autonomous-loops

Chain non-interactive claude -p calls in a shell script, one per focused step such as implement, cleanup, verify, and commit. Each call gets a fresh context window, and set -e stops the pipeline on failure.

A sequential pipeline runs fixed claude -p steps locally with no git automation. The Continuous Claude loop creates branches, opens PRs, waits for CI checks, auto-fixes failures, and merges, with limits via max-runs, max-cost, or max-duration.

Use a SHARED_TASK_NOTES.md file that Claude reads at the start of each iteration and updates at the end. This bridges the gap between independent claude -p invocations that otherwise start with fresh context.

Negative instructions like "don't test type systems" make the model hesitant about all testing and degrade quality unpredictably. A separate cleanup pass removes redundant tests and over-defensive checks while preserving business logic coverage.

Use it when a written spec decomposes into multiple interdependent work units needing parallel implementation with merge coordination. For single focused changes or quick iteration, a sequential pipeline or NanoClaw REPL is sufficient.

Always set an exit condition: --max-runs for iteration count, --max-cost for spending, --max-duration for elapsed time, or a completion signal phrase that stops the loop after consecutive signals indicate the work is done.