Cursor agent retry loop

Check whether your coding agent is stuck in retry mode.

CAUM Loop Check gives a local structural preflight for edit, search, terminal, and test sequences from coding agents. It looks for repeated patterns, not private code content.

local preflight zero upload observe-only not a truth score

The problem

When an agent keeps trying to repair the same issue, each step may look reasonable in isolation. The costly pattern appears across the whole run: same tool family, same error state, same retry loop.

Use CAUM when the question is: is this workflow still making structural progress, or is it repeating effort?

Useful event fields

You do not need prompts, source code, documents, customer messages, or private payloads. Structural fields are enough for a first check.

  • search/read/edit/test events
  • status or error marker
  • phase label
  • state_id or hashed issue signature
  • tokens/cost if available

Example shape

[{"event":"tool_call","tool":"search","phase":"inspect","status":"ok","state_id":"same_symbol"},
 {"event":"tool_call","tool":"edit","phase":"fix","status":"ok","state_id":"same_file"},
 {"event":"tool_call","tool":"test","phase":"verify","status":"error","state_id":"same_error"}]

What the free check returns

  • events analyzed
  • exact cycle coverage
  • structural pattern cycle coverage
  • work-yield proxy
  • burn proxy
  • trace quality

Run the free local check first.

Paste a trace into CAUM Loop Check. If it shows repeated structure, run a full receipt or put the recurring workflow into CAUM Live.

CAUM observes structure only. It does not judge correctness, read private content, block agents, or promise savings.