@nettee/beacon 0.1.5 → 0.1.6

package/README.md

@@ -115,12 +115,15 @@ does not read, initialize, or advance the Schedule's cron cursor. Unknown
 Profiles or Schedules and failed Runs or Deliveries exit non-zero; failure
 messages include the `run_id` whenever a Run record was created.
 
-## Feishu Final Outcome cards
+## Final Outcomes
 
-An Agent explicitly calls one of two Final Outcome tools:
+An Agent explicitly calls exactly one of three Final Outcome tools:
 
 - `submit_final_outcome_text` sends a normal Feishu text message.
 - `submit_final_outcome_card` sends an interactive card.
+- `submit_final_outcome_no_reply` completes the Run without creating a
+  Delivery. Its required `reason` is stored for internal audit and is never
+  sent to the user.
 
 The card tool accepts a title, Feishu-compatible Markdown body, and up to five
 HTTP(S) link buttons. The first button is styled as primary. The tools accept
@@ -141,11 +144,16 @@ these argument shapes:
 }
 ```
 
+```json
+{ "reason": "The group announcement is outside this Profile's role." }
+```
+
 The selected form is stored durably as part of the Final Outcome. Both quoted
 message replies and scheduled chat messages preserve that choice: text remains
 text, while cards are sent with Feishu's `interactive` message type and include
-the card generation time. Manual local triggers render card outcomes as
-readable Markdown on stdout.
+the card generation time. A no-reply outcome records a successful Run without
+a Delivery record. Manual local triggers render card outcomes as readable
+Markdown on stdout and print no user-facing content for no-reply outcomes.
 
 ## State and failure behavior
 
@@ -200,6 +208,7 @@ The repository also contains Chinese operator runbooks based on the verified
 
 - [Deploy or update Beacon on macmini](docs/macmini-deploy-and-update.md)
 - [Add a Beacon Profile on macmini](docs/macmini-add-profile.md)
+- [Write a role-bounded Profile Prompt](docs/profile-prompt-writing.md)
 
 ## Development checks
 
@@ -251,10 +260,13 @@ version change takes precedence over the automatic patch bump, and CI rejects
 a changed version that is not greater than the base branch version.
 
 The bump job accepts only branches in this repository, not pull requests from
-forks. It uses the job-scoped `github.token` to push the version commit, then
-dispatches a fresh CI run for the new commit. This avoids a long-lived personal
-access token while working around GitHub's suppression of workflows caused by
-bot pushes.
+forks. Configure a repository secret named `AUTO_BUMP_TOKEN` with a fine-grained
+GitHub token whose repository access is limited to `nettee/beacon` and whose
+Contents permission is read/write. The checkout persists that credential for
+the version commit push, so GitHub attributes the PR update to the token owner
+and starts follow-up PR checks without an approval prompt. When the secret is
+absent, CI falls back to the job-scoped `github.token`; the bump still succeeds,
+but GitHub requires a maintainer to approve the resulting PR workflow run.
 
 Submit the release candidate through the normal pull request checks. After the
 initial package bootstrap, each push to `main` runs

package/dist/feishu/gateway.js

@@ -22,6 +22,9 @@ function assertSucceeded(operation, result) {
     }
 }
 export function encodeFeishuFinalOutcome(outcome) {
+    if (outcome.kind === "no_reply") {
+        throw new Error("No-reply Outcomes must not be encoded for Delivery");
+    }
     if (outcome.kind === "text") {
         return {
             msgType: "text",
@@ -84,6 +87,18 @@ export class FeishuGateway {
             data: { reaction_type: { emoji_type: "OnIt" } },
         });
         assertSucceeded("add OnIt reaction", result);
+        const reactionId = result.data?.reaction_id;
+        if (!reactionId) {
+            throw new Error("add OnIt reaction returned no reaction_id");
+        }
+        return {
+            clear: async () => {
+                const deleteResult = await this.client.im.v1.messageReaction.delete({
+                    path: { message_id: messageId, reaction_id: reactionId },
+                });
+                assertSucceeded("remove OnIt reaction", deleteResult);
+            },
+        };
     }
     async fetchMessage(messageId) {
