Amistio

Session guide

Tool sessions without context drift

AI coding tools often keep conversation or task context. Reusing that context can save time when work is related, but it can also pollute a new task. Amistio treats session reuse as a safety and quality decision.

What a tool session is

A tool session is the local AI tool's working context for a piece of work. Depending on the provider, that might mean a conversation, an agent run, a resumable task, a model-specific thread, or a command invocation with related state.

The goal

Keep related work together when it helps, and start fresh when old context could make the tool overconfident, distracted, or expensive.

When continuing helps

Same workstream

A follow-up fix, verification pass, or polish request for the same ADR, plan, branch, and repository can benefit from remembered context.

Same runner boundary

The same paired runner, machine, repository link, and requester make reuse easier to reason about.

Same acceptance target

If the acceptance checks and files are still centered on the same task, continuing can reduce repeated exploration.

Provider supports it

Some tools can safely resume sessions; others work better as one-shot commands. The runner should respect the provider's real capability.

When fresh is safer

  • The new work belongs to a different feature, ADR, repository, branch, or requester.
  • The previous session included failed assumptions or stale architecture context.
  • The task changes from planning to implementation without an approved handoff.
  • A team workspace would cross member, runner, or machine boundaries.
  • The provider cannot expose enough session metadata for safe continuation.
  • The user explicitly wants a clean run.

Amistio's conservative policy

Amistio tracks provider-agnostic tool-session metadata so the local runner can choose whether to continue, create a new session, or avoid session reuse. The default should be conservative: related work can continue when the boundary is clear, and unrelated or ambiguous work starts fresh.

Review first

Generated project-brain artifacts should be reviewed before implementation work becomes claimable.

No cross-member reuse

Team work should not reuse sessions across different members, runners, machines, or unrelated workstreams by default.

Provider details stay local

Provider-specific session identifiers and credentials should stay behind local runner adapters, not in repository files.

User override matters

When a user knows the task history better than automation does, explicit session choices should take priority.