Skip to content

Troubleshooting

Troubleshooting

If a check fails or an operation stops, you can give your agent the result and ask it to investigate. discerndiscernA tool that installs and runs an agent development practice in a project.'s results describe the current state and the next action; these pages help you understand the recovery and recognize when a decision needs you.

Work through this result and explain what needs attention. Continue the repair you can make within this task, and bring me any decision that would change what we agreed to build or check.

Find the symptom

§
What you're seeing Where to go
Setup stopped, or a coding tool cannot connect. Setup and integrations
A build, test, or other project check failed. Fix a red gate
The gateGateThe configured checks a change must satisfy for ordinary completion. refuses to run, or it passes without usable Proof. Gate and Proof
A standardStandardA held limit for a repeatable project measurement. or checkpointCheckpointA review question presented when a relevant kind of change occurs. needs attention. Gate and Proof
A task's workspace, resources, or cleanup look wrong. Worktrees and resources
A session ended before the task finished. Recover an interrupted task
Tools disappeared, a long wait ended, or a page will not open. MCP, terminal, and docs
discern crashed, or you found local files you don't recognize. Crashes and local state
You want to upgrade or remove discern. Maintain or remove discern

Help your agent find the cause

§

Keep the original result available. It can contain the failing command, the location of captured output, and a way to reproduce the problem. That evidence gives your agent a starting point without repeating the entire operation.

If the session has lost track of the task, ask the agent to inspect discern status in the task's existing workspace. Status is read-only and reports what is true now. For an installation problem, discern doctor checks the configuration, commands, and agent connections without changing them.

The agent should explain what the evidence establishes and what remains uncertain. A successful recovery is something you can recognize: the tool connects, the operation completes, or the repaired change returns with fresh Proof.

When the next step needs you

§

A repair may reveal a choice about the project. For example, a feature may exceed a size limit you want to preserve, or an agent may find that a review questionQuestionSomething the agent is asked to judge about the project or a change. cannot be satisfied by the current change. Ask it to show the evidence, the available options, and its recommendation before you decide.

Removing a check or weakening a limit changes what future work must satisfy. It needs a decision about the practicePracticeThe connected way of working discern installs and the project carries between sessions., rather than being treated as an ordinary way to clear a failure. Cleanup also deserves care when it would discard work; use the operation's preview and recovery instructions to understand what will remain.

If the documented recovery still fails, keep the result and report the problem. It is useful evidence for finding the cause.

In this section

§
  1. Setup and integrationsResume an interrupted setup, fix installation problems, or activate discern in your coding agent.
  2. Gate and ProofFind why checks stopped or Proof is missing, repair the cause, and know when a decision is needed.
  3. Worktrees and resourcesRecover an interrupted task, resolve cleanup problems, and check its local services and environment settings.
  4. MCP, terminal, and docsRestore missing agent tools, continue a wait, find a documentation page, or read a result when an interface fails.
  5. Crashes and local stateRecover after a crash, preserve a useful report, and identify local files before removing them.
choose openEsc close