@desplega.ai/agent-swarm 2.5.2 → 2.5.4

package/src/workflows/task-step-routing.ts

@@ -7,6 +7,7 @@ import {
   updateWorkflowRun,
   updateWorkflowRunStep,
 } from "../be/db";
+import { getRunLineageBlock } from "../be/db/workflow-runs";
 import {
   type RetryPolicy,
   RetryPolicySchema,
@@ -18,6 +19,7 @@ import {
 import { checkpointStep, checkpointStepFailure } from "./checkpoint";
 import { getSuccessors } from "./definition";
 import type { ExecutorRegistry } from "./executors/registry";
+import { resolveStepNode } from "./foreach-body";
 import { type ForeachSettleResult, resolveForeachParent, settleForeachChild } from "./foreach-join";
 
 export interface TaskStepRoutingResult {
@@ -146,6 +148,7 @@ export async function checkpointPortStepAndResolveSuccessors(
     if (!run || (run.status !== "waiting" && run.status !== "running")) {
       return UNCLAIMED;
     }
+    if (await getRunLineageBlock(runId)) return UNCLAIMED;
     // A foreach child's port (approval rejected/timeout, wait event/timeout)
     // is stored for display and still counts as a completed item: the body's
     // ports do not route, the item's consumer reads the output.
@@ -182,8 +185,8 @@ export async function checkpointPortStepAndResolveSuccessors(
  *   or the failed task is no longer the one bound to the step (it was
  *   already retried and a fresh task now owns it); the caller stops.
  * - `not-eligible`: no retry applies (no `node.retry` and not a supersede
- *   cut, not an agent-task node, or retries exhausted); the caller applies
- *   `onNodeFailure` as before.
+ *   cut, not an agent-task node, retries exhausted, or a foreach child whose
+ *   run is no longer live); the caller applies `onNodeFailure` as before.
  */
 export type TaskStepRetryOutcome = "scheduled" | "not-claimed" | "not-eligible";
 
@@ -217,8 +220,13 @@ export const SUPERSEDE_RETRY_POLICY: RetryPolicy = {
  * only after ownership is confirmed: a stale caller's step snapshot may carry
  * the counter of a newer attempt.
  *
- * Foreach children (`<foreachId>#<key>`) are not matched here: the retry
- * poller resolves nodes by exact id and cannot re-dispatch a child yet.
+ * A foreach child (`<foreachId>#<key>`) with an `agent-task` body takes the
+ * parent foreach node's `retry`, else the same supersede default. While queued
+ * it does not settle, so the join stays open; the retry poller re-queues it and
+ * the foreach driver starts it under the fan-out's concurrency cap. A child is
+ * queued only while its run is live. Exhausted retries, a run that is no longer
+ * live and every other failure settle the item through `onNodeFailure` and the
+ * fan-out's failure policy as before.
  */
 export async function scheduleTaskStepRetry(
   def: WorkflowDefinition,
@@ -227,11 +235,14 @@ export async function scheduleTaskStepRetry(
   taskId: string,
   reason: string,
 ): Promise<TaskStepRetryOutcome> {
-  const node = def.nodes.find((n) => n.id === step.nodeId);
+  const foreachParent = resolveForeachParent(def, step.nodeId);
+  const node = resolveStepNode(def, step.nodeId);
   if (!node || node.type !== "agent-task") return "not-eligible";
+  // A synthesized child node carries only the body; its retry lives on the parent.
+  const declaredRetry = foreachParent ? foreachParent.retry : node.retry;
   let policy: RetryPolicy;
-  if (node.retry) {
-    const parsed = RetryPolicySchema.safeParse(node.retry);
+  if (declaredRetry) {
+    const parsed = RetryPolicySchema.safeParse(declaredRetry);
     if (!parsed.success) return "not-eligible";
     policy = parsed.data;
   } else {
@@ -243,6 +254,15 @@ export async function scheduleTaskStepRetry(
     const current = await getWorkflowRunStep(step.id);
     if (!current || current.status !== "waiting") return "not-claimed";
     if (!(await isTaskBoundToStep(taskId, step.id))) return "not-claimed";
+    // A sibling's fail-fast or another branch can fail the run while the
+    // caller awaited. A child queued then would revive the failed run, so the
+    // caller records the cut as the item's late outcome instead.
+    if (foreachParent) {
+      const run = await getWorkflowRun(runId);
+      if (run?.status !== "running" && run?.status !== "waiting") return "not-eligible";
+    }
+    // No retry is queued inside a cancelled tree.
+    if (await getRunLineageBlock(runId)) return "not-claimed";
     if (current.retryCount >= policy.maxRetries) return "not-eligible";
     await detachTaskFromWorkflowRunStep(taskId);
     await checkpointStepFailure(runId, step.id, reason, current.retryCount, policy);
@@ -280,6 +300,8 @@ export async function completeTaskStepAndResolveSuccessors(
     if (!current || current.status !== "waiting") return UNCLAIMED;
     // A stale snapshot of a step since redriven to a fresh task must not route it.
     if (ownerTaskId && !(await isTaskBoundToStep(ownerTaskId, step.id))) return UNCLAIMED;
+    // A late result inside a cancelled tree routes nothing and revives nothing.
+    if (await getRunLineageBlock(runId)) return UNCLAIMED;
 
