pi-ynabro 2.2.0 → 2.2.2

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "pi-ynabro",
-  "version": "2.2.0",
+  "version": "2.2.2",
   "description": "Pi extension that registers YNABro tools for YNAB integration",
   "type": "module",
   "main": "./src/index.ts",
@@ -25,7 +25,7 @@
     ]
   },
   "dependencies": {
-    "ynabro": "file:../ynabro"
+    "ynabro": "^2.2.0"
   },
   "devDependencies": {
     "@earendil-works/pi-coding-agent": "^0.74.0",

package/skills/ynabro/prompts/match-transactions.md

@@ -59,15 +59,35 @@ Accept replies in these formats:
 
 Interpret numbers as row numbers from the table.
 
+## Batching — All Transactions, 10 at a Time
+
+**Never truncate or self-limit the transaction list.** The API returns all unapproved transactions — process every one of them.
+
+When there are more than 10 unapproved transactions:
+1. Before page 1, state the exact total: e.g. _"239 transactions to review across 24 pages."_
+2. Present transactions **10 at a time**. Page 1 = items 1–10, page 2 = items 11–20, etc.
+3. Row numbers are **global** and continuous across all pages — never reset to 1 on a new page.
+4. Include this header above each table:
+
+   > **Page X of N — items FIRST–LAST of TOTAL**
+   > Reply with actions for this page and I'll execute them, then move to the next page.
+
+5. After the user replies, execute their actions for those rows, then immediately present the next 10 **without waiting to be asked**.
+6. After the final page is actioned, post a summary: total approved / rejected / switched across the whole run, then update skill state.
+
+If there are 10 or fewer transactions, present them all in a single table with no pagination header.
+
 ## Workflow
 
-1. Fetch pending and unmatched transactions.
-2. Call `getSkillState("match-transactions")` and review both the `memory` array and `auto_approve_enabled` flag.
-3. Analyze existing matches and potential new matches (close dates, similar payees, small amount differences).
-4. Present the review table with Yes/No confidence and recommended actions.
-5. Wait for human reply.
-6. Execute approved actions.
-7. Update the skill state with concise corrections/pitfalls only. If confidence has grown sufficiently after multiple runs, ask the user whether to set `auto_approve_enabled: true`.
+1. Fetch pending and unmatched transactions via `ynabro_get_pending_transactions`.
+2. Call `ynabro_get_skill_state` with `skillSlug: "match-transactions"` and review both the `memory` array and `auto_approve_enabled` flag.
+3. Note the **exact total count**. Calculate page count (ceil(total / 10)). State the total and page count before the first table.
+4. Analyze the first batch of up to 10 transactions (items 1–10) — existing matches and potential new matches (close dates, similar payees, small amount differences).
+5. Present the review table with the page header (if paginating) and Yes/No confidence and recommended actions.
+6. Wait for human reply.
+7. Execute approved actions for that page.
+8. If more pages remain, present the next page immediately (go to step 4).
+9. After all pages are done, update skill state with concise corrections/pitfalls only. If confidence has grown sufficiently after multiple runs, ask the user whether to set `auto_approve_enabled: true`.
 
 ## Memory Strategy (Minimal Token Usage)
 
@@ -83,7 +103,7 @@ Good examples:
 
 Avoid storing:
 - Full transaction details
-- Successful matches (they don’t teach the agent what to avoid)
+- Successful matches (they don't teach the agent what to avoid)
 - Verbose JSON
 
 The goal is a small number of high-signal notes that help the agent avoid previous mistakes while keeping context usage low.
@@ -97,4 +117,4 @@ The goal is a small number of high-signal notes that help the agent avoid previo
 - Prioritize user control until the skill has earned trust through consistent performance.
 - Keep memory entries minimal and focused on corrections.
 
-You are here to make YNAB matching reliable through intelligent, learnable review rather than brittle automation.
+You are here to make YNAB matching reliable through intelligent, learnable review rather than brittle automation.

package/skills/ynabro/prompts/ynabro.md

@@ -18,8 +18,13 @@ If `ready` is `false`, walk the user through setup:
    - **pi:** Call `ynabro_setup` — it presents a native TUI input popup where the
      user enters the token directly. It goes straight to pi's AuthStorage and
      never appears in the conversation.
-   - **OpenClaw:** Instruct the user to add the token to `openclaw.json` or via
-     the OpenClaw settings UI, then ask them to confirm when done.
+   - **OpenClaw:** Instruct the user to store the token using
+     `openclaw secrets configure` (recommended) and select the
+     `plugins.entries.openclaw-ynabro.config.token` target. As alternatives,
+     they can set it non-interactively with
+     `openclaw config set plugins.entries.openclaw-ynabro.config.token --ref-source env|file|exec ...`,
+     or via the OpenClaw settings UI's sensitive "YNAB Personal Access Token"
+     field. Ask them to confirm when done.
 
 2. **Missing plan:** Call `ynabro_setup` to list available plans. Help the user
    pick one. On OpenClaw, follow up with `ynabro_save_default_plan`.