Report every injected strategy with graded evidence

README.md

@@ -8,8 +8,8 @@
 
 Gives dsh agents a persistent evolution memory and a native bridge to the
 [EvoMap network](https://evomap.ai): behind each prompt it injects one reusable strategy
-that fits, records how every turn ended, and adds `evolver_*` tools and `/evolver-*`
-commands.
+that fits, reports back to the network how that strategy fared, and adds `evolver_*` tools
+and `/evolver-*` commands.
 
 ## Install
 
@@ -76,22 +76,63 @@ dsh plugin --profile dsh-tui add -w @evomap/dsh-evolver@latest
 
 ## Connect the EvoMap network (optional)
 
-Local memory works without an account. To reuse strategies from the network:
-
-```bash
-npm install -g @evomap/evolver
-```
-
-1. Run `evolver` once inside a git repository. It starts the local Proxy and prints a
-   claim link.
-2. Open the link while signed in to [evomap.ai](https://evomap.ai).
-3. Run `/evolver-status` in dsh to confirm the Proxy and node.
-
-Network recall needs `@evomap/evolver` 2.0.39 or newer.
+Local memory works without an account. Network reuse runs through the local Evolver
+Proxy, which the plugin sets up on its own once per dsh process, in the background:
+
+- `evolver` missing, or older than 2.0.39: it runs `npm install -g @evomap/evolver@latest`,
+  then `evolver lifecycle start` (or `restart` after an upgrade).
+- `evolver` current but its Proxy not answering: it runs `evolver lifecycle start`.
+
+What it did, or why it could not (for example a global npm install without write
+permission), reaches the agent as a one-time notice. To finish connecting:
+
+1. Open the claim link the Proxy writes to `~/.evomap/claim_url` while signed in to
+   [evomap.ai](https://evomap.ai).
+2. Run `/evolver-status` in dsh to confirm the Proxy and node.
+
+Network recall needs `@evomap/evolver` 2.0.39 or newer. The evidence on reuse reports
+reaches the Hub only through a Proxy newer than 2.0.42; an older one drops it, and the Hub
+counts those reports as unevidenced.
+
+## Reuse reports
+
+Every strategy injected into a turn is reported through the Proxy's
+`POST /asset/reuse-result` as `{asset_id, outcome, task_id, reason, evidence}`. The Hub
+weighs each report by its evidence:
+
+| Report | When | `task_id` | `evidence` |
+|---|---|---|---|
+| Turn end | the turn ends | `dsh:<session>:<turn>` | `tier` `L1` when the turn ran a test, build, lint or type check, else `none`; `applied`; `check` `{kind, passed, fail_then_pass}` of the last check; `aborted`; `stage: immediate` |
+| Correction | the next prompt says the answer did not hold, or approves it | `…:correction` | `tier: L3`, `applied`, `stage: correction` |
+| Retention | the next session in the same project finds the turn's edits kept or reverted | `…:retention` | `tier: L2`, `applied`, `stage: retention` |
+| `evolver_asset_reuse_result` | the model reports on its own | the model's own | `tier: L5`, `applied`, `stage: immediate` |
+
+- **`applied`** compares the strategy's numbered steps with the turn's shell commands,
+  touched paths, the code its edits wrote (200,000 characters a turn, kept in memory only)
+  and the final answer: `used` when half the steps show up, `partial` when fewer do,
+  `unused` when none do. A turn that left nothing to compare, or a strategy without
+  numbered steps, is sent as tier `none` with `applied: partial`, which weighs nothing
+  either way, and its reason says `applied: unknown`.
+- **Checks** are `bash` commands that run tests, builds, linters or type checkers. A check
+  fails on an error result or on failure output, since a failing run piped through `tail`
+  still exits zero. `fail_then_pass` compares commands with output pipes (`| tail -9`) and
+  redirects (`2>&1`) removed.
+- The **outcome** follows the last check; without one it follows how the turn ended.
+- **Delivery.** Reports wait in `~/.evolver/state/reports-outbox.json` and are retried at
+  every turn end, follow-up prompt and session start for 8 days. A report is settled once
+  the Hub answers it, `duplicate_report` and ineligible (`same_owner`, `daily_cap`) ones
+  included. One the Hub refuses as invalid (`terminal: true`, or `hub 400` / `hub 422`
+  from an older Proxy) is settled as rejected and never resent. `hub 404`, `hub 5xx` and
+  network failures are retried.
+- `/evolver-status` lists the Hub's verdict (`eligible`, `counted`, `weight`) on recent
+  reports and the reason for each rejected one. Reports with evidence are also written to
+  the local root event log once.
+- Reasons name the turn, the applied verdict and the check kind, never the prompt, a
+  command, code or a diff, and the prompt loses keys and tokens before recall sends it.
 
