Aglet

Investigate an Agent Trajectory Failure

Trajectory investigation needs more than the last failed step. Reconstruct the state before each decision, locate where the actual path diverged, and test whether the next failure was inherited. Preserve tool returns and recovery attempts so the first cause does not disappear.

Build a useful investigation brief

  1. Rebuild the path

    Collect prompt, task state, observations, tools, arguments, returns, decisions, retries, side effects, and final result in order. Record versions and timestamps. Mark any inferred transition instead of presenting it as observed.

  2. Find the first divergence

    Compare actual and acceptable trajectories at each state boundary. Identify the first missing observation, invalid call, stale assumption, skipped check, or unnecessary loop. Trace its downstream effects without treating every later symptom as a separate cause.

  3. Replay the boundary

    Rerun with one state, tool return, or instruction changed at a time. Test recovery and partial-completion cases. State whether the failure persists, moves, or disappears, and identify the smallest evidence gap that blocks attribution.

What to carry forward

The investigation brief should preserve ordered states, first divergence, downstream effects, replay, and evidence gaps. End with a localized trajectory fault or bounded uncertainty. Do not assign blame to the final step when an earlier unobserved state caused it.

Technical background: Agent evaluation research on arXiv.

Keep the decision with the work.

Use a Work Item in Aglet to record the problem, the evidence you have, and the next decision. Add an owner and priority, then keep updates in the discussion so the next person can follow the reasoning.

Create an account See the product workflow