RunTrim
How it works

RunTrim controls the run lifecycle around your agent.

RunTrim is the control, verification and recovery layer for AI coding agents. It enforces scope before edits, verifies changes after edits, and gives you safe recovery and continuation paths.

Install CLIRead FAQ

Five-step run flow

Step 01

Start a guarded run

runtrim start

RunTrim prepares project memory, agent instructions, and protocol state before edits start.

Step 02

Create scoped contract

runtrim agent "your task" --copy

The run gets a contract with objective, allowed scope, and stop rules so the agent does not drift silently.

Step 03

Agent works inside scope

agent executes task

Your agent applies changes inside the contract boundaries. Scope expansion requires explicit approval.

Step 04

Finish verification

runtrim finish

RunTrim checks changed files, risky paths, and proof gaps, then returns PASS, WARN, or BLOCKED.

Step 05

Restore or continue

runtrim restore last --preview

If needed, restore locally or continue from a generated continuation prompt without starting cold.

Finish verdicts

PASS

Run evidence and scope checks passed. Safe to review and proceed.

WARN

Non-blocking risk found. Proceed only after explicit human review.

BLOCKED

Needs review and is not trusted yet. Approve scoped amendment or restore.

What stays local by default

  • Source code stays local by default.
  • Restore apply happens locally.
  • Env file contents are not uploaded.
  • Cloud sync (paid plans) is metadata-oriented.

Why this is not Git or an orchestrator

  • Git tracks version history; RunTrim governs AI run behavior.
  • Orchestrators route models; RunTrim adds contracts, verification, and recovery.
  • RunTrim works with Claude, Codex, Cursor, ChatGPT, DeepSeek, and others.

Typical outcomes after guarded runs

  • Less repeated context loading across sessions.
  • Fewer off-scope edits that require cleanup.
  • Faster recovery when a run needs rollback or continuation.