Skip to main content

Documentation Index

Fetch the complete documentation index at: https://mintlify.com/lnardev/opencode-config-agent/llms.txt

Use this file to discover all available pages before exploring further.

When you invoke /sdd-new, /sdd-ff, or /sdd-continue for the first time in a session, the orchestrator asks you two configuration questions before running anything: which execution mode you want, and which artifact storage backend to use. Your answers are cached for the entire session — you won’t be asked again unless you explicitly request a change.

​
Execution Modes

The execution mode controls whether the orchestrator pauses between phases or runs everything back-to-back.
ModeBehaviorBest For
Interactive (default)Pauses after each phase, shows the result summary, asks “Continue?” before proceedingReviewing and steering each step; first-time use of SDD; high-stakes changes
AutomaticRuns all phases back-to-back without pausing; shows the final result onlySpeed and trust; familiar problem domains; iterating quickly
If you don’t specify a mode, the orchestrator defaults to Interactive — it’s the safer choice because it gives you control at every decision point.

​
Interactive Mode in Detail

In Interactive mode, between every two phases the orchestrator:
1

Shows a concise summary

Presents what the completed phase produced — key decisions, scope, or artifacts written.
2

Lists what the next phase will do

Describes the next phase and what sub-agent will run, so you know what to expect.
3

Asks to continue

Prompts: “¿Continuamos? / Continue?”You can respond with:
  • YES or continue → proceed to the next phase
  • NO or stop → halt the pipeline; you can resume later with /sdd-continue
  • Specific feedback → the orchestrator incorporates your feedback before launching the next phase
4

Incorporates feedback (if given)

If you provided feedback, the orchestrator adjusts the next sub-agent’s prompt to reflect your direction before delegating.

​
Automatic Mode in Detail

In Automatic mode, all phases run via sub-agents back-to-back. The orchestrator delegates each phase as soon as the previous one completes, without pausing. You see only the final combined result. This is equivalent to running /sdd-ff for the planning phases, but extends all the way through apply, verify, and archive if you invoke /sdd-continue in automatic mode.
Execution mode is cached for the session. If you want to switch mid-session, tell the orchestrator: “switch to automatic mode” or “switch to interactive mode.”

​
Artifact Storage Modes

The artifact store determines where SDD artifacts (proposals, specs, designs, tasks, apply-progress, verify reports, archive reports) are persisted.
ModeDescriptionBest For
engram (default)Fast, no files created. Artifacts stored in a local SQLite database via the engram MCP server. Cross-session. Re-running a phase overwrites the previous version — no history.Solo work, quick iteration
openspecFile-based. Creates an openspec/ directory with a full artifact trail. Committable and shareable. Full git history of every artifact version.Teams, PRs, audit trail
hybridBoth backends simultaneously — engram for cross-session recovery, files for team sharing. Higher token cost per operation (writes to two backends).Cross-session recovery + team sharing
noneReturns results inline only. No persistence. All artifacts lost when the conversation ends. Recommend enabling engram or openspec.One-off exploration only
If you don’t specify an artifact store, the orchestrator auto-detects: if engram is available (the MCP server is running), it defaults to engram. Otherwise, it defaults to none and explains the limitation.

​
Engram Mode Notes

No iteration history. Re-running a phase (e.g., /sdd-spec a second time) overwrites the previous artifact. Only the latest version is retained in engram. If you need a full audit trail or want to compare iterations, use openspec or hybrid instead.Not shareable. Engram is a local database — team members cannot see your SDD artifacts. For team projects, use openspec or hybrid.

​
OpenSpec Directory Structure

When using openspec or hybrid mode, sdd-init creates: 
openspec/
├── config.yaml          ← Project config: stack, conventions, strict_tdd, testing capabilities
├── specs/               ← Source of truth; populated by sdd-archive when changes close
└── changes/             ← Active changes
    └── archive/         ← Completed changes (date-prefixed)

​
Model Assignments

The orchestrator reads a model assignment table at session start, caches it, and passes the appropriate model alias in every sub-agent delegation via the model parameter. This ensures each phase uses the right model for its cognitive load.
PhaseDefault ModelReason
orchestratoropusCoordinates, makes decisions
sdd-exploresonnetStructural reading — not architectural
sdd-proposeopusArchitectural decisions
sdd-specsonnetStructured writing
sdd-designopusArchitecture decisions
sdd-taskssonnetMechanical breakdown
sdd-applysonnetImplementation
sdd-verifysonnetValidation against spec
sdd-archivehaikuCopy and close
default (non-SDD)sonnetNon-SDD general delegation
If the assigned model is not available in your current OpenCode session, the orchestrator automatically substitutes sonnet and continues without interrupting the workflow.

​
Configuration Summary

When prompted:
  Execution mode: interactive
  Artifact store: engram
The mode choices you make at the start of a session apply to all SDD commands for the rest of that session. If you start a new OpenCode session, the orchestrator will ask again.

Build docs developers (and LLMs) love

Get started for freeTalk to us