Persistence Modes - Agent Teams Lite

Overview

Agent Teams Lite is storage-agnostic. Artifacts can be persisted in three different modes:

engram (recommended default) — Persistent, repo-clean storage via Engram
openspec (file-based, optional) — File artifacts in project directory
none (ephemeral, no persistence) — No writes, artifacts exist only in conversation

The workflow engine is completely pluggable. All 9 sub-agent skills reference shared convention files to ensure consistent behavior across persistence modes.

Mode Resolution

The orchestrator passes artifact_store.mode with one of: engram | openspec | none.

Default Resolution Policy

When the orchestrator does not explicitly set a mode:

If Engram is available → use engram
Otherwise → use none

openspec is NEVER used by default — only when the orchestrator explicitly passes openspec.

When falling back to none, the system recommends the user enable engram or openspec for better results.

Quick Modes by Use Case

Recommended: Engram Mode

# Agent-team storage policy
artifact_store:
  mode: engram      # Recommended: persistent, repo-clean

Best for:

Production workflows
Team collaboration
Long-running feature development
Audit trails and lineage tracking

Privacy: None Mode

# Privacy/local-only (no persistence)
artifact_store:
  mode: none

Best for:

Privacy-sensitive projects
Exploratory work
Quick prototypes
When you don’t want any artifacts written

File-based: OpenSpec Mode

# File artifacts in project (OpenSpec flow)
artifact_store:
  mode: openspec

Best for:

When you explicitly want file artifacts
Projects that prefer file-based specs
When Engram is not available and you need persistence
Audit trails in version control

Behavior Per Mode

Mode	Read from	Write to	Project files
`engram`	Engram (see Engram Convention)	Engram	Never
`openspec`	Filesystem (see OpenSpec Convention)	Filesystem	Yes
`none`	Orchestrator prompt context	Nowhere	Never

Common Rules

Mode-Specific Rules

Engram Mode

DO NOT write any project files
Persist all artifacts to Engram
Return observation IDs in result envelope
Use deterministic naming: sdd/{change-name}/{artifact-type}
Always use two-step recovery: mem_search → mem_get_observation

OpenSpec Mode

Write files ONLY to paths defined in openspec-convention.md
Create openspec/ directory structure
Update tasks.md in-place with [x] marks
Follow archive structure: YYYY-MM-DD-{change-name}/

None Mode

DO NOT create or modify any project files
Return results inline only
All artifacts exist only in conversation context
No persistence between sessions

NEVER force openspec/ creation unless the orchestrator explicitly passed openspec mode. If you are unsure which mode to use, default to none.

Engram Mode Deep Dive

Deterministic Artifact Naming

All SDD artifacts persisted to Engram MUST follow this naming convention:

title:     sdd/{change-name}/{artifact-type}
topic_key: sdd/{change-name}/{artifact-type}
type:      architecture
project:   {detected or current project name}
scope:     project

Artifact Types

Artifact Type	Produced By	Description
`explore`	sdd-explore	Exploration analysis
`proposal`	sdd-propose	Change proposal
`spec`	sdd-spec	Delta specifications (all domains concatenated)
`design`	sdd-design	Technical design
`tasks`	sdd-tasks	Task breakdown
`apply-progress`	sdd-apply	Implementation progress (one per batch)
`verify-report`	sdd-verify	Verification report
`archive-report`	sdd-archive	Archive closure with lineage

Exception: sdd-init uses sdd-init/{project-name} as both title and topic_key (it’s project-scoped, not change-scoped).

Example: Writing to Engram

mem_save(
  title: "sdd/add-dark-mode/proposal",
  topic_key: "sdd/add-dark-mode/proposal",
  type: "architecture",
  project: "my-app",
  content: "# Proposal: Add Dark Mode\n\n..."
)

Two-Step Recovery Protocol

MANDATORY: To retrieve an artifact, ALWAYS use this two-step process.

Step 1: Search by topic_key pattern
  mem_search(query: "sdd/{change-name}/{artifact-type}", project: "{project}")
  → Returns a truncated preview with an observation ID

Step 2: Get full content (REQUIRED)
  mem_get_observation(id: {observation-id from step 1})
  → Returns complete, untruncated content

NEVER use mem_search results directly as the full artifact — they are truncated previews. ALWAYS call mem_get_observation to get the complete content.

