---
title: Autonomous Execution and Context Handoff
status: active
type: plan
archived: false
archived_on:
created: 2026-09-06
updated: 2026-09-11
owner: mixed
related_prs: []
related_issues: []
---

# Learnitive — Autonomous Execution and Context Handoff

## Summary

This is the durable execution rule for substantial Learnitive work. It is intentionally state-independent: project-specific progress belongs in `PROJECT_STATE_LEDGER.md`, while a copy-ready cross-chat continuation belongs in `NEXT_CHAT_CONTINUATION_PROMPT.md`.

## Execution contract

Continue from the current verified source state and execute all remaining safe phases autonomously, sequentially and continuously. Do not restart completed phases unless verification exposes a concrete dependency on them. Do not pause for routine confirmation.

For each phase:

**Execute → Verify → Identify problems → Fix → Verify again → Continue**

A phase is complete only when its objectives and reasonable success criteria are verified.

## Decision authority

Make reasonable independent decisions about architecture, implementation, research, prioritisation, UX/workflow, technical choices, documentation, testing, optimisation and error correction. Preserve established project decisions unless new evidence requires a change. When information is missing, make the most reasonable assumption unless it genuinely cannot be inferred and materially affects the objective.

Ask the user only when required information cannot be inferred, an external permission/credential is required, an irreversible/high-impact action needs explicit approval, or continuing would create a substantial risk of pursuing the wrong objective.

## State tracking

During substantial work, keep `PROJECT_STATE_LEDGER.md` current with:

- source authority and current working state;
- completed phases;
- current phase and exact stopping point;
- remaining phases in order;
- decisions and assumptions;
- requirements and dependencies;
- tests and verification;
- problems and fixes;
- artifacts/files produced; and
- outstanding risks or environment blockers.

## Context-limit handoff

Proactively monitor usable conversation context throughout execution. Before usable chat context becomes a practical limitation — and before response quality or state recall begins to degrade:

1. Finish the current safe subtask where possible.
2. Update the state ledger with the complete exact state.
3. Update `NEXT_CHAT_CONTINUATION_PROMPT.md` so a new ChatGPT instance can continue without the previous conversation.
4. Include the project objective, completed phases, current phase/stopping point, all remaining phases, important decisions/assumptions/requirements/dependencies, known problems and attempted fixes, verification results, relevant files/artifacts, and the autonomous execution/quality/human-interaction rules.
5. Preserve this context-handoff rule in the next prompt.

The continuation prompt must end exactly with:

> Continue autonomously from this exact state. Do not restart completed phases. Do not ask for confirmation. Execute all remaining phases sequentially, verify each phase, fix problems, and continue until the project is complete. If this chat again approaches its context limit, generate another complete continuation prompt for the next chat before stopping.

## Completion rule

At final completion report completed work, artifacts, testing/verification, remaining limitations or risks, any decisions requiring human attention, and recommended next steps. Never represent blocked dependency, browser, staging or production verification as completed.