     if (resolveForeachParent(def, step.nodeId)) {
       // onNodeFailure:"continue" completions persist the failure reason as

package/templates/ai-toolbox.manifest.json

@@ -1,5 +1,5 @@
 {
-  "commit": "77abe63716623f28177ebe29679c37b5859ef40e",
+  "commit": "5a577022800d9b43ccec5376bfe16c7b24eecd91",
   "excludedSkills": [
     "feedback"
   ],
@@ -16,8 +16,8 @@
     "templates/skills/comms/files/references/ste-rules.md": "6ee2ed5d6b49d2e421b6c9d22310a5b589fa1abe81de06213e4b89fbc1271e74",
     "templates/skills/comms/files/references/visual-shapes.md": "5e07e9250ad4cdc877af2281525185f0de89c01f90a077463a2c77c8a7b998e8",
     "templates/skills/delegate-work/config.json": "5b1542eb0be359bb8a17117fcf87917e0607d1b939ae3266ddffacdc14efb281",
-    "templates/skills/delegate-work/content.md": "c9562bd17846d3595d40172ecfdc6f85bb99eaf23597eb9d5aa9635d5c14d079",
-    "templates/skills/delegate-work/files/scripts/codex-exec.sh": "be9b33ba111afc8b38c20a4190eae19099025474d4ed3278d944ee0e6b6621af",
+    "templates/skills/delegate-work/content.md": "1046da2fbc5603c4deed92beb6d98c443df21e87cf83a7d7b3ce58dbc9d01997",
+    "templates/skills/delegate-work/files/scripts/codex-exec.sh": "4524a224fe0657d2e2872d97a70357c4a3ce84d2241bd935ed83febf1fac6e37",
     "templates/skills/design-docs/config.json": "b78141763541486d67a105e8fa23873c9ffee6a682832c4b7266b8320d55ead7",
     "templates/skills/design-docs/content.md": "fc860031defbd192c1b8e1c024b46509d78dcf10dbf1b8c93cc4d62396153a52",
     "templates/skills/design-docs/files/template.md": "72d2cdcf8713c08d525d69892c5fed536107594522e3e799ae44f51df7015a36",
@@ -38,13 +38,13 @@
     "templates/skills/planning/content.md": "543476e4dbb5c352acd686c3929c45cc992d7d033d06f7e9a30f78009b50785c",
     "templates/skills/planning/files/template.md": "c0d5ca63c07934546172322bbd1180e2901dc45e27b5d12ce3dc5c30f5e7a98b",
     "templates/skills/qa/config.json": "c53b01558a691c0343b6270a548e6172af86600a429b789eedb518849f7ea741",
-    "templates/skills/qa/content.md": "c885095ea814b4c410dbc7aac1e2fde09b94b735d818b6f06153901311a232a0",
+    "templates/skills/qa/content.md": "46b1147d8af1a292d9bcd7091385d2c98264f10bb5afe4fb61073113eb0a6574",
     "templates/skills/qa/files/template.md": "1902e4bb65b19ab58c54ab5b63693ad09adecee69079c2c70f58d7d6c6597896",
     "templates/skills/questioning/config.json": "da2d5fe6dfdac1915a0476a7063e62ebeb996b84caff07cc7527a55477656a80",
     "templates/skills/questioning/content.md": "66f826a191e96d0041adca00a7c197a458de8e6fe164c1599b8708098d4109ab",