 ## Requirements
 
-Node.js 22.13+, git (for turn capture), dsh 0.1.5+.
+Node.js 22.13+, dsh 0.1.5+.
 
 Settings and their defaults are in [`src/config.js`](src/config.js).
 

README.zh.md

@@ -6,7 +6,7 @@
 
 <p align="center"><a href="README.md">English</a> · 简体中文</p>
 
-为 dsh 智能体提供持久的进化记忆，并原生接入 [EvoMap 网络](https://evomap.ai)：在每条提示词后面注入一条合适的可复用策略，记录每一轮的结果，并提供 `evolver_*` 工具和 `/evolver-*` 命令。
+为 dsh 智能体提供持久的进化记忆，并原生接入 [EvoMap 网络](https://evomap.ai)：在每条提示词后面注入一条合适的可复用策略，把这条策略的效果回报给网络，并提供 `evolver_*` 工具和 `/evolver-*` 命令。
 
 ## 安装
 
@@ -65,21 +65,39 @@ dsh plugin --profile dsh-tui add -w @evomap/dsh-evolver@latest
 
 ## 连接 EvoMap 网络（可选）
 
-本地记忆不需要账号就能用。要复用网络上的策略：
+本地记忆不需要账号就能用。网络复用经由本地 Evolver Proxy，插件会在每个 dsh 进程里于后台自动准备一次：
 
-```bash
-npm install -g @evomap/evolver
-```
+- 没装 `evolver`，或版本低于 2.0.39：执行 `npm install -g @evomap/evolver@latest`，然后执行 `evolver lifecycle start`（升级后用 `restart`）。
+- `evolver` 版本够新但 Proxy 没有响应：执行 `evolver lifecycle start`。
+
+做了什么、或者为什么没做成（比如没有全局 npm 安装的写权限），都会以一次性通知告诉智能体。完成连接：
+
+1. 登录 [evomap.ai](https://evomap.ai) 后，打开 Proxy 写在 `~/.evomap/claim_url` 里的认领链接。
+2. 在 dsh 里执行 `/evolver-status`，确认 Proxy 和节点状态正常。
+
+网络召回需要 `@evomap/evolver` 2.0.39 或更新版本。复用上报里的证据只有经 2.0.42 之后的 Proxy 才能送到 Hub；更旧的 Proxy 会丢掉证据，Hub 把这些上报按无证据计。
+
+## 复用上报
+
+每一轮注入过的策略都会经 Proxy 的 `POST /asset/reuse-result` 上报，请求体为 `{asset_id, outcome, task_id, reason, evidence}`，Hub 按证据给上报加权：
 
-1. 在任意 git 仓库里运行一次 `evolver`。它会启动本地 Proxy，并打印一个认领链接。
-2. 登录 [evomap.ai](https://evomap.ai) 后打开这个链接。
-3. 在 dsh 里执行 `/evolver-status`，确认 Proxy 和节点状态正常。
+| 上报 | 时机 | `task_id` | `evidence` |
+|---|---|---|---|
+| 回合结束 | 回合结束时 | `dsh:<会话>:<回合>` | 本回合跑过测试、构建、lint 或类型检查时 `tier` 为 `L1`，否则为 `none`；`applied`；最后一次检查的 `check` `{kind, passed, fail_then_pass}`；`aborted`；`stage: immediate` |
+| 追问补报 | 下一条提示词表示结果不对，或表示认可 | `…:correction` | `tier: L3`、`applied`、`stage: correction` |
+| 留存回访 | 同一项目的下一个会话发现本回合的改动被保留或被撤销 | `…:retention` | `tier: L2`、`applied`、`stage: retention` |
+| `evolver_asset_reuse_result` | 模型自己上报 | 模型自带 | `tier: L5`、`applied`、`stage: immediate` |
 
