fix: verified findings from the analytics-plan research pass (consent boundary, soul fingerprint, dispatch exit status, doc counts)

.github/PULL_REQUEST_TEMPLATE.md

@@ -20,7 +20,7 @@
 
 - [ ] `npm run typecheck` passes
 - [ ] `npm run build` passes
-- [ ] If touching agent loop / tool / sandbox: added or updated a regression test under `test/`
+- [ ] If touching agent loop / tool / sandbox: added or updated a regression test co-located with the source (`src/**/*.test.ts` — this is what `npm test` runs; a `test/` dir would not be picked up)
 - [ ] If touching `src/soul/*` or `src/prompt.ts`: linked to a discussion issue (these change who every Lisa becomes)
 - [ ] Commit message is lowercase, imperative, no emoji
 - [ ] No `any` introduced (or comment explaining why)

README.md

@@ -210,7 +210,7 @@ which owns that auth. See [Coding plans](#coding-plans--use-a-subscription-inste
 
 LISA is also a control plane for the *other* coding agents on your machine. Three layers, increasing in how much she touches them:
 
-**1. Observe.** She watches the agents already running and tells you what you'd otherwise miss — a session stuck on the same error, two agents about to collide in one repo, a finished run sitting idle. Honest scope note: **all five observers (Claude Code, Codex, OpenCode, Aider, GitHub PRs) emit structural activity — tools, files touched, last command, errors — gated behind a per-integration `visibility` tier; fidelity varies by what each agent records on disk** (Claude Code is richest; Aider's markdown logs give files + turn counts but no tool stream; every adapter has a privacy test asserting prompts/replies/file-contents never leak). `lisa agents` prints a one-shot snapshot; the island shows it live. She can `dispatch_agent` headlessly (refusing directories another agent owns), `compare_agents` on the same task in parallel worktrees, and surface **advisor cards** — each with a one-click action that prefills the chat (nothing auto-runs) and a ✕ that teaches her to stop nagging about that category.
+**1. Observe.** She watches the agents already running and tells you what you'd otherwise miss — a session stuck on the same error, two agents about to collide in one repo, a finished run sitting idle. Honest scope note: **all ten observers — five coding-agent adapters (Claude Code, Codex, OpenCode, Aider, GitHub PRs) plus git, shell, takoapi, managed and pty — emit structural activity — tools, files touched, last command, errors — gated behind a per-integration `visibility` tier; fidelity varies by what each agent records on disk** (Claude Code is richest; Aider's markdown logs give files + turn counts but no tool stream; every adapter has a privacy test asserting prompts/replies/file-contents never leak). Three are enabled by default — Claude Code, managed and pty — and the other seven are opt-in per integration in `~/.lisa/agents.json`. `lisa agents` prints a one-shot snapshot; the island shows it live. She can `dispatch_agent` headlessly (refusing directories another agent owns), `compare_agents` on the same task in parallel worktrees, and surface **advisor cards** — each with a one-click action that prefills the chat (nothing auto-runs) and a ✕ that teaches her to stop nagging about that category.
 
 **2. Control her own agents.** A **managed agent** runs LISA's *own* agent loop in a child context she fully drives: delegate a task, approve/deny each mutating tool, send follow-ups, cancel — from the GUI agents card or `POST /api/agents/managed/<id>/{send,approve,cancel}`. These are hers, so the model and provider are hers too.
 
@@ -716,9 +716,9 @@ src/
 ├── idle/                   idle-time autonomous reflection (Reve)
 ├── autonomy/               run ledger — observable journal of self-driven runs (`lisa autonomy`)
 ├── agents/                 managed agents (LISA's own loop) + PTY agents (drive real claude/codex)
-├── integrations/           observers: claude-code · codex · opencode · aider · github-pr · pty · managed · …
+├── integrations/           observers (10): claude-code · codex · opencode · aider · github-pr · git · shell · takoapi · managed · pty
 ├── orchestrator/           cross-agent journal + "while you were away" recap synthesis
-├── advisor/                proactive advisor cards (stuck / conflict / ready / idle) + dismissal learning
+├── advisor/                proactive advisor cards (stuck / conflict / cost_spike / ready / idle) + dismissal learning
 ├── consent/                unified consent gate for ambient signals + mail (default all off)
 ├── control/                remote-control policy — gates high-risk actions from remote callers
 ├── sense/                  ambient signal sources (foreground app / window title), consent-gated

