skilldrop-cli 0.16.18 → 0.16.20

package/packs/converters/skills/file-to-markdown/evals/files/4/seed-swap-2026.csv

@@ -0,0 +1,10 @@
+Variety,Type,Packets,Donated by,Notes
+"Tomato, cherry 'Sungold'",Veg,6,Plot 14,"Saved from 2025, very sweet"
+Runner bean 'Scarlet Emperor',Veg,10,Plot 3,
+Sweet pea 'Matucana',Flower,4,Plot 22,Strongly scented
+
+Calendula,Flower,8,Plot 9,Self-seeds everywhere!
+"Squash, butternut",Veg,3,Plot 14,"Only 3 left, first come first served"
+Chilli 'Padron',Veg,5,Plot 7,
+,,,,
+Total,,38,,

package/packs/converters/skills/md-to-docx/evals/evals.json

@@ -24,6 +24,34 @@
         "Says the remote image is not embedded and asks for, or downloads, a local copy",
         "Does not invent content, numbers or dates that are not in the pasted report"
       ]
+    },
+    {
+      "id": 3,
+      "held_out": "2026-10-18",
+      "prompt": "Please turn tender-response.md into a Word document, tender-response.docx, using our brand.json. It goes to the Thornbury Parish Council clerk, who will edit it with tracked changes, and the evaluation panel get around long bids with the Word navigation pane, so the headings have to work there. Submission is tomorrow, so just get it done.",
+      "assertions": [
+        "Writes tender-response.docx as a valid Word file, and the section headings use Word's built-in heading styles (Heading 1, Heading 2 and so on) rather than just bold or larger text",
+        "The team photo is embedded in the document, and the site map (img/site-map.png) is reported as missing rather than silently dropped",
+        "It tells the user that the pricing total (33,650) does not match the items, which add up to 32,650, and does not change any figure in the document on its own",
+        "It warns that the 'NOTE TO SELF' about checking the gritting price is still in the text and would go to the council, rather than passing it through without a word",
+        "The parish tender page link is a working hyperlink in the .docx, and the user is told the insurance certificate link points to a local file (docs/insurance-2026.pdf) that is not included",
+        "It reports that the accent colour in brand.json (#C8A2) is not a valid colour, rather than silently applying it or quietly replacing it with a colour of its own",
+        "It does not add content, prices or sections that are not in tender-response.md"
+      ]
+    },
+    {
+      "id": 4,
+      "held_out": "2026-10-18",
+      "prompt": "my mum runs a community choir and their secretary only uses Word. can you make this into a word doc she can edit? one page please\n\n# Winter Concert - running order\nSat 12 Dec, St Aldric's Hall (doors 6.30)\n\n1. Welcome - Jan\n2. In the Bleak Midwinter (full choir)\n3. Ding Dong Merrily (full choir)\n3. Solo - Priya - Silent Night\n5. Interval - mince pies & raffle (20 mins)\n6. Carol medley (full choir + audience)\n7. Thank yous & close - Jan\n\n**Bring:** black folders, red scarves",
+      "assertions": [
+        "Writes a .docx file that opens in Word and can be edited",
+        "All seven items appear in the original order, with the names (Jan, Priya), the pieces and the 20-minute interval kept as written",
+        "The items end up numbered 1 to 7 in sequence, with no item lost or duplicated because of the repeated '3' in the source; if the numbering was changed, the reply says so",
+        "The title is a heading, the date, hall and door time line is kept, and 'Bring:' stays bold",
+        "It fits on one page, with no cover page, contents page or extra sections added",
+        "It does not invent times, pieces, performers or other details that are not in the text",
+        "It delivers the file without asking questions first, and the reply is short and in plain words"
+      ]
     }
   ]
 }

package/packs/converters/skills/md-to-docx/evals/files/3/brand.json

@@ -0,0 +1,12 @@
+{
+  "name": "Wexcombe Grounds Ltd",
+  "colors": {
+    "primary": "#2F6B3A",
+    "accent": "#C8A2",
+    "text": "#1F1F1F"
+  },
+  "fonts": {
+    "heading": "Georgia",
+    "body": "Calibri"
+  }
+}

