fix(mcp): keep gate_soft_continue true on unfinished soft-gap ticks

AGENTS.md

@@ -408,7 +408,7 @@ Parent。软缺口（`soft_continue=true`）下 `swarm_doctor` 同样给出
 这条引导上 `fixes` 不把 `evolver doctor --fix` 列为当前步骤；boot/status，以及
 第一次未完成 run 的 tick 与保险丝，其 `commands` 与 `next_steps` 同样不把它当成
 现在要执行的修复。没有未完成 run，或固化之后，列出这条命令仍然合适。本工具仍不执行它。
-tick / 保险丝上的 `gate_soft_continue` 仍缺席，不写成 false。
+有软缺口时，第一次 tick 与保险丝写 `gate_soft_continue=true`，不写成 false。
 硬阻断的 doctor 仍是 `ok=false`，不另给这个动作，也不写 `tick_again`。
 此时 `boot_once` 为空。忽略 `next_action` 再调 `swarm_tick` 时默认不启新周期。
 第一次 `ok=true`、`severity=soft`、`cycle_started=false`、`progress=false`、
@@ -422,8 +422,8 @@ tick / 保险丝上的 `gate_soft_continue` 仍缺席，不写成 false。
 不把裸码当成 `next_action`。软缺口留在 `warning`。两次返回都带
 `preserve_pending=true`、`abandon_run=false`、
 `action_required=swarm_solidify`（boot/status 在仍要固化时同样带上）。有软缺口时，
-第一次 tick 与保险丝都不带 `gate_soft_continue`（缺席不是 false；`ok` 第一次仍为 true，
-没有 `error`；`tick_again=false` 与 `retryable=false` 都表示不要再 tick，第一次也写 `retryable=false`，缺席不再当成可以再 tick）。两次 tick 都带
+有软缺口时，第一次 tick 与保险丝都写 `gate_soft_continue=true`（true 不是 false，也不是再 tick 的许可；`ok` 第一次仍为 true，
+没有 `error`；`tick_again=false` 与 `retryable=false` 都表示不要再 tick，第一次也写 `retryable=false`）。两次 tick 都带
 `session_not_failed=true`：先读它，会话没有失败。`loop_fuse` 不在 `error` 里，
 preserve 三元组仍指向 `swarm_solidify`。
 boot/status 的 `gate_soft_continue` 仍为 true，且不带 `session_not_failed`。未完成 run 的软缺口 boot/status 同时写 `tick_again=false`（只看 `gate_soft_continue=true` 的宿主仍可能再 tick）。

CHANGELOG.md

@@ -9,6 +9,7 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 ## Unreleased
 
 ### Fixed
+- **MCP 宿主可用性：未完成 run 的软缺口 tick 与保险丝写 `gate_soft_continue=true`**：第一次未完成 run 的 `swarm_tick`，以及同一条 run 再 tick 的 `loop_fuse=pending_redirect_loop`，在软缺口上保留 `apply_soft_gap` 写下的 `gate_soft_continue=true`，与 boot / status 同一口径。这个键不写成 false。`tick_again=false`、`retryable=false`、保险丝的 `ok=false`、`session_not_failed` 与 preserve 三元组不动。没有软缺口的重定向仍不写这个键。doctor 仍不写。不结算这条 run，不写 Parent baseline，不放宽固化门，不执行 `evolver doctor --fix`。
 - **MCP 宿主可用性：未完成 run 的第一次 tick 与保险丝也不把 `evolver doctor --fix` 当当前步骤**：软缺口固化引导上，第一次 `swarm_tick` 与 `loop_fuse=pending_redirect_loop` 的 `commands` 去掉精确的 `evolver doctor --fix`，`next_steps` 里点名它的句子改成先 `swarm_solidify`（`next_action` / preserve / `action_required`）。`gate_soft_continue` 在这两份信封上仍缺席，不写成 false。`tick_again=false` 与保险丝的 `ok=false` 不动。没有未完成 run 的 tick、boot、doctor 仍列出这条命令。本工具仍不执行 `--fix`。不结算这条 run，不写 Parent baseline，不放宽固化门。
 - **MCP 宿主可用性：软缺口固化引导不再把 `evolver doctor --fix` 当当前步骤**：未完成 run 且 soft-continue 时，`swarm_doctor` 的 `fixes` 去掉这条命令。boot / status 的 `commands` 同样去掉它，`next_steps` 里点名它的句子改成先 `swarm_solidify`（`next_action` / preserve / `action_required`）。`soft_continue` 仍为 true，`gate_soft_continue` 在 boot/status 仍为 true，doctor 仍不写这个键，也不写 `session_not_failed`。没有未完成 run 时，`fixes` / `commands` 仍列出 `evolver doctor --fix`。固化之后同样可以再列。本工具仍不执行 `--fix`。第一次 tick 与保险丝的 `commands` 不动。不结算这条 run，不写 Parent baseline，不放宽固化门。
 - **MCP 宿主可用性：未完成 run 的软缺口 boot/status 写 `tick_again=false`**：`swarm_boot` 与 `swarm_status` 在未完成 run 且软缺口、`next_action=swarm_solidify` 时带 `tick_again=false`。`gate_soft_continue` 仍为 true，不写 `session_not_failed`。preserve 三元组不动。没有未完成 run 的 soft-continue，以及压缩检查点盖过固化时，不写 `tick_again`。doctor、第一次 tick 与保险丝本来的 `tick_again=false` 不动。不结算这条 run，不写 Parent baseline，不放宽固化门。

