Troubleshooting
Start with:Agent executable was not found
Check the launch plan:PATH. Supported
names are codex, claude, claudecode, and aider.
An agent launches but its commands are not intercepted
This is expected with the current limited agent runner.termyte run <agent> prepares local policy, logs, memory, and session context,
but does not guarantee subprocess interception. Use termyte shell to evaluate
the experimental governed shell, and verify it first with termyte doctor.
A policy file is rejected
Review the effective policy and error:version: 1;- known preset names;
- rules with
name, optionaldescription,action, andmatch; - matcher groups
semantic_ids,commands, andpaths.
A plain-English policy is unsupported
The natural-language compiler supports fixed templates, not arbitrary language. Use:YAML policy and runtime policy disagree
They are separate policy surfaces.policycommands manage built-in/global/repository YAML used bycheck.policiescommands manage SQLite semantic lists used byrun --andshell.
Check logs are empty
Onlytermyte check writes .termyte/logs.jsonl.
policy test, logs, memory, and the experimental SQLite runtime do not
write stable check events. Use termyte replay for the runtime ledger.
A command was unexpectedly allowed
Unknown or obfuscated command forms can fall back to a generic shell action.- Inspect it with
termyte check "<command>" --json. - Add a global or repository policy rule.
- Test the rule with
termyte policy test. - Do not rely on Termyte as a sandbox for untrusted commands.
A shim blocks because the guard is unavailable
The governed shell fails closed when an intercepted command cannot reach the local guard. Exit the failed session, runtermyte doctor, and start a new
termyte shell. Do not bypass the block until you understand why the guard
failed.