feat(mode): ask_user_question joins the plan allowlist; the section steers to the exit tool

Live-session feedback (a real Zed elicitation round-trip): the model
presented its finished plan as a plain reply and asked the USER to
switch modes — the exact reversal the roadmap warns about — because the
shipped section's 'present it with the exit_plan_mode tool' read as a
suggestion. The section now says a finished plan is delivered by
calling exit_plan_mode, preferred over pasting it as a plain reply or
asking the user to switch modes — firmer, without imperatives.

ask_user_question enters the shipped plan allowlist (asking is
read-only-safe), and the section points a blocked decision at it. The
plan-acp-agent example composes the bash family (default mode only —
plan's allowlist keeps excluding it, so the two modes now demo a real
difference) plus tool-ask-user; both recorded scenarios re-recorded:
the pin now shows plan = [ask_user_question, exit_plan_mode, read,
todo_write] and post-exit default = the full eight-tool surface.
This commit is contained in:
kingwl
2026-07-10 15:05:26 +08:00
parent 14f7976e3d
commit 02e0756b78
12 changed files with 1127 additions and 1029 deletions
+1 -1
View File
@@ -4,7 +4,7 @@ The coding agent as an ACP server with **session modes** composed — the live c
## What it demonstrates
`session/new` advertises the mode picker (`default` / `plan`); the editor's `session/set_mode` switches the session, applied at the next turn boundary. In plan mode the model sees only the read-only allowlist (`read`, `todo_write`, `exit_plan_mode` here — this tree loads no web tools) plus the plan-mode guidance section, and every call outside the allowlist is denied at `tools/pre-execute` with a reason that steers it back to planning. The model leaves by presenting its plan through `exit_plan_mode`: the plan markdown renders as the tool's call card, the review question arrives as an elicitation form (approve / keep planning, free text welcome), and a keep-planning answer returns the feedback to the model verbatim.
`session/new` advertises the mode picker (`default` / `plan`); the editor's `session/set_mode` switches the session, applied at the next turn boundary. The default mode carries the full composition — `bash`, `read`/`write`/`edit`, `todo_write`, `ask_user_question` — while plan mode narrows it to the read-only allowlist (`read`, `todo_write`, `ask_user_question`, `exit_plan_mode` here — this tree loads no web tools) plus the plan-mode guidance section; every call outside the allowlist is denied at `tools/pre-execute` with a reason that steers it back to planning, and a blocking decision goes to the user through `ask_user_question`. The model leaves by presenting its plan through `exit_plan_mode`: the plan markdown renders as the tool's call card, the review question arrives as an elicitation form (approve / keep planning, free text welcome), and a keep-planning answer returns the feedback to the model verbatim.
## Run