DEBUG.md

@@ -1265,3 +1265,11 @@
 - **修复**：第一次未完成 run 的 tick 与保险丝在 `next_action=swarm_solidify` 时，`commands` 去掉精确的 `evolver doctor --fix`，`next_steps` 里点名它的句子改成先 `swarm_solidify`。`gate_soft_continue` 仍缺席，不写成 false。`tick_again=false` 不动。保险丝的 `ok` 仍为 false。没有未完成 run 的 tick 仍列出这条命令。不结算，不写 Parent baseline，不放宽固化门，不执行 `--fix`。目录缓存（DEBUG #61）未做假修复。
 - **经验**：**当前步骤的清单要盖住宿主下一步会读的每一份信封。** 开机回执拿掉一条命令之后，重定向信封若仍列出它，宿主会从 tick 的返回里把它捡回来。
 - **仍在**：只看 `soft_continue=true` 或把缺席的 `gate_soft_continue` 当成继续、完全不看 `tick_again` / `next_action` / `commands` 的宿主，仍可能再 tick。没有把这些旗标写成 false（会重开 DEBUG #67）。`commands` 里仍可能有 `evolver bench freeze` 与 `evolver bench baseline`。把保险丝 `ok=false` 当成整段失败、且不读 `session_not_failed` 与 `loop_fuse` 的宿主（DEBUG #68 余项 2）仍会停掉会话；`ok` 留着 false。目录缓存（DEBUG #61）没有做假修复。
+
+### 75. 只认 `gate_soft_continue` 的宿主在 tick 与保险丝上看不到软缺口（MCP 易用性，已修）
+
+- **症状**：#67 之后，未完成 run 的软缺口 boot / status 写 `gate_soft_continue=true`。同一条 run 的第一次 `swarm_tick` 与 `loop_fuse=pending_redirect_loop` 把这个键删掉。只在键为 true 时才认出 soft-continue 的宿主，在这两份信封上错过软缺口，尽管 `warning` 里仍是建议、`tick_again=false` 已经表示不要再 tick。doctor 仍不写这个键。
+- **根因**：`apply_soft_gap` 把旗标写成 true。`_prefer_solidify_steps(hold_tick=True)` 再 `pop` 掉，用来同时躲开「只在 true 时 tick」和「只在 false 时判死」。boot / status 没有这次 pop。缺键对只认这个键的宿主等于「没有软缺口」。
+- **修复**：软缺口上的第一次 tick 与保险丝保留 true，并在 `hold_tick` 上再写一次 true，不写 false。停 tick 仍看 `tick_again=false`、`retryable=false`、`next_action=swarm_solidify`，保险丝的 `ok` 仍为 false。`session_not_failed` 与 preserve 三元组不动。没有软缺口的重定向不写这个键。doctor 仍不写。`commands` 里 `evolver doctor --fix` 的延后不动。不结算，不写 Parent baseline，不放宽固化门，不执行 `--fix`。目录缓存（DEBUG #61）未做假修复。
+- **经验**：**软缺口旗标的原义是「缺口是建议」，停 tick 要写在 `tick_again` 上。** 把 true 从 tick 信封拿掉，只认这个键的宿主就和 boot 对不上。写成 false 会重开 DEBUG #67。
+- **仍在**：只看 `gate_soft_continue=true`、完全不看 `tick_again` / `next_action` / `ok` / `retryable` 的宿主，第一次 tick 与保险丝之后仍可能再 tick。没有把这个旗标写成 false。doctor 仍不写 `gate_soft_continue`，只读 `soft_continue=true` 的冒烟余项仍在。把保险丝 `ok=false` 当成整段失败、且不读 `session_not_failed` 与 `loop_fuse` 的宿主（DEBUG #68 余项 2）仍会停掉会话；`ok` 留着 false。`commands` 里仍可能有 `evolver bench freeze` 与 `evolver bench baseline`。目录缓存（DEBUG #61）没有做假修复。

README.md

@@ -239,9 +239,9 @@ After `uv sync`, a pull, or any install, **restart the MCP server**. A long-live
 
 Live `tools/list` already publishes `swarm_tick` parameters `prompt_mode` (`full` or `short`) and `checklist_only`, with descriptions on those fields, and read-only `swarm_doctor`. A host catalog that still lists only `agent_name` and `include_prompt`, or that cannot find `swarm_doctor`, is a stale client session. Restarting the server reloads the code; start a new host session so the client reads `tools/list` again.
 
