Coding agents

AI coding loops

AI coding loops are repo-safe agent cycles for repeated coding work: inspect, patch, test, review, and stop with evidence.

Abstract developer workbench showing inspect patch test and review loop

Answer-first definition

An AI coding loop gives a coding agent durable task state, an isolated workspace when needed, allowed actions, a verifier such as tests or CI, and an approval boundary before merge or production change.

Common coding loop types

Loop type	Verifier	Safe output
CI Failure Sweeper	CI job passes or failing log is summarized	Patch or report
PR Babysitter	PR checks and reviewer comments are resolved	Updated PR, not auto-merge
Flaky Test Stabilizer	Repeated test runs prove stability	Patch plus evidence log
Maker-Checker Review	Separate agent or human reviews diff against rubric	Review notes or gated approval
Dependency Update Loop	Tests, lockfile diff, vulnerability status	Approval-gated PR

Details

Relevant loops

Loop	Category	Difficulty	Cadence	Verification
API Contract Drift	Engineering	Intermediate	After API changes or SDK releases	Server behavior, client types, examples, and docs agree on request/response contracts.
Acceptance Scenario Lockstep	Engineering	Intermediate	Before and during ambiguous feature work	The same scenarios written before implementation pass after the change, and any scope expansion is explicitly approved.
Agent Instructions After-Action	Engineering	Beginner	After a successful or painful agent coding session	Repo instructions contain only reusable, source-grounded lessons and the next similar task can start without rediscovering the same trap.
Architecture Rubric Refactor	Engineering	Advanced	When architecture work has a defined scope	Scoped module meets the written rubric, tests pass, and unresolved objections are explicit.
Behavior Ladder TDD	Engineering	Intermediate	When implementing logic-heavy features	Each behavior test fails before implementation, passes after the smallest change, and remains green through final refactor.
CI Optimization	Engineering	Advanced	Monthly or when CI is painful	CI p50/p95 improves against the same workflow without weakening tests or hiding failures.
Claude Code Repo Readiness	Engineering	Beginner	Before major agent work	Repo has agent instructions, documented commands, architecture notes, risk areas, and a docs/loops scaffold.
Cold Load Trim	Engineering	Advanced	When first visit feels heavy	Initial screen downloads fewer bytes while screenshots and behavior remain unchanged.
Completion Promise Loop	Engineering	Intermediate	For scoped implementation tasks where half-finished output is the main risk	Every acceptance criterion is checked with tests, browser evidence, logs, screenshots, or a clear blocker report before the agent stops.
Fresh Clone Onboarding	Engineering	Intermediate	Before onboarding	A clean machine reaches the documented ready state using only the README.
Parallel Agent Worktree Sweep	Engineering	Advanced	When several independent repo improvements can run at once	Each agent branch has isolated scope, passing checks, a summary, and no conflicting files before integration review.
Project Docs Freshness	Engineering	Beginner	Nightly or after meaningful code changes	Changed behavior, APIs, CLI commands, config, and workflows are reflected in docs. Docs checks pass.
Reference Oracle Implementation	Engineering	Advanced	Before implementing tricky behavior with an external source of truth	Generated outputs match the reference oracle across the agreed fixture set, with tolerances documented for legitimate differences.
Spec to Task Shards	Engineering	Intermediate	Before multi-file feature work or migrations	A written spec, non-goals, acceptance checks, and ordered task shards exist before implementation begins.
Test Flake Stabilizer	Engineering	Intermediate	When tests are inconsistent	The repaired test and full suite pass for the required consecutive-run streak.
Test and Logging Coverage	Engineering	Intermediate	Weekly or before release	Critical flows have useful tests and structured logs for representative success and failure paths.
Trace-First Debugging	Engineering	Intermediate	When an agent is tempted to patch a bug from a hunch	The bug is reproduced, the root cause is evidenced by traces or tests, and the fix includes a regression check.
Adversarial PR Review	Evaluation	Advanced	For meaningful PRs	An independent critic approves the unchanged version or only accepted findings remain.
Agent Merge Queue Review	Evaluation	Advanced	After multiple agent-generated PRs or branches accumulate	Only branches with passing checks, clear intent, non-conflicting scope, and human-readable evidence are merged or promoted.
Browser Quality Streak	Evaluation	Intermediate	Before release	N realistic scenarios pass consecutively, and earlier failures have regression coverage.
Dependency CVE Burndown	Security	Advanced	After security scan	No exploitable high or critical CVE remains without an explicit risk decision.
Sandboxed YOLO Probe	Security	Advanced	Before allowing autonomous shell-heavy agent runs	The agent can run needed commands inside the sandbox, cannot reach forbidden files/secrets, and produces a replayable diff or report before host-side changes.