     "templates/skills/questioning/files/template.md": "a68d72801a0b10380bf773b07e9d2f8d4c5b4b1c609ce954289e38ffc78cc223",
     "templates/skills/researching/config.json": "921c062cab599097cc7677d2dad4797ec4d5844d2ee73607fbd417ec3fa487d4",
-    "templates/skills/researching/content.md": "0e8f98a7a5c75059fba6cd2c908b8268efce13192c1220ba594f2a9e3bf1e60a",
+    "templates/skills/researching/content.md": "92951142327415696726a2577defdb2379cc611cdafa9cdac92431a5e60f618f",
     "templates/skills/researching/files/template.md": "411d34b98c610ef715edf0f24d09f828ba2d0f65f688e7879ef33a9453007949",
     "templates/skills/reviewing/config.json": "09c40bb098e9738c60a848995fa9fe7568d158631a089c76ba02c0dd955445c7",
     "templates/skills/reviewing/content.md": "1bbc285c9ad8cc3312f1ee2c558b04ffee914ffb6f5c63fb3741fcbb3e300fa5",
@@ -130,13 +130,13 @@
       },
       {
         "from": "`${CLAUDE_PLUGIN_ROOT}/skills/delegate-work/scripts/codex-exec.sh`",
-        "line": 66,
+        "line": 70,
         "skill": "delegate-work",
         "to": "`bash ~/.claude/skills/delegate-work/scripts/codex-exec.sh`"
       },
       {
         "from": "| codex-exec.sh -m",
-        "line": 69,
+        "line": 73,
         "skill": "delegate-work",
         "to": "| bash ~/.claude/skills/delegate-work/scripts/codex-exec.sh -m"
       },
@@ -178,7 +178,7 @@
       },
       {
         "from": "cc-plugin/base/skills/researching/template.md",
-        "line": 113,
+        "line": 115,
         "skill": "researching",
         "to": "template.md"
       },

package/templates/community/open-prompt-showreel/PLAYBOOK.md

@@ -10,7 +10,7 @@ The worked example is a 24-second video about a model checker finding race condi
 - Every number on screen read live from a named source into `facts.json` right before render.
 - The real Agent Swarm logo on the end card, never a redrawn one.
 - A contact sheet per critique round, so a reviewer sees what the maker saw.
-- Delivery to agent-fs, with a share link per mp4.
+- Delivery to agent-fs, with a live app link per mp4.
 
 ## Template Files
 
@@ -120,10 +120,9 @@ The maker uploads to `thoughts/<maker-agent-id>/videos/<slug>/` (versions as `<s
 
 ```bash
 agent-fs --org <ORG_ID> --drive <DRIVE_ID> write thoughts/<maker-agent-id>/videos/<slug>/<file>.mp4 --file ./out/<file>.mp4
-agent-fs --org <ORG_ID> --drive <DRIVE_ID> share-create thoughts/<maker-agent-id>/videos/<slug>/<file>.mp4 --expires-in 604800 --json
 ```
 
