dcdpr / jp / 19031308849

58%

Build Type

push

github

Committed by web-flow

Commit Message

feat(cli, config): Add Git-style interactive prompts with turn-level answer persistence (#307)

Replace simple `yes`/`no` confirmations with Git-style inline select
prompts that provide users with more runtime control over tool
execution. When tools request user input, users can now choose between
multiple options presented inline (e.g., `[y,n,e,r,?]`), similar to `git
add -p`.

For tool execution confirmations, users can now:

- `y` - Run the tool as-is
- `n` - Skip execution with default message
- `e` - Edit tool arguments before running
- `r` - Skip with custom reasoning message

For boolean tool questions (e.g., "overwrite existing file?"), users can
additionally choose to remember their answer for the duration of the
turn (`Y`/`N`), automatically applying the same answer to subsequent
identical questions within the same LLM interaction.

This eliminates the need to reconfigure tools at configuration time for
different interaction scenarios. Users maintain full runtime control
while automated workflows can pre-configure answers via the new
`tool.answers` configuration field.

The implementation introduces a new `jp_inquire` crate that wraps the
upstream `inquire` library with Git-style prompt patterns. It also adds
`TurnState` to track persisted answers across tool calls within a single
turn, and extends `ToolConfig` with an `answers` field for
configuration-level automation.

Tool execution flow is refactored into `prepare_run()` and
`prepare_result()` methods for clearer separation of concerns, and new
`RunMode::Never` and `ResultMode::Never` variants enable explicit
rejection workflows with custom reasoning.

Related: #199

---------

Signed-off-by: Jean Mertz <git@jeanmertz.com>

Run Details

47 of 345 new or added lines in 7 files covered. (13.62%)

5 existing lines in 4 files now uncovered.

7699 of 16008 relevant lines covered (48.09%)

14.5 hits per line