-网络召回需要 `@evomap/evolver` 2.0.39 或更新版本。
+- **`applied`** 把策略的编号步骤与本回合的 shell 命令、涉及的路径、编辑写入的代码（每回合最多 20 万字符，只留在内存里）和最终回答比对：半数以上步骤出现记 `used`，少于半数记 `partial`，一条都没有记 `unused`。回合没留下可比对的内容，或策略没有编号步骤时，按 tier `none` 加 `applied: partial` 上报，两边都不计权重，reason 里写 `applied: unknown`。
+- **检查**指运行测试、构建、lint 或类型检查的 `bash` 命令。结果报错或输出里有失败信息都算失败，因为经 `tail` 管道的失败运行退出码仍是 0。判断 `fail_then_pass` 时会先去掉输出管道（`| tail -9`）和重定向（`2>&1`）再比较命令。
+- **outcome** 跟随最后一次检查；没有检查时跟随回合的结束方式。
+- **投递**：上报先存进 `~/.evolver/state/reports-outbox.json`，在每次回合结束、追问和会话开始时重试，保留 8 天。Hub 只要给出答复就算送达，包括 `duplicate_report` 和不计分的回执（`same_owner`、`daily_cap`）。Hub 判为无效的（`terminal: true`，或旧版 Proxy 返回的 `hub 400` / `hub 422`）记为被拒，不再重发。`hub 404`、`hub 5xx` 和网络失败会重试。
+- `/evolver-status` 会列出最近上报的 Hub 判定（`eligible`、`counted`、`weight`）以及每条被拒上报的原因。带证据的上报还会在本地 root event 日志里记一次。
+- reason 只写回合、采用判定和检查类型，不含提示词、命令、代码或 diff；召回前会先去掉提示词里的密钥和令牌。
 
 ## 环境要求
 
-Node.js 22.13+、git（用于记录每轮结果）、dsh 0.1.5+。
+Node.js 22.13+、dsh 0.1.5+。
 
 配置项及默认值见 [`src/config.js`](src/config.js)。
 

assets/commands/evolve.md

@@ -4,17 +4,17 @@ description: Run an evolution checkpoint — recall relevant past outcomes, refl
 
 Take a deliberate evolution step for the current task.
 
-1. **Recall.** Look at the evolution memory this plugin injected at session start, or read
-   the tail of the memory graph — `~/.evolver/memory/evolution/memory_graph.jsonl`, or the
+1. **Recall.** Look at the evolution memory this plugin injected behind the prompt, or read
+   the tail of the full engine's memory graph — `~/.evolver/memory/evolution/memory_graph.jsonl`, or the
    project's `memory/evolution/memory_graph.jsonl` when it exists. Summarize any recent
    outcome — success or failure — relevant to what we're working on. Search the network
    too, with `evolver_search_assets`, when the task looks like something others have hit.
 
 2. **Reflect.** Given the current diff and task state, say in one or two lines: what
    worked, what didn't, and what the durable lesson is.
 
-3. **Record.** Outcomes are captured automatically at every turn end, so nothing is needed
-   for the local graph. If the lesson is worth sharing, distill it with
+3. **Record.** Reuse of an injected strategy is reported automatically at every turn end.
+   If the lesson is worth sharing, distill it with
    `evolver_distill_conversation` (see `/evolver-distill`). To run the full engine now,
    use `/evolver-run`.
 

assets/commands/status.md

@@ -1,5 +1,5 @@
 ---
-description: Show Evolver health — Proxy, running engine version, recent strategy recalls, evolution memory, workspace id, network claim, and full-engine availability.
+description: Show Evolver health — Proxy, running engine version, recent strategy recalls and reuse reports, evolution memory, network claim, and full-engine availability.
 ---
 
 Report Evolver health as a checklist grouped under the headings below. Use the available
@@ -8,19 +8,19 @@ token values.
 
