Lint Feedback Loop

chimera.core.loops.lint_feedback runs a linter after each agent turn and feeds any errors back as context for the agent to fix. Inspired by Aider’s lint-and-fix workflow where lint_edited feeds errors back as a reflected_message.

Key Classes

Class	Description
LintFeedbackLoop	Wraps an inner loop with post-turn linting and iterative fixing

Quick Start

from chimera.core.loops.lint_feedback import LintFeedbackLoop

loop = LintFeedbackLoop(
    linter="ruff",
    lint_args=["check", "--no-fix"],
    max_lint_rounds=3,
)

result = loop.run(provider, tools, context, env)
print(f"Lint rounds: {len(loop.lint_history)}")

How It Works

1. The inner loop runs the original task to completion.
2. The configured linter runs on the workspace.
3. If lint errors are found, the agent gets a new context with the errors and instructions to fix them without changing functionality.
4. Steps 2-3 repeat up to max_lint_rounds times or until the linter reports no errors.

Custom Linter

Any linter that writes errors to stdout/stderr works:

# Use flake8 instead of ruff
loop = LintFeedbackLoop(linter="flake8", lint_args=["--max-line-length=100"])

# Use eslint for JavaScript projects
loop = LintFeedbackLoop(linter="eslint", lint_args=["src/"])

Inspecting Lint History

After run() completes, lint_history contains the raw output from each lint round:

for i, output in enumerate(loop.lint_history):
    if output.strip():
        print(f"Round {i+1}: {len(output.splitlines())} lines of errors")
    else:
        print(f"Round {i+1}: clean")

Import Reference

from chimera.core.loops.lint_feedback import LintFeedbackLoop

Related

• Loops — overview of all loop types
• Retry Loop — retry wrapper with scoring
• Plan/Act Loop — two-phase plan then execute