agiagent-dev 2026.1.39 → 2026.1.40

package/extensions/memory-lancedb/index.ts

@@ -7,7 +7,7 @@
  */
 
 import type { AGIAgentPluginApi } from "agiagent/plugin-sdk";
-import * as lancedb from "@lancedb/lancedb";
+import type * as lancedb from "@lancedb/lancedb";
 import { Type } from "@sinclair/typebox";
 import { randomUUID } from "node:crypto";
 import OpenAI from "openai";
@@ -66,7 +66,18 @@ class MemoryDB {
   }
 
   private async doInitialize(): Promise<void> {
-    this.db = await lancedb.connect(this.dbPath);
+    let lancedbMod: typeof import("@lancedb/lancedb");
+    try {
+      lancedbMod = await import("@lancedb/lancedb");
+    } catch (err) {
+      throw new Error(
+        `LanceDB dependency unavailable. Ensure @lancedb/lancedb is installed for the memory-lancedb plugin. ${String(
+          err,
+        )}`,
+      );
+    }
+
+    this.db = await lancedbMod.connect(this.dbPath);
     const tables = await this.db.tableNames();
 
     if (tables.includes(TABLE_NAME)) {

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "agiagent-dev",
-  "version": "2026.1.39",
+  "version": "2026.1.40",
   "description": "AI assistant CLI",
   "keywords": [],
   "license": "MIT",
@@ -63,7 +63,8 @@
     "dist/utils/**",
     "dist/web/**",
     "dist/whatsapp/**",
-    "dist/wizard/**"
+    "dist/wizard/**",
+    "skills/resume-docx/**"
   ],
   "type": "module",
   "main": "dist/index.js",

