Diasync Memory

DiaSync Memory

End State (Begin With The End)

At any moment, the active agent instance should be able to recover and communicate:

• current goal and current stage,
• active decisions and commitments,
• unresolved conflicts and risks,
• and the next concrete action,

while memory remains auditable, convergent, and healthy without step-by-step human intervention.

diasync-memory uses a hyphenated name to comply with the Agent Skills naming specification.

First Principles

• Concurrency is normal, not exceptional.
• Write correctness is stricter than read convenience.
• Autonomy is preferred over manual prompting.
• History is append-only; corrections are explicit (supersedes).
• Soft policies guide behavior; avoid rigid hard-coded orchestration.
• Manage both diachronic and synchronic complexity continuously.

Complexity Model

• Diachronic complexity (over time): continuity and drift control across sessions, interruptions, and long-running work.
• Synchronic complexity (same-time concurrency): contention and divergence control across multiple active instances.

Activation Signals (Soft, Not Hard Logic)

High-confidence activation signals:

• user asks to continue prior work, handoff, or "remember context",
• work spans many steps or sessions,
• multiple agents/instances work in parallel,
• quality/risk/governance of memory is requested.

Medium-confidence activation signals:

• major decisions, commitments, or constraints appear,
• planning requires stable shared context,
• unresolved collisions or ambiguity appear.

Low-confidence activation signals:

• brief task with no durable value and no collaboration/concurrency.

Bootstrap

Use any command and let the script auto-initialize .memory, or initialize explicitly:

python .opencode/skills/diasync-memory/scripts/memoryctl.py init --root .memory

Proactive Baseline Loop

Default autonomous cadence (adapt as needed):

1. sync start
2. attach
3. recall before planning/answering
4. capture during meaningful execution updates
5. distill at milestones or scope shifts
6. publish + reduce for cross-instance knowledge
7. lease + reconcile for contested keys
8. diagnose then optimize periodically
9. checkpoint during long sessions
10. handoff and sync stop at session end

Memory Debt Prioritization

When time is constrained, prioritize by highest observed memory debt:

• convergence debt (published but unreduced, duplicate active decision keys),
• contention debt (unresolved conflicts, stale leases),
• continuity debt (missing attach/checkpoint/handoff freshness),
• governance debt (stale findings, low health trend),
• retrieval debt (stale indexes or poor capsule quality).

Choose actions that reduce the largest debt first.

Autonomy Guardrails

• Never rewrite historical ledger lines.
• Prefer explicit conflict visibility over silent overwrite.
• Keep uncertainty explicit (confidence, assumptions, evidence).
• Use --dry-run when safety is unclear.
• Preserve machine-readable outputs and integrity contracts.

Activation Router

• Session start or resume: use references/ATTACH_AND_SYNC.md
• During execution updates: use references/CAPTURE_AND_DISTILL.md
• Cross-instance knowledge sharing: use references/PUBLISH_AND_REDUCE.md
• Contention on shared decision keys: use references/LEASE_AND_RECONCILE.md
• Long-session anti-drift: use references/CHECKPOINT_AND_HANDOFF.md
• Recall before planning/answering: use references/RECALL_PROTOCOL.md
• Memory health checks and self-improvement: use references/GOVERNANCE_LOOP.md
• Autonomous cadence policy: use references/PROACTIVE_CADENCE.md
• Memory debt triage policy: use references/MEMORY_DEBT.md
• Diachronic/synchronic radar: use references/COMPLEXITY_RADAR.md

Command Path

All tool commands are executed through:

python .opencode/skills/diasync-memory/scripts/memoryctl.py <command> ...

Converted and distributed by TomeVault — claim your Tome and manage your conversions.