Agent Handoffs

Goal

Explain how AletheIA treats handoffs between agents as compact restartable transitions between work-slice boundaries.

Core rule

Inter-agent handoffs should behave as:

restart packages, not transcripts

They should preserve:

• work-slice boundary clarity
• validation continuity
• next-action explicitness
• low ambiguity for the next boundary

Handoff in the broader model

A handoff sits between:

• a current Work Slice
• a possible related Work Item in an external coordination system
• the current and next Execution Surface

The handoff should preserve what the next boundary needs without pretending that the previous runtime surface is durable memory.

Minimum restart package

A useful operational handoff should usually make these things explicit:

1. current status
2. what was completed
3. what remains pending
4. the next action
5. the main risks
6. the minimum validation expectation
7. the related work-slice or work-item refs when they matter

Richness rule

Keep three layers distinct:

1. conceptual richness — the full meaning the boundary may need
2. template richness — optional fields that reduce drift
3. schema coverage — the minimum structured continuity record

The schema may stay smaller than the template. That is a feature, not a bug.

When the minimum package is enough

Prefer the minimum restart package when:

• the next action is obvious
• the receiving boundary is narrow
• the main risk is execution, not reinterpretation
• validation continuity fits in one compact line

When to step up to the richer template

Use the richer operational template when one or more of these are true:

• the work crosses a meaningful ownership boundary
• file scope must be frozen explicitly
• semantic guardrails matter more than raw file paths
• acceptance criteria need to be preserved to avoid reinterpretation
• the receiving boundary should report back in a specific format

Compact and restartable still beats exhaustive and noisy, but the richer template is justified when it materially reduces drift.

Multi-boundary continuity

If work will cross more than one meaningful boundary, prefer:

a chain of compact handoffs

instead of one oversized artifact.

A healthy flow still looks like:

boundary A -> restart package -> boundary B -> restart package -> boundary C

Model-agnostic requirement

A good handoff should still preserve its meaning whether the next executor is:

• Codex
• Claude Code
• another coding agent
• a future workflow using a different provider

Runtime-local examples may differ. The handoff meaning should not.