fix(pi-permission-system): bash 命令枚举下降进入控制流与赋值载荷

for/while/if/case/{ }/函数体此前只按整段文本匹配，
"for d in ...; do n=$(find ...); done" 中的 find 绕过了 find* deny（真实会话复现）；
x=$(grep ...)、export X=$(find ...)、! grep ... 同样绕过。

控制流语句仍整段判定，并逐条枚举条件与体内命令；do/else/case 分支容器只下降不单独成单元；
其余语句兜底额外枚举其中的 $(…)。结果只会更严，hard guard 共享同一枚举。

Generated with [Devin](https://devin.ai)

Co-Authored-By: Devin <158243242+devin-ai-integration[bot]@users.noreply.github.com>

pi-permission-system/CHANGELOG.md

@@ -9,6 +9,7 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 
 ### Changed
 
+- **Descend into bash control-flow bodies when enumerating command units.** `for`/`while`/`until` loops, `if`/`elif`/`else` branches, `case` items, `{ … }` brace groups, function definitions, and `!`-negated commands were previously matched only as whole-statement text, so a denied command hidden in a loop body or behind `!` (e.g. `for d in */; do find "$d" …; done`, `! rm -rf x`) bypassed every bash pattern — as did a substitution inside a value (`x=$(grep foo f)`, `export X=$(find .)`). The enumerator still emits each enclosing statement as a unit and now also emits every condition, body, and negated command — plus substitutions hosted in value lists, case subjects, assignments, declarations, and other non-command positions — so the most restrictive decision wins; pure body containers (`do …; done`, `elif`/`else` clauses, `case` items) no longer produce their own units. Hard guards share the same enumeration and now reach those nested commands too.
 - **Replace the AutoReview text-JSON chat reviewer with a conversational tool-call reviewer.** The reviewer makes one independent model call per ask — a fixed review prompt, the rendered transcript, and a bounded JSON description of the planned action — and reaches its verdict only by calling one of two tools, `approve` or `reject`, each carrying a `reason`. A reply with no tool call counts as no verdict; multiple calls, unknown tools, and missing reasons count as invalid responses; neither is retried. An experimental TypeSafe Jev classifier rewrite explored in the working tree was dropped before release and never shipped.
 - **The review model follows the session by default.** With `autoReview.provider` and `autoReview.model` both unset, the reviewer calls the session's current model (`ctx.model`), resolved live per ask so a mid-session switch applies to the next request. Setting both pins a fixed reviewer; setting exactly one is a misconfiguration that warns once at `session_start` and defers every ask.
 - **A rejection defers instead of denying.** `reject` maps to `defer` and the model's `reason` is shown in the human permission prompt, labelled as model-generated. Every review failure — timeout, provider error, malformed or truncated reply, unresolved model — also defers with a fixed system-authored note. Pressing Esc during a review aborts it and denies outright, so no human dialog appears after the user already refused.