-A share link opens a player with no login and expires in 7 days at most. Report the url, the share id and the expiry. A raw signed URL downloads instead of playing, and agent-fs serves a raw `.mp4` as `application/octet-stream`. For a byte-exact copy back down: `agent-fs download <path> -o <file>` (`cat` is paginated and truncates).
+Hand each mp4 over as its live app link `${AGENT_FS_LIVE_URL}/file/~/<org>/<drive>/<path>`: the viewer plays it for a signed-in teammate. Make a `share-create` link only when the brief asks for one or names a reader without an agent-fs login (a customer, a public post, an outsider). Then run `agent-fs --org <ORG_ID> --drive <DRIVE_ID> share-create <path> --expires-in 604800 --json`; the share page plays with no login for 7 days at most. Report the url, the share id and the expiry. A raw signed URL downloads instead of playing, and agent-fs serves a raw `.mp4` as `application/octet-stream`. For a byte-exact copy back down: `agent-fs download <path> -o <file>` (`cat` is paginated and truncates).
 
 **Not run** for this playbook: `write` to a real drive and `share-create` (it publishes a public link). `write --file`, `share-create` and `signed-url --inline` flags were checked against the CLI's `--help`. `agent-fs download` was run on the example master.
 
@@ -261,8 +260,8 @@ The maker's output is not the finish line. Before relaying, the Lead:
 2. Opens the end-card frame: real logo, credit line present and readable.
 3. Checks every on-screen number against the maker's `facts.json`, and `facts.json` against the source named in the brief.
 4. Checks the ffprobe line: `h264`, `yuv420p`, exactly 1920x1080 (and 1080x1350), duration as briefed, and an `aac` audio stream when the video has music. (`ffprobe` after `setup.sh`, or `ffmpeg -i` stderr in the stock image.)
-5. Plays the share link once.
-6. Relays the share link, the share id and the expiry, plus the `attribution` string from `music/current.json` (source link, license link, changes made). The maker never posts.
+5. Plays the live app link once.
+6. Relays the live app link (plus the share url, id and expiry when one was asked for) and the `attribution` string from `music/current.json` (source link, license link, changes made). The maker never posts.
 
 ## Known Gotchas
 
@@ -274,7 +273,7 @@ The maker's output is not the finish line. Before relaying, the Lead:
 | PNG frames encoded by libx264 without `-pix_fmt` | The output is `yuv444p` (High 4:4:4 Predictive), which many players and uploaders reject | Always `-pix_fmt yuv420p`. `render.mjs` does. |
 | Full-range `yuvj420p` | Rendered by `remotion render` by default | `--color-space=bt709`, or `x-reencode.sh` |
 | Remotion's bundled ffmpeg is stripped | no `fps`, `tile`, `hstack` filters | Use the image's `/usr/local/bin/ffmpeg` for analysis and sheets |
-| agent-fs serves `.mp4` as `application/octet-stream` | A content-type check on a raw link fails | Use a `share-create` link. Do not retry the content-type check. |
+| agent-fs serves `.mp4` as `application/octet-stream` | A content-type check on a raw link fails | Open it through the live app link, or a `share-create` link for a reader without a login. Do not retry the content-type check. |
 | Signed URLs expire and default to attachment | A `<video src>` pointing at one downloads instead of playing, and stops working at the expiry | Never use one as a public URL. For a page, commit a compressed copy. `signed-url --inline` renders in a browser tab but still expires. |
 | `agent-fs cat` truncates | Pagination hides the tail of a big file | `agent-fs download <path> -o <file>` for exact bytes |
 | Container restart mid-render | The scratch dir is the only thing that survives | Keep it in `/workspace/personal/<slug>`, name it in the first progress note, upload an interim cut before the long render |

package/templates/community/open-prompt-showreel/lead-prompt.md

@@ -65,7 +65,7 @@ How:
 