         const result = await this.client.im.v1.message.get({

package/dist/feishu/intake.js

@@ -31,12 +31,14 @@ export class FeishuIntake {
         });
         if (!claim.created)
             return "duplicate";
-        void this.options
+        const acknowledgement = this.options
             .acknowledge(event.message.message_id)
             .catch((error) => {
             console.error(`[beacon] non-critical acknowledgement failure message_id=${event.message.message_id}: ${error instanceof Error ? error.message : String(error)}`);
+            return undefined;
         });
-        const task = this.options.process(claim.record.triggerKey, async () => {
+        const task = this.options
+            .process(claim.record.triggerKey, async () => {
             const normalized = await buildFeishuTriggerInput({
                 messageId: event.message.message_id,
                 chatId: event.message.chat_id,
@@ -54,6 +56,17 @@ export class FeishuIntake {
                 quotedMessages: normalized.quotedMessages,
                 currentMessage: normalized.message,
             };
+        })
+            .finally(async () => {
+            const activeAcknowledgement = await acknowledgement;
+            if (!activeAcknowledgement)
+                return;
+            try {
+                await activeAcknowledgement.clear();
+            }
+            catch (error) {
+                console.error(`[beacon] non-critical acknowledgement cleanup failure message_id=${event.message.message_id}: ${error instanceof Error ? error.message : String(error)}`);
+            }
         });
         this.active.add(task);
         void task

package/dist/manual-trigger.js

@@ -50,7 +50,8 @@ export async function runManualTrigger(configPath, profileId, input, output = wr
         }));
         const record = (await store.list()).find((candidate) => candidate.triggerKey === claim.record.triggerKey);
         if (record?.run?.state !== "succeeded" ||
-            record.delivery?.state !== "delivered") {
+            (record.finalOutcome?.content.kind !== "no_reply" &&
+                record.delivery?.state !== "delivered")) {
             throw new Error(`Manual Trigger failed: ${record?.run?.failure?.code ?? record?.delivery?.failure?.code ?? "unknown"}`);
         }
     }

package/dist/outcome/content.js

@@ -12,6 +12,7 @@ const buttonSchema = z
     .strict();
 export const finalOutcomeContentSchema = z.discriminatedUnion("kind", [
     z.object({ kind: z.literal("text"), text: nonBlank }).strict(),
+    z.object({ kind: z.literal("no_reply"), reason: nonBlank }).strict(),
     z
         .object({
         kind: z.literal("card"),
@@ -27,6 +28,8 @@ export function parseFinalOutcomeContent(value) {
 export function renderFinalOutcomeAsText(outcome) {
     if (outcome.kind === "text")
         return outcome.text;
+    if (outcome.kind === "no_reply")
+        return "";
     return [
         `# ${outcome.title}`,
         "",

package/dist/outcome/instructions.js

@@ -1,5 +1,5 @@
 export const finalOutcomeSystemInstructions = [
-    "When your work is complete, call exactly one of submit_final_outcome_text or submit_final_outcome_card with the exact user-facing response.",
-    "Use submit_final_outcome_text for a normal response, or submit_final_outcome_card when a titled Markdown summary and link buttons materially improve the result.",
+    "When your work is complete, call exactly one of submit_final_outcome_text, submit_final_outcome_card, or submit_final_outcome_no_reply.",
+    "Use submit_final_outcome_text for a normal response, submit_final_outcome_card when a titled Markdown summary and link buttons materially improve the result, or submit_final_outcome_no_reply when the triggering message is outside the Profile's role and should receive no response.",
     "Beacon ignores ordinary assistant final text for Delivery.",
 ];

package/dist/run/orchestrator.js

@@ -209,7 +209,8 @@ export class RunOrchestrator {
                     submittedAt: this.timestamp(),
                 },
             }));
-            await this.deliver(triggerKey);
+            if (outcome.kind !== "no_reply")
+                await this.deliver(triggerKey);
         }
         catch (error) {
             submission.cancel();
@@ -297,6 +298,7 @@ export class RunOrchestrator {
                 continue;
             }
             if (record.finalOutcome &&
+                record.finalOutcome.content.kind !== "no_reply" &&
                 !record.delivery &&
                 (record.run.state === "succeeded" || record.run.state === "failed")) {
                 await this.deliver(record.triggerKey);

package/dist/runtime/pi-outcome-extension.js

@@ -11,6 +11,9 @@ export function cardOutcomeFromToolParams(params) {
         buttons: params.buttons ?? [],
     });
 }
+export function noReplyOutcomeFromToolParams(params) {
+    return parseFinalOutcomeContent({ kind: "no_reply", reason: params.reason });
+}
 async function invokeBeaconCli(outcome, signal) {
     const cliPath = process.env.BEACON_CLI_PATH;
     if (!cliPath)
@@ -47,7 +50,7 @@ export default function registerOutcomeTools(pi) {
     pi.registerTool({
         name: "submit_final_outcome_text",
         label: "Submit Final Outcome Text",
-        description: "Submit the final response as a normal text message. Call either this tool or submit_final_outcome_card exactly once after completing the task.",
+        description: "Submit the final response as a normal text message. Call exactly one Final Outcome tool after completing the task.",
         parameters: {
             type: "object",
             properties: {
@@ -68,7 +71,7 @@ export default function registerOutcomeTools(pi) {
     pi.registerTool({
         name: "submit_final_outcome_card",
         label: "Submit Final Outcome Card",
-        description: "Submit the final response as a structured Feishu card. Call either this tool or submit_final_outcome_text exactly once after completing the task.",
+        description: "Submit the final response as a structured Feishu card. Call exactly one Final Outcome tool after completing the task.",
         parameters: {
             type: "object",
             properties: {
@@ -104,4 +107,25 @@ export default function registerOutcomeTools(pi) {
             return acceptedResult();
         },
     });
+    pi.registerTool({
+        name: "submit_final_outcome_no_reply",
+        label: "Submit No Reply Outcome",
+        description: "Complete the run without sending a reply. Use this exactly once when the triggering message is outside the Profile's role.",
+        parameters: {
+            type: "object",
+            properties: {
+                reason: {
+                    type: "string",
+                    minLength: 1,
+                    description: "A concise internal reason why the message should not receive a reply.",
+                },
+            },
+            required: ["reason"],
+            additionalProperties: false,
+        },
+        async execute(_toolCallId, params, signal) {
+            await invokeBeaconCli(noReplyOutcomeFromToolParams(params), signal);
+            return acceptedResult();
+        },
+    });
 }

package/examples/profiles/example/prompt.md

@@ -1 +1,7 @@
-You are the configured Beacon Profile. Work only within the provided workspace and return a concise, user-facing result through the required Final Outcome tool.
+You are a workspace-status assistant. Your only role is to answer requests about the configured workspace's files, version-control state, and verification results.
+
+Before reading files or running commands, decide whether the input is within that role. A group `@ All` does not by itself address this bot, and a direct message is not automatically relevant. Use the quoted-message chain when the current message is a follow-up.
+
+If the input is outside this role, immediately call `submit_final_outcome_no_reply` with a concise internal reason. Do not inspect the workspace or submit a text or card response.
+
+For relevant input, work only within the configured workspace and return a concise, user-facing result through the appropriate Final Outcome tool. Required operations must fail visibly; do not fabricate fallback data or report success after incomplete work.

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@nettee/beacon",
-  "version": "0.1.5",
+  "version": "0.1.6",
   "description": "A single-host bridge from messaging and schedules to isolated agent runs",
   "license": "Apache-2.0",
   "repository": {