README.zh-CN.md

@@ -203,7 +203,7 @@ OpenAI 模型 (`gpt-*`) 还需要 `OPENAI_API_KEY`。
 
 LISA 也是你机器上*其它* coding agent 的一个控制面。三层，对它们的介入逐层加深：
 
-**1. 观察。** 她看着已经在跑的 agent，把你会错过的事告诉你——哪个会话卡在同一个报错上、哪两个要在同一个仓库里打架、哪个早就跑完在干等。诚实地说明范围：**五个 observer（Claude Code、Codex、OpenCode、Aider、GitHub PR）都能产出结构化活动——工具、改动的文件、最近命令、错误——由每个集成的 `visibility` 档位门控；精细度取决于各 agent 在磁盘上记录了什么**（Claude Code 最丰富；Aider 的 markdown 日志只给文件 + 轮次、没有工具流；每个 adapter 都有隐私测试断言提示词/回复/文件内容绝不泄漏）。`lisa agents` 打印一次性快照，灵动岛实时显示。她可以 `dispatch_agent` 无头派发（拒绝把新 agent 丢进已被占用的目录）、`compare_agents` 在并行 worktree 里对比多个 agent 做同一任务，并给出**顾问卡片**——每条带一个一键动作（预填到聊天框，**绝不自动执行**）和一个 ✕（教她少唠叨这一类）。
+**1. 观察。** 她看着已经在跑的 agent，把你会错过的事告诉你——哪个会话卡在同一个报错上、哪两个要在同一个仓库里打架、哪个早就跑完在干等。诚实地说明范围：**十个 observer——五个 coding agent 适配器（Claude Code、Codex、OpenCode、Aider、GitHub PR），外加 git、shell、takoapi、managed、pty——都能产出结构化活动——工具、改动的文件、最近命令、错误——由每个集成的 `visibility` 档位门控；精细度取决于各 agent 在磁盘上记录了什么**（Claude Code 最丰富；Aider 的 markdown 日志只给文件 + 轮次、没有工具流；每个 adapter 都有隐私测试断言提示词/回复/文件内容绝不泄漏）。默认开启的是三个——Claude Code、managed、pty——其余七个需要在 `~/.lisa/agents.json` 里逐个开启。`lisa agents` 打印一次性快照，灵动岛实时显示。她可以 `dispatch_agent` 无头派发（拒绝把新 agent 丢进已被占用的目录）、`compare_agents` 在并行 worktree 里对比多个 agent 做同一任务，并给出**顾问卡片**——每条带一个一键动作（预填到聊天框，**绝不自动执行**）和一个 ✕（教她少唠叨这一类）。
 
 **2. 控制她自己的 agent。** **managed agent** 跑的是 LISA *自己*的 agent loop，在一个她完全驱动的子上下文里：派发任务、逐个审批/拒绝改写类工具、追加追问、取消——从 GUI 的 agents 卡片或 `POST /api/agents/managed/<id>/{send,approve,cancel}`。它们是她的，所以用的模型和 provider 也是她的。
 
@@ -664,9 +664,9 @@ src/
 ├── idle/                   空闲自主反思（梦境 Reve）
 ├── autonomy/               run ledger —— 自驱运行的可观察日志（`lisa autonomy`）
 ├── agents/                 managed agent（LISA 自己的 loop）+ PTY agent（操纵真实 claude/codex）
-├── integrations/           observer：claude-code · codex · opencode · aider · github-pr · pty · managed · …
+├── integrations/           observer（10 个）：claude-code · codex · opencode · aider · github-pr · git · shell · takoapi · managed · pty
 ├── orchestrator/           跨 agent 日志 + "你不在的时候"回顾合成
-├── advisor/                主动顾问卡片（卡住 / 冲突 / 就绪 / 空闲）+ 关闭学习
+├── advisor/                主动顾问卡片（卡住 / 冲突 / 成本飙升 / 就绪 / 空闲）+ 关闭学习
 ├── consent/                环境信号 + 邮箱的统一授权门控（默认全关）
 ├── control/                远程控制策略 —— 对远程调用者门控高危动作
 ├── sense/                  环境信号源（前台 app / 窗口标题），授权门控