Retrieving Multiple Artifacts

When a skill needs multiple artifacts (e.g., sdd-tasks needs proposal + spec + design):

// Step 1: Search for all required artifacts
mem_search(query: "sdd/{change-name}/proposal", project: "{project}") → get ID
mem_search(query: "sdd/{change-name}/spec", project: "{project}") → get ID
mem_search(query: "sdd/{change-name}/design", project: "{project}") → get ID

// Step 2: Get full content for EACH
mem_get_observation(id) for EACH → full content

Loading Project Context

mem_search(query: "sdd-init/{project}", project: "{project}") → get ID
mem_get_observation(id) → full project context

Browsing All Artifacts for a Change

mem_search(query: "sdd/{change-name}/", project: "{project}")
→ Returns all artifacts for that change

Updating Artifacts

When updating an artifact you already retrieved (e.g., marking tasks complete):

// If you have the observation ID
mem_update(
  id: {observation-id},
  content: "{updated full content}"
)

// For upserts (Engram deduplicates by topic_key)
mem_save(
  title: "sdd/{change-name}/{artifact-type}",
  topic_key: "sdd/{change-name}/{artifact-type}",  // Same topic_key
  type: "architecture",
  project: "{project}",
  content: "{updated content}"
)

Why This Convention Exists

Deterministic titles → recovery works by exact match, not fuzzy search
topic_key → enables upserts (updating same artifact without creating duplicates)
sdd/ prefix → namespaces all SDD artifacts away from other Engram observations
Two-step recovery → mem_search previews are always truncated; mem_get_observation is the only way to get full content
Lineage → archive-report includes all observation IDs for complete traceability

OpenSpec Mode Deep Dive

Directory Structure

When openspec mode is enabled, a change produces this structure:

openspec/
├── config.yaml              ← Project-specific SDD config
├── specs/                   ← Source of truth (main specs)
│   └── {domain}/
│       └── spec.md
└── changes/                 ← Active changes
    ├── archive/             ← Completed changes (YYYY-MM-DD-{change-name}/)
    └── {change-name}/       ← Active change folder
        ├── exploration.md   ← (optional) from sdd-explore
        ├── proposal.md      ← from sdd-propose
        ├── specs/           ← from sdd-spec
        │   └── {domain}/
        │       └── spec.md  ← Delta spec
        ├── design.md        ← from sdd-design
        ├── tasks.md         ← from sdd-tasks (updated by sdd-apply)
        └── verify-report.md ← from sdd-verify

Artifact File Paths

Skill	Creates / Reads	Path
sdd-init	Creates	`openspec/config.yaml`, `openspec/specs/`, `openspec/changes/`, `openspec/changes/archive/`
sdd-explore	Creates (optional)	`openspec/changes/{change-name}/exploration.md`
sdd-propose	Creates	`openspec/changes/{change-name}/proposal.md`
sdd-spec	Creates	`openspec/changes/{change-name}/specs/{domain}/spec.md`
sdd-design	Creates	`openspec/changes/{change-name}/design.md`
sdd-tasks	Creates	`openspec/changes/{change-name}/tasks.md`
sdd-apply	Updates	`openspec/changes/{change-name}/tasks.md` (marks `[x]`)
sdd-verify	Creates	`openspec/changes/{change-name}/verify-report.md`
sdd-archive	Moves	`openspec/changes/{change-name}/` → `openspec/changes/archive/YYYY-MM-DD-{change-name}/`
sdd-archive	Updates	`openspec/specs/{domain}/spec.md` (merges deltas into main specs)

Reading Artifacts

Each skill reads its dependencies from the filesystem:

Proposal:  openspec/changes/{change-name}/proposal.md
Specs:     openspec/changes/{change-name}/specs/  (all domain subdirectories)
Design:    openspec/changes/{change-name}/design.md
Tasks:     openspec/changes/{change-name}/tasks.md
Verify:    openspec/changes/{change-name}/verify-report.md
Config:    openspec/config.yaml
Main specs: openspec/specs/{domain}/spec.md

Writing Rules

OpenSpec Writing Guidelines

ALWAYS create the change directory (openspec/changes/{change-name}/) before writing artifacts
If a file already exists, READ it first and UPDATE it (don’t overwrite blindly)
If the change directory already exists with artifacts, the change is being CONTINUED
Use the openspec/config.yaml rules section to apply project-specific constraints per phase