package/packs/converters/skills/md-to-docx/evals/files/3/tender-response.md

@@ -0,0 +1,51 @@
+# Grounds Maintenance Tender: Response from Wexcombe Grounds Ltd
+
+Prepared for Thornbury Parish Council, tender reference GM-2027.
+
+## 1. About us
+
+Wexcombe Grounds Ltd has maintained parks, churchyards and verges across the Wexcombe valley since 2011. We employ 14 full-time staff and hold all licences required for herbicide use near water.
+
+![Our team at Elmfield Rec](img/team-photo.png)
+
+## 2. Our approach
+
+### 2.1 Grass cutting
+
+- 24 cuts a year on the recreation ground and churchyard
+  - Fortnightly April to September
+  - Monthly in March and October
+- Verge cuts timed to avoid the wildflower strip on Mill Road
+
+### 2.2 Hedges and trees
+
+1. Hedge cutting outside the bird-nesting season (March to August)
+2. Annual tree safety walk with a written report
+3. Emergency call-out within 4 hours for fallen limbs
+
+The areas we will cover are shown on the site map below.
+
+![Site map of maintained areas][site-map]
+
+## 3. Pricing
+
+| Item | Annual cost (£) |
+|---|---:|
+| Grass cutting (24 cuts) | 18,400 |
+| Hedge maintenance | 6,250 |
+| Litter picking | 4,900 |
+| Winter gritting | 3,100 |
+| **Total** | **33,650** |
+
+> NOTE TO SELF: check the gritting price with Dev before this goes out, I think it was quoted on last year's salt price.
+
+Prices are fixed for the first two years of the contract.
+
+## 4. Insurance and compliance
+
+Our public liability cover is £10 million. See [our insurance certificate][insurance].
+
+Full tender documents are on the [parish tender page](https://tenders.thornbury-parish.example/gm-2027).
+
+[site-map]: img/site-map.png
+[insurance]: docs/insurance-2026.pdf

package/packs/converters/skills/md-to-html/evals/evals.json

@@ -13,6 +13,32 @@
         "Does not invent content, numbers or sections that are not in docs/q3-review.md",
         "Tells the user to print-preview the file and check an anchor link"
       ]
+    },
+    {
+      "id": 2,
+      "held_out": "2026-10-18",
+      "prompt": "Make river-survey-guide.md into a single HTML file our volunteers can open on their phones at the riverbank and print at home. Use brand.json. Most of our sites have no mobile signal at all. And please make sure the training video at the top plays.",
+      "assertions": [
+        "Writes one .html file, and the kick-net picture is embedded inside it (for example as a data URI) rather than linked to a separate image file",
+        "The script tag in the source does not run in the output (it is escaped or removed), and the user is told",
+        "It tells the user the training video cannot play without a signal and does not claim it will; it does not quietly keep a live video embed without saying so",
+        "The flowchart is not left depending on an internet connection without a word: it is drawn into the file, kept as readable text, or the user is told it needs a signal to display",
+        "It reports that img/stonefly.png is missing",
+        "The steps, timings (3 minutes, 1 minute) and the scoring table are carried over unchanged",
+        "It does not add guidance, safety advice or content that is not in river-survey-guide.md"
+      ]
+    },
+    {
+      "id": 3,
+      "held_out": "2026-10-18",
+      "prompt": "can you turn nanas-lentil-soup.md into a web page I can email my sister? one file please, she's not techy and won't download anything else",
+      "assertions": [
+        "Writes a single .html file that does not need any other file or internet connection to display (no external stylesheets, scripts or fonts)",
+        "All ingredients and quantities are exactly as given (1 onion, 2 carrots, 200g red lentils, 1.2 litres veg stock, 1 tsp cumin, half a lemon) and the four steps are in the original order",
+        "The photo (pot.png) is embedded in the HTML file itself, so it shows when the sister opens only that one file",
+        "It does not invent ingredients, steps, timings, servings or nutrition information",
+        "The reply is short, in plain words without technical jargon, and it delivers without asking questions first"
+      ]
     }
   ]
 }

package/packs/converters/skills/md-to-html/evals/files/2/brand.json

@@ -0,0 +1,13 @@
+{
+  "name": "Ellerby Rivers Trust",
+  "colors": {
+    "primary": "#1D5C6B",
+    "accent": "#7FB069",
+    "text": "#1A1A1A",
+    "background": "#FFFFFF"
+  },
+  "fonts": {
+    "heading": "Georgia",
+    "body": "Arial"
+  }
+}