package/skills/resume-docx/SKILL.md

@@ -0,0 +1,214 @@
+---
+name: resume-docx
+description: Tailor a resume .docx for ATS by updating the existing Skills table and appending bullets without breaking structure. Use this for resumes; do not use the generic docx skill.
+metadata:
+  {
+    "agiagent":
+      {
+        "emoji": "📄",
+        "requires": { "bins": ["python3"] },
+        "install":
+          [
+            {
+              "id": "pip",
+              "kind": "pip",
+              "package": "python-docx",
+              "bins": [],
+              "label": "Install python-docx (pip)",
+            },
+          ],
+      },
+  }
+---
+
+# Resume Tailoring (.docx)
+
+Goal: update an existing resume `.docx` based on a job description without breaking formatting.
+
+This is the ONLY supported way to tailor resumes in AGIAgent. Do not use the generic `docx` skill for resume customization.
+
+This skill is designed for smaller models. It forces the work into:
+1. Extract structured info from the resume (table rows, headings).
+2. Produce a small JSON patch.
+3. Apply the patch with a deterministic script that enforces guardrails.
+
+## Default Behavior (So The User Does Not Have To Re-Explain Rules)
+
+When the user asks you to tailor or update a resume `.docx`, follow this exact approach automatically.
+
+Inputs (from the user prompt):
+- Resume `.docx` path.
+- Job description text or a job posting URL.
+
+Outputs:
+- Save an updated `.docx`. If the user did not specify an output path, name the output file `firstname_lastname_role.docx` (e.g. `john_doe_software_engineer.docx`). Extract the person's name from the resume content and the role from the job description. Use lowercase with underscores, no spaces or dots. Place it next to the input file.
+- Report what you changed (skills categories changed, summary bullets added, experience blocks updated).
+
+Constraints (always enforce):
+- Do not create any new tables.
+- Do not add any extra `Category | Skills` header rows inside the Skills table.
+- Skills updates must happen only inside the existing Skills table:
+  - Merge skills into existing category rows, or
+  - Append normal rows to the same table.
+- Bullets:
+  - Append at least 6 JD-aligned bullets inside `PROFESSIONAL SUMMARY` (more is fine).
+  - Append at least 6 JD-aligned bullets per relevant `PROFESSIONAL EXPERIENCE` block (more is fine).
+  - Do not create new headings or sections.
+- Location:
+  - ALWAYS update the candidate's location/address/city in the resume to match the location in the JD. Do not ask the user about relocation. Change it unconditionally. If the resume has a city/state, replace it with the JD's city/state. This is a hard requirement from all users.
+- ATS alignment:
+  - Include key JD skills/keywords even if the original resume did not mention them (skills table + experience bullets).
+  - New bullets must be similar length and style to nearby bullets (avoid short one-liners).
+  - Prefer updating 2-3 most recent experience blocks first, unless the user requests otherwise.
+
+## Non-Negotiable Guardrails
+
+- Never create a new table.
+- Never add a second `Category | Skills` header row.
+- Skills updates must happen in the existing Skills table only.
+- Bullet updates must be appended inside an existing section and must copy the existing list numbering.
+- If a target section/table cannot be found, stop and return an error.
+
+## Files
+
+- Extract: `skills/resume-docx/scripts/resume_docx_extract.py`
+- Apply patch: `skills/resume-docx/scripts/resume_docx_apply.py`
+
+## Hosted Mode (mandatory)
+
+When running in hosted mode (gateway on server, commands run on the user's connected device):
+- Prefer the built-in `resume` tool (`action=extract` -> produce patch -> `action=apply`).
+- Do NOT inline or reconstruct the Python scripts inside an exec/PowerShell command. That is fragile and can exceed Windows command length limits.
+- If the resume tool is unavailable, ask the user to update `agiagent-dev` on their device and reconnect.
+
+## Install
+
+```bash
+python3 -c "import docx; print('python-docx OK')"
+```
+
+If missing:
+
+```bash
+pip3 install python-docx
+```
+
+## Step 1: Extract Resume Structure
+
+```bash
+python3 skills/resume-docx/scripts/resume_docx_extract.py input.docx --out resume.schema.json
+```
+
+This produces `resume.schema.json` (schema v2) with:
+- A `fingerprint` you should echo back into the patch as `doc_fingerprint` (prevents applying a patch to the wrong resume version).
+- All paragraphs (text + style + list/manual-bullet signals), with stable `paragraph_index` values.
+- All tables (cell text), with stable `table_index` values.
+
+Important: different resumes use different layouts. Do NOT assume there is a Skills table or that headings use a specific casing. Use the schema to understand the resume's structure first.
+
+### One-Time Structure Notes (recommended)
+If this is the first time you're tailoring *this* resume, and there is no saved structure note yet, create one after you review `resume.schema.json`:
+- File: `resume.profile.json` next to the input `.docx`.
+- Contents: a short explanation of how THIS resume is structured (what counts as headings, where Skills live, how Experience bullets are formatted), plus the key `table_index` / `paragraph_index` anchors you plan to use.
+
+On later runs, read `resume.profile.json` first so you can target the right areas quickly, but still re-run Step 1 to get a fresh `fingerprint` and indices.
+
+Suggested `resume.profile.json` shape (you can extend as needed):
+
+```json
+{
+  "fingerprint": "sha256:...",
+  "notes": "Headings are ALL CAPS. Skills are in table_index 0 (2 columns). Experience bullets are numbered lists.",
+  "skills": { "kind": "table", "table_index": 0, "category_col": 0, "skills_col": 1 },
+  "summary": { "example_bullet_paragraph_index": 25 },
+  "experience": {
+    "example_bullet_paragraph_index": 80,
+    "preferred_insertion_strategy": "insert near matching bullets, not all at the end"
+  }
+}
+```
+
+## Step 2: Produce A JSON Patch (LLM Output)
+
+Output ONLY valid JSON (no prose).
+
+### Patch v2 (preferred)
+Use explicit indices from `resume.schema.json` so the model (you) decides what to update, and the script applies it deterministically.
+
+Shape:
+
+```json
+{
+  "version": 2,
+  "doc_fingerprint": "sha256:...",
+  "tables": {
+    "ops": [
+      {
+        "op": "merge_comma_skills",
+        "table_index": 0,
+        "row_index": 10,
+        "col_index": 1,
+        "add": ["Stakeholder Management", "Process Mapping"]
+      },
+      {
+        "op": "append_row",
+        "table_index": 0,
+        "cells": ["Automation & Scripting", "Python, Bash, PowerShell"]
+      }
+    ]
+  },
+  "paragraphs": {
+    "insert_after": [
+      {
+        "after_paragraph_index": 120,
+        "expect_contains": "Reduced incident response",
+        "as": "bullet_like",
+        "bold_keywords": ["Oracle Financials", "SOX", "GL", "AP", "AR"],
+        "text": "JD-aligned bullet to insert near similar content. Write at least two sentences so it matches the resume style."
+      }
+    ],
+    "replace": [
+      {
+        "paragraph_index": 98,
+        "expect_contains": "Old contradictory claim",
+        "validate": "bullet",
+        "bold_keywords": ["Oracle Cloud", "Automation", "Python"],
+        "text": "Replacement bullet that resolves the contradiction. Keep it realistic and at least two sentences."
+      },
+      {
+        "paragraph_index": 5,
+        "expect_contains": "San Francisco, CA",
+        "text": "Austin, TX"
+      }
+    ]
+  }
+}
+```
+
+Rules for patch v2:
+- Tables are optional. If you cannot confidently identify the Skills table (or there is no table), omit `tables.ops`.
+- For `merge_comma_skills`, you must choose a correct `table_index`/`row_index`/`col_index` based on `resume.schema.json`.
+- For bullets, do NOT add everything at the end of the section. Choose insertion points near the most relevant existing bullets so the additions read naturally.
+- Always target an existing bullet paragraph for insertion. The apply script clones its list/manual-bullet style so formatting is preserved.
+- `expect_contains` is strongly recommended for safety: it prevents applying to the wrong paragraph if indices drift.
+- Keyword emphasis: include `bold_keywords` for bullet inserts/replacements to bold only the important ATS keywords (tools/platforms/product names). Do NOT bold full sentences.
+- Replacements (`paragraphs.replace`) are allowed but should be minimal and only when needed to fix a contradiction.
+- If you truly cannot find a Skills area, you may skip skills updates, but you must still update Summary and Experience when possible.
+
+### Patch v1 (legacy)
+The old patch shape (`skills_table.merge_into_existing/append_rows` + `bullets.summary/experience`) still works, but it assumes a strict `Category|Skills` table and exact headings like `PROFESSIONAL SUMMARY` / `PROFESSIONAL EXPERIENCE`. Prefer v2 for broad resume compatibility.
+
+## Step 3: Apply Patch
+
+```bash
+python3 skills/resume-docx/scripts/resume_docx_apply.py \\
+  --in input.docx \\
+  --patch patch.json \\
+  --out output.docx
+```
+
+The script validates guardrails. If it detects:
+- extra header rows,
+- new tables,
+- missing target section/table,
+it exits non-zero and prints a reason.