 Deliver to agent-fs thoughts/<maker-agent-id>/videos/<slug>/: <1920x1080 [+ 1080x1350]> h264 yuv420p mp4,
 contact sheet(s), facts.json, reel.html.
-Output: agent-fs paths, a share-create link per mp4 (url + share id), ffprobe line, model you ran on,
+Output: agent-fs paths, a live app link per mp4, ffprobe line, model you ran on,
 each on-screen number with its source, music attribution (source link, license link, changes made).
 ```
 

package/templates/community/open-prompt-showreel/skills/open-prompt-showreel/SKILL.md

@@ -54,7 +54,7 @@ How:
 
 Deliver to agent-fs thoughts/<maker-agent-id>/videos/<slug>/: <1920x1080 [+ 1080x1350]> h264 yuv420p mp4,
 contact sheet(s), facts.json.
-Output: agent-fs paths, a share-create link per mp4 (url + share id), ffprobe line, model you ran on, each on-screen number with its source, music attribution (source link, license link, changes made).
+Output: agent-fs paths, a live app link per mp4, ffprobe line, model you ran on, each on-screen number with its source, music attribution (source link, license link, changes made).
 ```
 
 Optional style add-on for a calmer, premium cut (the "Apple framework"): 3 colors, one background, one font, no HUD/grain; one shot = one idea; ~3 s open and ~2 s held end; key object centered; eased, overlapping keyframes; no hard cuts; ~108 BPM, phrase hit on the first product frame; no SFX. Add it only when asked.
@@ -127,9 +127,9 @@ window.frame = (f) => { render(f); return cv.toDataURL("image/png"); };
 
 - agent-fs: `thoughts/<maker-agent-id>/videos/<slug>/` (versions as `<slug>-v2`, `<slug>-v3`, never overwrite an earlier version).
 - Upload the mp4s, contact sheet(s), `facts.json`, the music json, and `reel.html` so the next round can start from it. After every non-trivial write, `agent-fs stat <path> --json` and check the size.
-- Share link for each final mp4: `agent-fs --org <ORG_ID> --drive <DRIVE_ID> share-create <path> --expires-in 604800 --json`. The share page plays the video inline with no login. Max 7 days; report the url, the share id and the expiry. Do not hand out raw signed URLs or viewer URLs for video.
-- agent-fs serves a raw `.mp4` download as `application/octet-stream`; the share page's player gets `video/mp4`. Mention it rather than retrying.
-- Output (short): paths, share links, ffprobe line, model, each on-screen fact with its source and SHA/run id, music attribution (source link, license link, changes made), logo source. No Slack post by the maker; Lead relays.
+- Link for each final mp4: the live app link `${AGENT_FS_LIVE_URL}/file/~/<org>/<drive>/<path>`; its viewer plays the mp4 for a signed-in teammate. Make a `share-create` link only when the brief asks for one or names a reader without an agent-fs login (a customer, a public post, an outsider). Then run `agent-fs --org <ORG_ID> --drive <DRIVE_ID> share-create <path> --expires-in 604800 --json`; the share page plays the video with no login for 7 days at most. Report the url, the share id and the expiry. Do not hand out raw signed URLs for video.
+- agent-fs serves a raw `.mp4` download as `application/octet-stream`; the viewer and the share page play it. Mention it rather than retrying.
+- Output (short): paths, live app links (plus share url, id and expiry only if asked), ffprobe line, model, each on-screen fact with its source and SHA/run id, music attribution (source link, license link, changes made), logo source. No Slack post by the maker; Lead relays.
 
 ## Feedback rounds (v2, v3)
 

package/templates/extensions/comms-adherence-gate/README.md

@@ -1,14 +1,14 @@
 # comms-adherence-gate
 
-Agents finish Slack tasks with a final message a human reads. This extension judges that message against the requester's stated comms preferences and their latest request with Jev, [TypeSafe](https://typesafe.ai)'s decision model. In `enforce` it blocks a non-adherent completion once and tells the agent what to fix.
+Agents finish Slack tasks with a final message a human reads, and post to Slack threads directly. This extension judges that text with Jev, [TypeSafe](https://typesafe.ai)'s decision model: against the requester's stated comms preferences and their latest request, and against a plain-language rubric that applies to every message. In `enforce` it blocks a failing call once and tells the agent what to fix. A `plain` failure points the agent at the `comms` skill.
 