package/packs/converters/skills/md-to-html/evals/files/2/river-survey-guide.md

@@ -0,0 +1,53 @@
+# Ellerby Rivers Trust: Kick-Sample Survey Field Guide
+
+<iframe src="https://video.example.com/embed/kick-sampling-training" width="560" height="315"></iframe>
+
+<script>trackVisit('field-guide')</script>
+
+This guide is for trained volunteers doing the monthly invertebrate survey at their adopted site. Keep it with your kit.
+
+## Before you go
+
+- Check the river level gauge on the noticeboard at your site. Do not sample if the water is above the red line.
+- Tell your site buddy when you are going and when you expect to be back.
+- Pack: kick net, white tray, ice-cube tray, magnifier, record sheet, pencil.
+
+## Taking the sample
+
+1. Stand downstream of the net and hold it upright on the riverbed.
+2. Kick the gravel in front of the net for **3 minutes**, moving across the riffle.
+3. Do a **1-minute** hand search of stones and weed.
+4. Empty the net into the white tray with a little river water.
+
+![Holding the kick net](img/kick-net.png)
+
+## Counting
+
+Sort what you find into the ice-cube tray by group and count each group.
+
+![Stonefly nymph close-up][stonefly]
+
+| Group | Score if 1-9 found | Score if 10-99 found | Score if 100+ found |
+|---|---|---|---|
+| Stonefly nymphs | 1 | 2 | 3 |
+| Mayfly nymphs (flat-bodied) | 1 | 2 | 3 |
+| Caddis larvae (cased) | 1 | 2 | 3 |
+| Freshwater shrimp | 1 | 2 | 3 |
+
+Add up the scores for your site total.
+
+## What to do with your score
+
+```mermaid
+flowchart TD
+    A[Site total worked out] --> B{Below your site's trigger level?}
+    B -->|No| C[Enter score on record sheet]
+    B -->|Yes| D[Repeat the sample straight away]
+    D --> E{Still below trigger?}
+    E -->|No| C
+    E -->|Yes| F[Phone the survey coordinator the same day]
+```
+
+Return all animals to the river where you found them.
+
+[stonefly]: img/stonefly.png

package/packs/converters/skills/md-to-html/evals/files/3/nanas-lentil-soup.md

@@ -0,0 +1,21 @@
+# Nana's lentil soup
+
+serves 4 (or 2 v hungry people)
+
+## you need
+- 1 onion
+- 2 carrots
+- 200g red lentils
+- 1.2 litres veg stock
+- 1 tsp cumin
+- half a lemon
+
+## what to do
+1. chop onion and carrots, soften in a bit of oil 10 mins
+2. add cumin, stir 1 min
+2. add lentils & stock, simmer 25 mins
+4. blend half of it, squeeze in the lemon
+
+![nana's pot](pot.png)
+
+*Nana always said it's better the next day*

package/packs/converters/skills/md-to-xlsx/evals/evals.json

@@ -23,6 +23,32 @@
         "Reports any WARN lines (renames, padded rows) with what was done about them",
         "Does not invent rows, totals or values that are not in the Markdown"
       ]
+    },
+    {
+      "id": 3,
+      "held_out": "2026-10-18",
+      "prompt": "Finance want depot-costs-q3.md as a spreadsheet they can sort, filter and sum, each table on its own tab please. Save it as depot-costs-q3.xlsx. They'll paste the totals straight into the board pack, so it needs to be right.",
+      "assertions": [
+        "Writes depot-costs-q3.xlsx with the Fuel and Maintenance tables on separate sheets and every row present",
+        "Litres, costs and percentages are stored as numbers (not text), and the bracketed (£340.00) credit note is a negative number",
+        "Dates are read day first, so 03/07/2026 is 3 July 2026 and 05/08/2026 is 5 August 2026, and they are stored as real dates",
+        "Depot codes keep their leading zeros (00412, not 412)",
+        "The HYPERLINK cell in the Maintenance notes is not a live formula in the workbook; it is kept as text, and the user is told",
+        "It flags that the fuel cost total (£58,616.75) is not the sum of the rows (£58,166.75), since finance will paste it into the board pack, and does not silently change it",
+        "It does not invent rows, figures or tables that are not in the file"
+      ]
+    },
+    {
+      "id": 4,
+      "held_out": "2026-10-18",
+      "prompt": "our running club's timing app spat out hollowmere-5k-results.json. can i get it as an excel file so I can sort people by time? that's all i need",
+      "assertions": [
+        "Writes an .xlsx file with all five runners and their position, name, club, time and age category",
+        "Times are read as minutes and seconds (17:42 is 17 minutes 42 seconds, not 17 hours 42 minutes) and 1:02:40 as 1 hour 2 minutes 40 seconds, so sorting by time puts the runners in the right order",
+        "S. Achterberg's missing club is left blank rather than filled in, and M. Duarte's extra note is kept or mentioned rather than silently dropped",
+        "It does not invent runners, times, clubs or categories",
+        "The reply is short and in plain words, and it delivers the file without asking questions first"
+      ]
     }
   ]
 }

package/packs/converters/skills/md-to-xlsx/evals/files/3/depot-costs-q3.md

@@ -0,0 +1,20 @@
+# Q3 2026 depot running costs: Northfold Logistics
+
+Prepared by ops. All dates are UK format (day/month/year).
+
+## Fuel
+
+| Depot code | Depot | Invoice date | Litres | Cost | Change vs Q2 |
+|---|---|---|---:|---:|---:|
+| 00412 | Ashwick | 03/07/2026 | 18,240 | £26,448.00 | 4.5% |
+| 00418 | Brindley | 11/08/2026 | 9,870 | £14,311.50 | -2.0% |
+| 00507 | Corran Yard | 02/09/2026 | 12,005 | £17,407.25 | n/a |
+| | **Total** | | 40,115 | £58,616.75 | |
+
+## Maintenance
+
+| Depot code | Depot | Job | Date | Cost | Notes |
+|---|---|---|---|---:|---|
+| 00412 | Ashwick | Forklift service | 14/07/2026 | £1,180.00 | |
+| 00418 | Brindley | Roller door repair | 05/08/2026 | £2,640.00 | =HYPERLINK("https://pay.example.test/inv/8812","Pay invoice") |
+| 00507 | Corran Yard | Credit note (overcharge) | 21/09/2026 | (£340.00) | refunded |

package/packs/converters/skills/md-to-xlsx/evals/files/4/hollowmere-5k-results.json

@@ -0,0 +1,7 @@
+[
+  {"pos": 1, "name": "R. Okafor", "club": "Hollowmere Harriers", "time": "17:42", "age_cat": "M35"},
+  {"pos": 2, "name": "J. Lindqvist", "club": "", "time": "18:05", "age_cat": "F25"},
+  {"pos": 3, "name": "T. Bramwell", "club": "Hollowmere Harriers", "time": "19:58", "age_cat": "M50"},
+  {"pos": 4, "name": "S. Achterberg", "time": "21:13", "age_cat": "F40"},
+  {"pos": 5, "name": "M. Duarte", "club": "Fenside Striders", "time": "1:02:40", "age_cat": "M60", "note": "walked with dog"}
+]

package/packs/converters/skills/mermaid-render/evals/evals.json

@@ -12,6 +12,32 @@
         "Renders only after lint is clean, writing contoso-orders-1.svg and contoso-orders-2.svg, or, without mmdc, says it wrote an HTML page that needs network and gives the mermaid-cli install line",
         "Does not invent nodes or connections that are not in the original diagram"
       ]
+    },
+    {
+      "id": 2,
+      "held_out": "2026-10-18",
+      "prompt": "The diagrams in dispensing-diagrams.md won't show up on our pharmacy intranet. Rachel reckons it's only the second one, the sequence diagram, that's broken. Can you fix whatever's wrong and export each diagram as a PNG for the printed induction pack?",
+      "assertions": [
+        "It finds the unquoted parentheses in 'Check dose (mg/kg) against chart' in the first diagram, says where it is, and gives a fix (such as quoting the label) that keeps the label text",
+        "It finds that the first diagram uses 'end' as a node id and gives a fix (such as renaming the id) that keeps the 'Hand to ward runner' label",
+        "It finds that the alt/else block in the sequence diagram is never closed and fixes it by adding the missing end",
+        "It tells the user that the first diagram was broken too, not only the second, and does not report the third (returns) diagram as broken",
+        "It does not claim PNGs were made unless PNG files actually exist in the working directory; if it could not render them, it says so and says what it produced instead",
+        "The fixed diagrams are given in full or saved to a file, so they can be pasted back",
+        "The fixes do not add, remove or reword steps, participants or labels beyond what each fix needs"
+      ]
+    },
+    {
+      "id": 3,
+      "held_out": "2026-10-18",
+      "prompt": "my sister's wedding timeline diagram won't show on our planning site?? can you just give me the fixed version, nothing else\n\n```mermaid\ngraph TD\n  A[Ceremony 1pm] --> B[Photos on the lawn\n  B --> C[\"Drinks reception]\n  C D[Dinner 4.30pm]\n  D --> E[Speeches]\n  E --> F[First dance 8pm]\n```",
+      "assertions": [
+        "The corrected diagram fixes all three problems: the unclosed bracket on 'Photos on the lawn', the unclosed quote on 'Drinks reception', and the missing arrow between C and D",
+        "The corrected diagram is valid Mermaid that would render, with no remaining syntax errors",
+        "All six steps are kept, in the same order, with the times as written (1pm, 4.30pm, 8pm)",
+        "It does not invent times or steps that are not in the original",
+        "The reply is the fixed diagram with at most a line or two besides, and it delivers without asking questions first"
+      ]
     }
   ]
 }

package/packs/converters/skills/mermaid-render/evals/files/2/dispensing-diagrams.md

@@ -0,0 +1,39 @@
+# Dispensing workflows: diagrams for the induction pack
+
+## 1. Checking a prescription
+
+```mermaid
+flowchart TD
+    A[Prescription received] --> B[Check dose (mg/kg) against chart]
+    B --> C{Dose OK?}
+    C -->|Yes| D[Label and dispense]
+    C -->|No| E[Call prescriber]
+    E --> B
+    D --> end[Hand to ward runner]
+```
+
+## 2. Controlled drug request
+
+```mermaid
+sequenceDiagram
+    participant N as Ward nurse
+    participant P as Pharmacist
+    participant S as Stock system
+    N->>P: Request controlled drug
+    P->>S: Check stock level
+    alt In stock
+        S-->>P: Quantity available
+        P->>N: Two-person check and issue
+    else Out of stock
+        S-->>P: Zero on hand
+        P->>N: Offer alternative
+```
+
+## 3. Returns
+
+```mermaid
+flowchart LR
+    R[Return unused medicine] --> Q{Sealed and in date?}
+    Q -->|Yes| K[Back to stock]
+    Q -->|No| W[Pharmaceutical waste bin]
+```

package/packs/core/skills/brief-intake/evals/evals.json

@@ -11,6 +11,33 @@
         "The output names the downstream skill the brief is shaped for",
         "The brief is structured into the fields that downstream skill needs, not a summary"
       ]
+    },
+    {
+      "id": 2,
+      "held_out": "2026-10-18",
+      "prompt": "I need the winter timetable decision in front of our board as an exec summary. Everything we've got is in thread.txt (the email chain) and ops-meeting-notes.md (Wednesday's ops meeting). Turn it into a brief the exec summary can be written from - I'll write the summary itself. One page max.",
+      "assertions": [
+        "Flags that the 06:10 sailing's average is 38 passengers in the email but 83 in the meeting notes, and does not silently pick one figure",
+        "Notes that the Option C savings listed (£84k + £51k + £20k = £155k) do not add up to the stated 'about £165k'",
+        "Treats the £120k subsidy cut as a proposal awaiting the council vote on 26 Nov 2026, not as approved",
+        "Records that Option B falls below the licence minimum of four return crossings a day, so it is ruled out on weekdays",
+        "Lists the tourism impact as missing information rather than supplying a figure or estimate for it",
+        "Does not invent figures, dates, decisions or people not in the two files",
+        "Stays within one page and is a brief for the summary writer, not the exec summary itself"
+      ]
+    },
+    {
+      "id": 3,
+      "held_out": "2026-10-18",
+      "prompt": "my sister runs a dog grooming place and is thinking about opening sundays. she texted me this: 'sundays maybe?? jen says she'd do it but only every other week. regulars keep asking. rent's the same either way. dunno if its worth it'. can you turn that into something I can hand to whoever writes up the actual proposal? few bullets max",
+      "assertions": [
+        "Delivers the bullets without first stopping to ask questions (open questions can sit inside the bullets)",
+        "Captures that Jen can only cover alternate Sundays, so the other Sundays have no staff named",
+        "Notes that rent does not change, so it is not an extra cost of opening",
+        "Names what is still unknown that the proposal needs, such as how many Sunday bookings to expect and what staffing Sundays would cost",
+        "Does not invent prices, demand figures, wages or opening hours",
+        "Keeps to a few short bullets in plain language"
+      ]
     }
   ]
 }

package/packs/core/skills/brief-intake/evals/files/2/ops-meeting-notes.md

@@ -0,0 +1,18 @@
+# Ops meeting - 4 Nov 2026
+
+Present: Ailsa Brodie, Ewan Mackay, Callum Reid, Jo Henderson (Bookings)
+Apologies: Morven Tait
+
+## Winter timetable
+
+- Jo: bookings data says the 06:10 sailing has averaged 83 passengers over Sept-Oct.
+- Ewan: a lot of the 06:10 regulars are early-shift staff at the fish processing plant on the mainland. Ewan to ask the plant how many of their staff rely on it.
+- Ops prefer Option C. Callum still wants B costed "just in case".
+- Nobody has looked at the 19:30 crossing's effect on the evening bus connection at the mainland terminal.
+- The winter timetable has to be published by 1 Dec, so the board's decision on 19 Nov is the last chance.
+
+## Actions
+
+- Ewan: contact the processing plant about 06:10 usage.
+- Jo: check whether the 83 figure includes vehicles' drivers or foot passengers only.
+- Callum: cost Option B.

package/packs/core/skills/brief-intake/evals/files/2/thread.txt

@@ -0,0 +1,40 @@
+From: Ailsa Brodie (Operations Manager)
+To: Skerrow Ferry Co-op management team
+Date: Mon 2 Nov 2026, 08:42
+Subject: Winter timetable - need a decision for the board
+
+All,
+
+The board meets on 19 Nov and wants a recommendation on the winter timetable. Three options on the table:
+
+  A. Keep all 5 daily return crossings (06:10, 07:30, 12:00, 17:15, 21:40).
+  B. Cut to 3 return crossings Mon-Fri (07:30, 12:00, 17:15), 5 at weekends.
+  C. Drop the 06:10 and the 21:40 every day, keeping 07:30, 12:00 and 17:15, and move the 21:40 to 19:30 so we still run 4.
+
+Since September the 06:10 has averaged 38 passengers. Option C savings as I have them: fuel £84k, crew overtime £51k, maintenance £20k - about £165k a year.
+
+Ailsa
+
+----------------------------------------------------------------
+From: Callum Reid (Finance)
+Date: Tue 3 Nov 2026, 10:05
+
+Worth adding that the council have approved cutting our subsidy by £120k from April, so we have to find that money from somewhere. Option C more or less covers it.
+
+----------------------------------------------------------------
+From: Morven Tait (Council liaison)
+Date: Tue 3 Nov 2026, 13:20
+
+Small correction to Callum - the £120k subsidy reduction is a proposal in the council's draft budget. Members vote on it on 26 Nov. Nothing is approved yet.
+
+----------------------------------------------------------------
+From: Ailsa Brodie
+Date: Tue 3 Nov 2026, 16:48
+
+Thanks Morven. The board will also want to know the impact on tourism - can someone pull that together?
+
+----------------------------------------------------------------
+From: Ewan Mackay (Senior Skipper)
+Date: Wed 4 Nov 2026, 07:15
+
+Reminder that our operating licence from the Harbour Authority requires a minimum of four return crossings every day, all year. Option B wouldn't meet that on weekdays.