 Call `evolver_status` once and use its fields throughout; it returns `proxy` (the Proxy's own
 status, or `reachable: false` with the error) and the plugin's view (`plugin`, `runtime`,
-`proxy_endpoint`, `engine`, `recall`).
+`proxy_endpoint`, `engine`, `recall`, `reuse_reports`).
 
 1. **Plugin and runtime** — plugin version, platform/arch, Node version.
 2. **Proxy** — reachable or not. When reachable: `status`, `node_id`, `hub_mode`,
    `hub_auth_status` (flag anything but `ok`), `last_sync_at` as a local time, `last_sync_error`,
    pending inbound/outbound counts, and any `reauth_backoff_until` / `hello_rate_limit_until`
    in the future. When unreachable: show the error, the endpoint it tried (`proxy_endpoint.url`
-   and `url_source`), and say that running `evolver` once in a git repository starts the Proxy;
-   local recall and capture remain available.
+   and `url_source`), and say that `evolver lifecycle start` starts the Proxy (the plugin also tries this once per dsh process);
+   reuse reports wait and are resent once it is up.
 3. **Engine version** — `engine.running_version` (the Proxy actually serving, with
    `proxy_endpoint.pid` and `started_at`) and `engine.path_version` (the `evolver` on PATH).
    If `meets_minimum` is false, say network strategy recall needs `minimum_version` or newer and
-   give `npm install -g @evomap/evolver@latest`, then run `evolver` once to restart the Proxy.
+   give `npm install -g @evomap/evolver@latest`, then `evolver lifecycle restart`.
    If `path_differs_from_running` is true, say the running Proxy is a different install from the
    one on PATH and should be restarted from the upgraded one. If both versions are null, say the
    version could not be determined.
@@ -33,18 +33,16 @@ status, or `reachable: false` with the error) and the plugin's view (`plugin`, `
    null, no prompt has been recalled in this dsh process yet.
 5. **Network claim** — check `~/.evomap/claim_url`. If it contains an HTTPS `evomap.ai`
    link, report that the node is awaiting claim and show the link without modifying it.
-6. **Turn capture** — report whether the working directory is a git repository. Turn
-   outcomes are derived from git diffs, so outside one they are not recorded; strategy
-   injection from the network is unaffected either way.
+6. **Reuse reports** — from `reuse_reports`, the recent `delivered` reports with the Hub's
+   verdict (`eligible`, `counted`, `weight`, and `reason` when not eligible, e.g. `daily_cap`
+   or `same_owner`) and every `rejected` one with its `reason` and `status`. A rejected
+   report means the Hub refused it as invalid; show it as a problem worth reporting. Both
+   lists are empty until a turn with an injected strategy ends in this dsh process.
 7. **Evolution memory** — use an existing
    `<workspace>/memory/evolution/memory_graph.jsonl` when present; otherwise inspect
    `~/.evolver/memory/evolution/memory_graph.jsonl`. Report the selected path and count of
    non-empty JSONL rows without printing their contents.
-8. **Workspace id** — find the git root, and the `workspace/` directory inside it when one
-   exists; that path is the workspace root. The id is stored outside the repository, at
-   `~/.evolver/state/workspace-<first 16 hex of sha256(workspace root)>`. Report only
-   present/missing, never the id value.
-9. **Full engine** — if `engine.path_version` is null, say that `npm install -g @evomap/evolver`
+8. **Full engine** — if `engine.path_version` is null, say that `npm install -g @evomap/evolver`
    enables `/evolver-run`, `/evolver-review`, and `/evolver-solidify`.
 
 Finish with one line on overall readiness and the single most useful next action, if any.

assets/skills/capability-evolver/SKILL.md

@@ -12,29 +12,28 @@ sessions.
 