 Install config:
 
 ```json
-{ "mode": "shadow" }
+{ "mode": "shadow", "plainMode": "shadow" }
 ```
 
-Start in `shadow`, read the verdicts in KV for a few days, then switch to `enforce`.
+Start in `shadow`, read the verdicts in KV for a few days, then switch to `enforce`. `mode` and `plainMode` move independently.
 
 ## Modes
 
@@ -16,14 +16,22 @@ Start in `shadow`, read the verdicts in KV for a few days, then switch to `enfor
 |---|---|
 | `off` | Never evaluates. |
 | `shadow` (default) | Asks Jev and records the verdict. Never blocks. |
-| `enforce` | Asks Jev, records the verdict, and blocks the `store-progress` call when a dimension fails, at most `maxBlocksPerTask` times per task. The agent gets the failed dimensions, their scores and the stated preference, and calls again. |
+| `enforce` | Asks Jev, records the verdict, and blocks the call when a dimension fails, at most `maxBlocksPerTask` times per task. The agent gets the failed dimensions, their scores and the stated preference, and calls again. |
+
+`mode` governs the preference dimensions (`verbosity`, `tone`, `language`). `plainMode` governs `plain`. A dimension blocks only when its own mode is `enforce`. Both share one block budget per task, across every gated tool.
 
 ## What it gates
 
-Hook: `pre.tool.call` on `store-progress` with `status: "completed"` and a non-empty `output`. Calls made by extensions are skipped. The task must:
+Hook: `pre.tool.call` on:
+
+- `store-progress` with `status: "completed"` and a non-empty `output`, on a task that came from Slack (`source = "slack"` or a `slackThreadTs`). Other completions pass without a Jev call.
+- `slack-reply`, `slack-post`, `slack-start-thread` and `slack-update`, judged on `message`. The task is `slack-reply`'s `taskId`, else the caller's task context. A post with no task context passes.
+
+Only a call the tool itself accepts is judged, so no other task's text reaches Jev. For `store-progress` the caller must own the task, the task must be unassigned, or the caller must be a lead. For the Slack tools the caller must be the task's assignee or its creator, the rule `slack-reply` enforces. Calls made by extensions are skipped. Script SDK calls never reach `pre.tool.call` (the tool registrar dispatches it only for MCP calls); if one ever does, the gate records it and never blocks.
 
-1. Come from Slack (`source = "slack"` or a `slackThreadTs`). Other tasks pass without a Jev call.
-2. Have a requester with stated preferences in `users.metadata.comms`. A requester with none passes. Set them with the `manage-user` tool.
+A requester with no stated preferences in `users.metadata.comms` still gets the `plain` question. Set preferences with the `manage-user` tool.
+
+In a thread with several people, the requester is the person whose message created the task. A message that @-mentions someone (`<@U123>`) is judged against the first mentioned person with stored preferences, resolved through their Slack identity in `user_external_ids`; otherwise against the requester. The lookup is one SQL read, only when the text mentions someone and `mode` is not `off`. The verdict records who was judged in `judgedFor` (`mention` or `requester`) and `audience`.
 
 ## Decision
 
@@ -31,7 +39,21 @@ The gate reads the task row and asks one `noul` question per stated preference i
 
 Jev sees the preferences, the requester's latest message (the text after the last `</thread_context>`, from the last 1,400 characters of the task), the first 400 characters of the task, and the output. Its judging rules say an explicit request outranks the stored default. A requested long answer, story or document is not a verbosity failure.
 
