage] helpful=2 harmful=0 :: Check if new code paths have corresponding tests, especially for edge cases and error handling.

Bullet Format

Each instruction follows this structure:

[semantic-slug] helpful=N harmful=N :: Actionable instruction

See Creating Playbooks for the full format reference and examples.

Playbook Properties

PropertyDescription
idUnique identifier (UUID)
nameHuman-readable name
descriptionBrief summary of purpose
statusPlaybook status: active, paused, or archived
sourceOrigin: starter, user_created, or imported
current_versionReference to the active version
created_atCreation timestamp
updated_atLast modification timestamp

Versions

Every playbook maintains a version history. Versions are created when:

  1. You manually edit the playbook
  2. The system evolves the playbook based on outcomes

Version Properties

PropertyDescription
version_numberSequential version identifier
contentPlaybook content at this version (Markdown)
bullet_countNumber of ACE-format bullets in this version
diff_summaryDescription of what changed
created_by_job_idEvolution job ID if created by evolution (null for manual edits)
created_atWhen this version was created

Viewing Version History

From the dashboard, click on a playbook and navigate to the Versions tab to:

  • See all historical versions
  • View evolution summaries
  • Compare versions to see what changed

Outcomes

An outcome is a record of how a playbook performed on a specific task. Outcomes are the fuel for playbook evolution.

Recording Effective Outcomes

Good outcomes include:

{
"playbook_id": "abc-123",
"task_description": "Reviewed authentication refactor PR with 500+ line changes",
"outcome": "success",
"notes": "Identified race condition in token refresh logic. Suggested using mutex.",
"reasoning_trace": "Analyzed auth flow, found concurrent access issue..."
}

Outcome Fields

FieldRequiredDescription
playbook_idYesWhich playbook was used
task_descriptionYesWhat task was performed
outcomeYesResult: "success", "partial", or "failure"
notesNoAdditional context or feedback
reasoning_traceNoAgent's reasoning process

Outcome Values

  • success - Task completed correctly
  • partial - Task completed but with issues
  • failure - Task did not complete or was incorrect
tip

Include detailed notes even for successful outcomes. They help the evolution process understand why something worked well.

Evolution

Evolution is the process by which ACE improves playbooks based on accumulated outcomes.

How Evolution Works

Outcomes → Reflector → Insights → Curator → New Version
  1. Collect Outcomes - System gathers unprocessed outcomes
  2. Reflect - Reflector agent analyzes patterns and issues
  3. Generate Insights - Identifies what's working and what isn't
  4. Curate - Curator agent drafts improved playbook version
  5. Publish - New version becomes active

Evolution Triggers

Evolution happens:

  • Automatically - After threshold outcomes are recorded (default: 5)
  • Manually - When you trigger it from the dashboard or MCP

Evolution Status

StatusDescription
queuedWaiting to start
runningCurrently processing
completedSuccessfully created new version
failedError occurred during evolution

API Keys

API keys authenticate your MCP tool access to ACE.

Scopes

Each key has specific permissions:

ScopeAllows
playbooks:readRead playbook content and metadata
playbooks:writeCreate and update playbooks
outcomes:readRead task outcomes
outcomes:writeRecord task outcomes
evolution:readRead evolution job status
evolution:writeTrigger playbook evolution
*Full access to all operations

Key Types

  • Full Access - All scopes (for development/testing)
  • Read Only - Only read scopes (for production agents)
  • Custom - Choose specific scopes
warning

Never expose API keys with write scopes in client-side code or public repositories.

MCP (Model Context Protocol)

MCP is a protocol for connecting AI agents to external tools and data sources. ACE provides an MCP server that exposes playbooks as tools.

Available MCP Tools

ToolDescription
list_playbooksList all your playbooks
get_playbookGet content of a specific playbook
create_playbookCreate a new playbook
create_versionCreate a new version of a playbook
record_outcomeRecord a task outcome
trigger_evolutionManually trigger evolution
get_evolution_statusCheck evolution job status

MCP Endpoint

https://aceagent.io/mcp/sse

Subscriptions & Usage

See Billing & Subscriptions for plan details and pricing.

Next Steps

Now that you understand the core concepts: