agent-afk 5.192.1 → 5.194.0

package/dist/agent/daemon/handoff-store.d.ts

@@ -19,6 +19,7 @@ export interface HandoffRecord {
     answeredAt?: string;
     answerSource?: string;
     originalCommand: string;
+    telegramMessageId?: number;
 }
 export declare function writeHandoff(record: HandoffRecord, handoffsDir?: string): Promise<void>;
 export declare function readHandoff(taskId: string, handoffsDir?: string): Promise<HandoffRecord | null>;

package/dist/agent/daemon/handoff-wiring.d.ts

@@ -1,15 +1,19 @@
 import type { ElicitationHandler } from '../elicitation-router.js';
+import type { Telegraf } from 'telegraf';
 export interface DaemonElicitationOpts {
     taskId: string;
     originalCommand: string;
     queueDir?: string;
     handoffsDir?: string;
+    bot?: Telegraf;
+    chatId?: number;
+    threadId?: number;
 }
 export declare function makeDaemonElicitationHandler(opts: DaemonElicitationOpts): ElicitationHandler;
 export interface RecoveryResult {
     renotified: number;
     expired: number;
 }
-export declare function recoverPendingHandoffs(handoffsDir?: string, queueDir?: string): Promise<RecoveryResult>;
+export declare function recoverPendingHandoffs(handoffsDir?: string, queueDir?: string, bot?: Telegraf): Promise<RecoveryResult>;
 export declare function answerHandoff(taskId: string, answer: unknown, source: string, handoffsDir?: string): Promise<boolean>;
 export declare function cleanupHandoff(taskId: string, handoffsDir?: string): Promise<void>;

package/dist/bundled-plugins/awa-bundled/skills/orient/SKILL.md

@@ -0,0 +1,29 @@
+---
+name: orient
+description: At end-of-session, dispatches a read-only sub-agent to survey recent work and drafts a vaguely-specific priming prompt the next session can paste to orient itself and plan its next sprint. Produces orientation, not prescription — candidate directions are framed as questions, not decisions.
+context: load
+---
+
+## Sub-agent contract
+/contract
+
+Dispatch one read-only research sub-agent to survey recent work across four narrow targets, then produce a single markdown prompt block the user can paste into a fresh session.
+
+**Sub-agent surveys:**
+- Memory: the **`memory_search` tool** (2–3 keyword queries on the current work; FTS5 syntax) + hot memory at `~/.afk/state/memory/HOT.md`. Do **not** grep `~/.claude/projects/**/memory/` — that path is vestigial and empty on both trees; the archive is SQLite and only `memory_search` reaches it. Report which stores you consulted.
+- Git activity in this repo + sibling repos the session touched (`git log --oneline -10`, open PRs, dirty state)
+- `experiments/` docs (if present) relevant to the current work
+- Telemetry/state files the session interacted with (e.g. `~/.afk/agent-framework/*.jsonl`)
+
+**Return exactly this shape (≤ 60 lines total):**
+1. One-line project identity + one-line "where we are" (current phase / latest milestone)
+2. 3–5 bullets on what shipped recently — commits, PRs, empirical closures; be specific
+3. Memory pointers by filename only (next session's memory system auto-loads — don't restate)
+4. 2–4 **candidate directions** framed as questions or possibilities, **never** as a decided plan
+5. Closing line inviting the next session to run `/ground-state` first, then propose a plan
+
+**Discipline the contract enforces:** orientation, not execution. Handoff briefs drift toward prescription by default; the "questions-as-possibilities" framing is the guardrail so the receiving session plans with fresh judgment.
+
+Relay the prompt to the user. On request, save to `/tmp/orient-<ISO8601>.md`.
+
+**Skip when:** no substantive work to carry forward, user wants a bare `/compact`, or user already has a decided plan (use `/mint` or draft directly).