Bash Hooks

Kent post-processes shell command output before it is shown to the model to normalize output, reduce command noise, and add useful execution context.

Config

Configure command post-processing under [shell] in ~/.kent/config.toml:

[shell]
postprocessing_mode = "all" # none | builtin | user | all
postprocess_hook = "~/.kent/shell_postprocess_hook"

Omit postprocess_hook when no hook is configured.

postprocessing_mode

Allowed values:

• none: disable command post-processing.
• builtin: run Kent’s output cleanup and built-in processing.
• user: run Kent’s output cleanup, then your configured hook.
• all: run Kent’s output cleanup, built-in processing, then your configured hook.

In builtin, user, and all, Kent’s final model-visible command-output pass limits each line to 1,000 Unicode code points; oversized lines keep only their prefix and end with … [N characters omitted], where N is exact. This runs after user-hook replacement; none bypasses the limit, and Kent operational warnings are not command-output lines.

Protocol

Input

Kent sends JSON like:

{
  "tool_name": "exec_command",
  "command": "go test ./...",
  "parsed_args": ["go", "test", "./..."],
  "command_name": "go",
  "workdir": "/abs/workdir",
  "original_output": "...sanitized command output...",
  "current_output": "...built-in processed output or original output...",
  "exit_code": 0,
  "backgrounded": false,
  "max_display_chars": 16000
}

Your hook receives both:

• original_output: sanitized command output before built-in processing
• current_output: command output after built-in processing, or original_output when unchanged

Hook must return JSON like:

{
  "processed": true,
  "replaced_output": "...new output..."
}

Return {"processed": false} for no-op passthrough. If the hook is missing, times out, exits nonzero, or returns invalid JSON, Kent falls back to the current output and reports a warning.