-A dimension fails when its score is below `threshold`. Any failed dimension fails the output.
+It also asks `plain` on every message unless `plainMode` is `off` or the text is a JSON document (structured output for an `outputSchema`). The rubric comes from the `comms` skill: everyday words, one idea per sentence, every number with a unit and a direction word, and no unexplained shorthand (n, median, p90, cohort, baseline, delta, `±X at 95%`). Code identifiers, task ids, PR numbers, links and versions are fine. Jev gets contrasting plain and dense examples in `plain_language_rules`; a prose-only rubric scored plain rewrites as low as the dense originals.
+
+All questions go in one Jev request. A preference dimension fails below `threshold`; `plain` fails below `plainThreshold`. Any failed dimension fails the message.
+
+### Calibration of `plain`
+
+Jev scores dense status text near 0 and plain text near 0.8 to 0.9, with a wide middle of compact but readable engineering notes. Measured on 400 Slack-facing completions over 7 days, at `threshold` 0.75 for preferences:
+
+| `plainThreshold` | Newly blocked (of messages that pass today) |
+|---|---|
+| 0.75 | 81% to 83% |
+| 0.3 | 9% to 13% |
+| 0.25 (default) | 5% to 7% |
+
+At 0.25 the dense prompt-cache-monitor post scores about 0.15 and its plain rewrite about 0.81.
 
 ## Jev transport
 
@@ -51,14 +73,16 @@ Every error lets the completion through: a missing key, an HTTP error, a network
 
 ## Self-test
 
-Call `store-progress` with a non-completed status and `progress` starting with `comms-gate-selftest`. The gate judges a known-bad and a known-good message against a terse preference and writes `selftest:last`: both scores, the threshold, the provider, and `controlsOk` (bad below threshold and good at or above it). Real traffic is untouched.
+Call `store-progress` with a non-completed status and `progress` starting with `comms-gate-selftest`. The gate judges a known-bad and a known-good message against a terse preference, and a dense and a plain message on `plain`, then writes `selftest:last`: every score, both thresholds, the provider, `controlsOk` (verbosity: bad below `threshold`, good at or above it) and `plainControlsOk` (the same against `plainThreshold`). Real traffic is untouched.
 
 ## Config
 
 | Key | Default | Notes |
 |---|---|---|
 | `mode` | `shadow` | `off`, `shadow`, or `enforce` |
-| `threshold` | `0.5` | A dimension scoring below this fails |
+| `threshold` | `0.5` | A preference dimension scoring below this fails |
+| `plainMode` | `shadow` | `off`, `shadow`, or `enforce` for the `plain` dimension |
+| `plainThreshold` | `0.25` | `plain` scoring below this fails. See [Calibration](#calibration-of-plain) |
 | `timeoutMs` | `4000` | Jev call timeout, capped at 4000 (the dispatcher cancels a handler at 5 s) |
 | `maxBlocksPerTask` | `1` | Blocks per task in `enforce`; after that the gate records and lets it through |
 | `provider` | `auto` | `auto`, `typesafe`, or `openrouter`. See [Jev transport](#jev-transport) |
@@ -76,13 +100,14 @@ No key is written to config, KV, logs, or a block reason.
 
 ## KV namespace `ext:comms-gate`
 
-- `last`: the latest verdict, with `scores`, `failed`, `provider`, `latencyMs`, the preferences, a 400-character `askPreview` and a 700-character `outputPreview`.
+- `last`: the latest verdict, with `tool`, `scores`, `failed`, `mode`, `plainMode`, `provider`, `latencyMs`, the preferences, a 400-character `askPreview` and a 700-character `outputPreview`.
 - `fail:<epochMs>:<taskId>`: one row per failed verdict, with `action` (`blocked` or `recorded`).
 - `blocks:<taskId>`: blocks issued for that task.
 - `error:last`, `selftest:last`.
-- Counters: `total`, `pass`, `fail`, `blocked`, `escaped` (enforce, block budget spent), `shadowed`, `error`, and `dim:<dimension>`.
+- Counters: `total`, `pass`, `fail`, `blocked`, `escaped` (a blocking dimension failed, block budget spent), `shadowed` (only non-blocking dimensions failed), `error`, and `dim:<dimension>`.
 
 ## Known limits
 
 - Only the last 1,400 characters of the task are searched for the latest request. A very long final message is cut to its last 900 characters.
-- The block budget is per task, so a retried completion after a block goes through once the budget is spent, even if it still fails.
+- The block budget is per task, so a retried call after a block goes through once the budget is spent, even if it still fails.
+- Each gated Slack post waits for one Jev call (about 0.2 s measured, capped by `timeoutMs`).