-If a previous run is unfinished, `swarm_boot` keeps `next_action=swarm_solidify`, sets top-level `pending_solidify`, and adds `readonly_smoke` (`tools`: `swarm_doctor` and `swarm_status`, `settles_run`: false). On a soft gap it also sets `tick_again=false` while `gate_soft_continue` stays true. `hint` says to solidify first. A soft gap (`no_baseline` / `pack_absent` / `unmeasured`) still attaches `warning` and `next_steps`, and it does not replace that hint or weaken the gate. A read-only smoke can call `swarm_doctor` or `swarm_status`; neither settles the run nor invents a Parent baseline. A `swarm_tick` in that state does not start a cycle. The first call stays soft (`ok` true, `severity` soft, `cycle_started` false, `progress` false, `redirect_count` 1, `retryable` false, `next_action` still `swarm_solidify`, `last_run` unchanged). An omitted `retryable` is not a license to tick again. The same call again on that run returns `ok` false, `loop_fuse=pending_redirect_loop`, `redirect_reason=pending_redirect_loop`, `severity` soft, `retryable` false, and no `error` field. It still does not start a cycle. `loop_fuse` and that `redirect_reason` are the tick-loop fuse. The code is not an `error` and not a `next_action`. `next_action` stays `swarm_solidify`. `dispatch_reason` stays `pending_solidify`. The first redirect keeps `redirect_reason=pending_solidify` and does not set `loop_fuse`. Connect-time instructions name `loop_fuse=pending_redirect_loop` the same way. The fuse is not a gate failure and not an abandon signal; a soft gap stays on `warning` and `next_steps`. Both returns set `preserve_pending` true, `abandon_run` false, and `action_required` to `swarm_solidify` (boot and status do too while solidify is still next). Do not drop the unfinished run. When a soft gap is present, `gate_soft_continue` is omitted on the first redirect and on the fuse (`tick_again` is false). Omitted is not false: the first redirect keeps `ok` true and sets no `error`, so a host that treats only a strict false as session death stays in the session and does not tick again. Both tick returns set `session_not_failed` true. Read that field before other flags: true means the session is not failed, and `loop_fuse` does not abandon the run when the preserve flags say to call `swarm_solidify`. `next_steps` starts with `swarm_solidify`. Boot and status keep `gate_soft_continue` true while a run is still unfinished, set `tick_again` false on that soft gap, and they do not set `session_not_failed`. Pass `supersede_pending=true` to replace that run. Pause and the compaction checkpoint still refuse the cycle.
+If a previous run is unfinished, `swarm_boot` keeps `next_action=swarm_solidify`, sets top-level `pending_solidify`, and adds `readonly_smoke` (`tools`: `swarm_doctor` and `swarm_status`, `settles_run`: false). On a soft gap it also sets `tick_again=false` while `gate_soft_continue` stays true. `hint` says to solidify first. A soft gap (`no_baseline` / `pack_absent` / `unmeasured`) still attaches `warning` and `next_steps`, and it does not replace that hint or weaken the gate. A read-only smoke can call `swarm_doctor` or `swarm_status`; neither settles the run nor invents a Parent baseline. A `swarm_tick` in that state does not start a cycle. The first call stays soft (`ok` true, `severity` soft, `cycle_started` false, `progress` false, `redirect_count` 1, `retryable` false, `next_action` still `swarm_solidify`, `last_run` unchanged). An omitted `retryable` is not a license to tick again. The same call again on that run returns `ok` false, `loop_fuse=pending_redirect_loop`, `redirect_reason=pending_redirect_loop`, `severity` soft, `retryable` false, and no `error` field. It still does not start a cycle. `loop_fuse` and that `redirect_reason` are the tick-loop fuse. The code is not an `error` and not a `next_action`. `next_action` stays `swarm_solidify`. `dispatch_reason` stays `pending_solidify`. The first redirect keeps `redirect_reason=pending_solidify` and does not set `loop_fuse`. Connect-time instructions name `loop_fuse=pending_redirect_loop` the same way. The fuse is not a gate failure and not an abandon signal; a soft gap stays on `warning` and `next_steps`. Both returns set `preserve_pending` true, `abandon_run` false, and `action_required` to `swarm_solidify` (boot and status do too while solidify is still next). Do not drop the unfinished run. When a soft gap is present, `gate_soft_continue` is true on the first redirect and on the fuse (`tick_again` is false). True is not false and not a license to tick: the first redirect keeps `ok` true and sets no `error`, so a host that treats only a strict false as session death stays in the session. Stop-tick stays on `tick_again` and `retryable`. This path never writes `gate_soft_continue=false`. Both tick returns set `session_not_failed` true. Read that field before other flags: true means the session is not failed, and `loop_fuse` does not abandon the run when the preserve flags say to call `swarm_solidify`. `next_steps` starts with `swarm_solidify`. Boot and status keep `gate_soft_continue` true while a run is still unfinished, set `tick_again` false on that soft gap, and they do not set `session_not_failed`. Pass `supersede_pending=true` to replace that run. Pause and the compaction checkpoint still refuse the cycle.
 