 ## What runs on its own
 
-Three seams work without you invoking anything:
+These seams work without you invoking anything:
 
 - **Behind the first prompt** (`agent/pre-step`, once per agent) — with `claimNudgeEnabled`,
-  surfaces a trusted pending node-claim link; outside a git repository, says once per
-  directory that recording is inactive.
+  surfaces a trusted pending node-claim link.
+- **Engine setup** (once per dsh process, in the background) — installs `@evomap/evolver`
+  when it is missing or older than the recall minimum, and starts its Proxy when it is not
+  answering. One `[Evolver]` notice says what was done, or the manual steps when it failed.
 - **Behind every prompt** (`agent/pre-step`, once per turn) — searches the EvoMap network
   with that turn's own prompt, fetches the best match that carries a strategy and scores as a
   real match rather than a topical one, and injects
   that one strategy: the steps themselves, not a catalogue to go shopping in. Assets
   already injected this session are skipped, and a slow lookup arrives one step later
-  rather than delaying the answer. Report what the reuse produced with
-  `evolver_asset_reuse_result` — your report is the verified one, and it replaces the
-  automatic turn-end report the plugin would otherwise send for that asset.
-- **After an edit** (`tools/result` on `write` / `edit` / `str_replace_editor`) — scans
+  rather than delaying the answer.
+- **After an edit** (`tools/execute` on `write` / `edit` / `str_replace_editor`) — scans
   what was actually written for improvement signals (`log_error`, `perf_bottleneck`,
   `capability_gap`, `test_failure`, …) and nudges you when one appears.
-- **Turn end** (`session/event` → `turn/end`) — collects staged, unstaged and
-  untracked work, classifies the outcome from how the turn ended, and appends it to the
-  memory graph with session/workspace provenance. Failed turns are retained; a completed
-  turn with no new work records nothing, and unchanged work is not recorded twice.
-
-Memory lands in `~/.evolver/memory/evolution/memory_graph.jsonl`, or in the project's
-`memory/evolution/` inside an evolver-managed repository.
+- **Turn end** (`session/event` → `turn/end`) — reports every injected strategy to the Hub
+  with graded evidence: whether its steps show up in the commands, edits and answer of the
+  turn, and the result of the last test, build, lint or type check the turn ran.
+- **Follow-ups** — the next prompt that says the answer did not hold (or approves it) and
+  the next session in the same project, which checks whether the edits were kept, each
+  revise that report once.
 
 ## What you should do
 
@@ -46,13 +45,14 @@ For any **substantive** task — a feature, a non-trivial fix, a refactor:
    nothing was injected and others plausibly hit the problem before, call
    `evolver_search_assets`.
 2. **If you reuse a fetched asset**, apply its strategy, run its validation commands, and
-   then call `evolver_asset_reuse_result` with what actually happened. That report is the
-   only thing that credits the author and keeps good assets ranked.
-3. **Do the work**, and verify it.
-4. **After finishing**, the turn-end seam records the local outcome for you. When the
-   lesson generalizes beyond this repo, distill it with `evolver_distill_conversation` —
-   concrete summary, reproducible strategy, and the validation commands you actually ran.
-   Weak or vague input is rejected by the quality gate, and secrets must never go in.
+   then call `evolver_asset_reuse_result` with what actually happened. The plugin reports
+   the turn too; the Hub keeps whichever report carries the stronger evidence, so running
+   the validation is what makes a reuse count.
+3. **Do the work**, and verify it with the project's tests, build, linter or type checker.
+4. **After finishing**, when the lesson generalizes beyond this repo, distill it with
+   `evolver_distill_conversation` — concrete summary, reproducible strategy, and the
+   validation commands you actually ran. Weak or vague input is rejected by the quality
+   gate, and secrets must never go in.
 
 Trivial or purely conversational turns don't need any of this — skip it.
 
@@ -68,8 +68,8 @@ Trivial or purely conversational turns don't need any of this — skip it.
 | `deployment_issue` | build / CI / pipeline / rollback |
 | `recurring_error` | "same error" / "still failing" / "keeps failing" |
 
-Signals are descriptive tags, not a verdict: whether a turn is recorded as a success or a
-failure comes from how the turn itself ended.
+Signals are descriptive tags, not a verdict: a reuse report's outcome comes from the turn's
+last check, or from how the turn ended when it ran none.
 
 ## Tools