Config File Reference

# openspec/config.yaml
schema: spec-driven

context: |
  Tech stack: {detected}
  Architecture: {detected}
  Testing: {detected}
  Style: {detected}

rules:
  proposal:
    - Include rollback plan for risky changes
  specs:
    - Use Given/When/Then for scenarios
    - Use RFC 2119 keywords (MUST, SHALL, SHOULD, MAY)
  design:
    - Include sequence diagrams for complex flows
    - Document architecture decisions with rationale
  tasks:
    - Group by phase, use hierarchical numbering
    - Keep tasks completable in one session
  apply:
    - Follow existing code patterns
    tdd: false           # Set to true to enable RED-GREEN-REFACTOR
    test_command: ""     # e.g., "npm test", "pytest"
  verify:
    test_command: ""     # Override for verification
    build_command: ""    # Override for build check
    coverage_threshold: 0  # Set > 0 to enable coverage check
  archive:
    - Warn before merging destructive deltas

Archive Structure

When archiving, the change folder moves to:

openspec/changes/archive/YYYY-MM-DD-{change-name}/

Use today’s date in ISO format. The archive is an AUDIT TRAIL — never delete or modify archived changes.

Detail Level Control

The orchestrator may also pass detail_level: concise | standard | deep. This controls output verbosity but does NOT affect what gets persisted — always persist the full artifact.

Shared Convention Files

All skills reference these shared convention files for consistent behavior:

File	Purpose
`persistence-contract.md`	Mode resolution rules — how `engram`, `openspec`, and `none` modes behave, what each mode reads/writes, and the fallback policy
`engram-convention.md`	Deterministic artifact naming (`sdd/{change-name}/{artifact-type}`), two-step recovery protocol (search then get full content), and write/update patterns via `topic_key` upserts
`openspec-convention.md`	Filesystem paths for each artifact, directory structure, config.yaml reference, and archive layout

Why they exist: Previously each skill inlined its own persistence logic (~224 lines of duplication across 9 skills). Now each skill references the shared files for DRY principles and consistent behavior.

Choosing the Right Mode

Engram

Best for most teams

Persistent across sessions
No repo clutter
Full audit trail
Deterministic naming

OpenSpec

When you need files

File-based workflow
Version control integration
Explicit artifact request
Self-contained changes

None

Privacy first

No persistence
Ephemeral artifacts
Quick exploration
No file writes

Next Steps

Complete Workflow

Learn the full SDD workflow

Delta Specs

Understand how delta specs work

​Overview

​Mode Resolution

​Default Resolution Policy

​Quick Modes by Use Case

​Recommended: Engram Mode

​Privacy: None Mode

​File-based: OpenSpec Mode

​Behavior Per Mode

​Common Rules

​Engram Mode

​OpenSpec Mode

​None Mode

​Engram Mode Deep Dive

​Deterministic Artifact Naming

​Artifact Types

​Example: Writing to Engram

​Two-Step Recovery Protocol

​Retrieving Multiple Artifacts

​Loading Project Context

​Browsing All Artifacts for a Change

​Updating Artifacts

​Why This Convention Exists

​OpenSpec Mode Deep Dive

​Directory Structure

​Artifact File Paths

​Reading Artifacts

​Writing Rules

​Config File Reference

​Archive Structure

​Detail Level Control

​Shared Convention Files

​Choosing the Right Mode

Engram

OpenSpec

None

​Next Steps

Complete Workflow

Delta Specs

Overview

Mode Resolution

Default Resolution Policy

Quick Modes by Use Case

Recommended: Engram Mode

Privacy: None Mode

File-based: OpenSpec Mode

Behavior Per Mode

Common Rules

Engram Mode

OpenSpec Mode

None Mode

Engram Mode Deep Dive

Deterministic Artifact Naming

Artifact Types

Example: Writing to Engram

Two-Step Recovery Protocol

Retrieving Multiple Artifacts

Loading Project Context

Browsing All Artifacts for a Change

Updating Artifacts

Why This Convention Exists

OpenSpec Mode Deep Dive

Directory Structure

Artifact File Paths

Reading Artifacts

Writing Rules

Config File Reference

Archive Structure

Detail Level Control

Shared Convention Files

Choosing the Right Mode

Next Steps