-A cold `swarm_status` (this process has not called `swarm_boot`) returns `next_action=swarm_boot` and a short hint. After boot, status keeps `agent_name`, `source_revision`, and any soft `gap` / warnings. `tool_asset_search`, `tool_mailbox_poll`, `tool_cycle_timeline`, and `swarm_approvals` always return `{ok, items, hint}` — an empty match is not a blank body. `swarm_approvals` also keeps `pending` (the same rows as `items`) and `recent`. Read-only `swarm_doctor` mirrors `evolver doctor` without `--fix` (`ok`, soft warnings, command hints in `fixes`). `no_baseline` / `pack_absent` / `unmeasured` stay soft-continue. When a run is unfinished and that report is soft-continue, doctor also sets `next_action=swarm_solidify`, `tick_again=false`, `pending_solidify`, and the preserve triad (`preserve_pending`, `abandon_run=false`, `action_required=swarm_solidify`). `soft_continue` stays true. It still does not settle the run, write a Parent baseline, or execute `evolver doctor --fix`. On that steer, `fixes` does not list `evolver doctor --fix` as a current step. Boot and status `commands` and `next_steps` do not present it as the current action either. The first unfinished-run tick and the `loop_fuse=pending_redirect_loop` envelope do the same in `commands` and `next_steps`. `gate_soft_continue` stays omitted on those tick envelopes. The command stays listed when no run is unfinished, and after solidify. `gate_soft_continue` and `session_not_failed` stay absent. A hard blocker keeps `ok` false and does not add that action or `tick_again`. The first unfinished-run tick and the fuse already set `tick_again` false.
+A cold `swarm_status` (this process has not called `swarm_boot`) returns `next_action=swarm_boot` and a short hint. After boot, status keeps `agent_name`, `source_revision`, and any soft `gap` / warnings. `tool_asset_search`, `tool_mailbox_poll`, `tool_cycle_timeline`, and `swarm_approvals` always return `{ok, items, hint}` — an empty match is not a blank body. `swarm_approvals` also keeps `pending` (the same rows as `items`) and `recent`. Read-only `swarm_doctor` mirrors `evolver doctor` without `--fix` (`ok`, soft warnings, command hints in `fixes`). `no_baseline` / `pack_absent` / `unmeasured` stay soft-continue. When a run is unfinished and that report is soft-continue, doctor also sets `next_action=swarm_solidify`, `tick_again=false`, `pending_solidify`, and the preserve triad (`preserve_pending`, `abandon_run=false`, `action_required=swarm_solidify`). `soft_continue` stays true. It still does not settle the run, write a Parent baseline, or execute `evolver doctor --fix`. On that steer, `fixes` does not list `evolver doctor --fix` as a current step. Boot and status `commands` and `next_steps` do not present it as the current action either. The first unfinished-run tick and the `loop_fuse=pending_redirect_loop` envelope do the same in `commands` and `next_steps`. `gate_soft_continue` is true on those soft-gap tick envelopes and is never false. The command stays listed when no run is unfinished, and after solidify. Doctor still leaves `gate_soft_continue` and `session_not_failed` absent. A hard blocker keeps `ok` false and does not add that action or `tick_again`. The first unfinished-run tick and the fuse already set `tick_again` false.
 
 `swarm_boot(project=...)` binds a directory only when that path exists. `bound_paths.repo_root` is the code root (the git root walked from that directory, or the directory itself). `bound_paths.workspace_root` is the mutation directory. When they differ, or when `project` is a free-text label rather than a directory, the same object carries `warning` and `next_steps`. That is a soft warning: `ok` stays true and the loop still continues. A label is not turned into a new workspace.