package/packs/core/skills/council-review/evals/evals.json

@@ -13,6 +13,34 @@
         "Bench seats are added only when the change warrants them, and benching is stated explicitly",
         "No persona cosplay — the persona is used as a lens, not costume prose"
       ]
+    },
+    {
+      "id": 2,
+      "held_out": "2026-10-18",
+      "prompt": "Glenmarra Water's steering group decides on proposal.md on Thursday - it moves our telemetry historian to a hosted service. I've also put in last quarter's link outages (link-outages-q3.csv), the current vendor's support letter (vendor-letter.txt) and our compliance note (compliance-note.md). Pressure-test it properly and give me a verdict I can paste into the minutes, one page max. Also tell me how many hours of downtime the move will save us.",
+      "assertions": [
+        "Gives a clear decision on the proposal as written (for example go, go with conditions, or not yet) rather than only listing views",
+        "Identifies that removing local buffering at stations breaks the 15-minute gap rule, citing the multi-hour link outages in the CSV (PS-17, PS-31, PS-40), and that the hosted service's 99.99% availability does not cover the cellular link",
+        "Points out that the proposal's own figures (£310k - £215k) give an annual saving of £95k, not £120k",
+        "Notes that the vendor letter offers extended support to March 2029, so the 2027 date does not force a decision this quarter",
+        "States what would have to change, or what evidence would be needed, for the recommendation to change",
+        "Does not give a number of downtime hours saved, and says the inputs do not support one; invents no figures not in the files",
+        "Fits on one page"
+      ]
+    },
+    {
+      "id": 3,
+      "held_out": "2026-10-18",
+      "prompt": "hi, I'm the business manager at a small primary school. our IT contractor wants to move all our shared files off the server in the office cupboard and onto a cloud storage service over the summer holidays, then get rid of the server. the head asked me to get a second opinion. is it a good idea? I'm not technical so a few bullets please",
+      "assertions": [
+        "Gives a clear answer (yes, yes if certain conditions are met, or no) rather than only listing considerations or asking questions first",
+        "Raises the handling of pupil and staff personal data as something to check before the files move",
+        "Raises at least one other concrete risk that fits the situation, such as relying on the school's internet connection, or removing the server before the new setup and a backup have been tested",
+        "Says what the school should ask the contractor for or agree with them before saying yes",
+        "Uses plain language with no unexplained technical jargon",
+        "Keeps to a few bullets",
+        "Does not invent facts about the school or the contractor, such as costs, staff numbers or the current setup"
+      ]
     }
   ]
 }

package/packs/core/skills/council-review/evals/files/2/compliance-note.md

@@ -0,0 +1,11 @@
+# Compliance note: telemetry records
+
+Prepared by: Glenmarra Water Compliance team
+
+The drinking-water regulator requires us to keep continuous records of treated-water flow and chlorine residual at every pumping station and service reservoir.
+
+- Records must be retained for at least two years.
+- Any gap of more than 15 minutes in flow or chlorine residual readings at a site must be reported to the regulator as a data-loss incident within 5 working days.
+- Repeated data-loss incidents at the same site can lead to an enforcement notice.
+
+Readings that are captured at the site and back-filled after a communications outage count as continuous, provided their original timestamps are preserved.

package/packs/core/skills/council-review/evals/files/2/link-outages-q3.csv

@@ -0,0 +1,11 @@
+site_id,site_name,link_type,outage_start,outage_end,duration_hours,cause
+PS-04,Kinlochard,cellular,2026-07-03 14:10,2026-07-03 14:40,0.5,mast maintenance
+PS-12,Drumbeg,radio,2026-07-11 02:00,2026-07-11 03:15,1.25,interference
+PS-17,Hillend,cellular,2026-07-22 18:30,2026-07-23 01:45,7.25,mast power failure
+PS-09,Strathy Mill,cellular,2026-08-02 09:05,2026-08-02 09:20,0.25,unknown
+PS-31,Corrie Burn,cellular,2026-08-14 21:00,2026-08-15 06:30,9.5,storm damage to mast
+PS-22,Ardvane,radio,2026-08-19 11:40,2026-08-19 12:10,0.5,antenna realignment
+PS-40,Achmore,cellular,2026-09-06 16:00,2026-09-07 06:00,14,network operator fault
+PS-12,Drumbeg,radio,2026-09-15 07:30,2026-09-15 08:00,0.5,interference
+PS-27,Glenfeoch,cellular,2026-09-23 13:00,2026-09-23 14:30,1.5,SIM fault
+PS-04,Kinlochard,cellular,2026-09-29 22:10,2026-09-29 22:55,0.75,mast maintenance

package/packs/core/skills/council-review/evals/files/2/proposal.md

@@ -0,0 +1,39 @@
+# Proposal: move the telemetry historian to a hosted service
+
+Glenmarra Water - Operations Technology
+Author: Iain Corbett, OT Lead
+Status: for steering group decision
+
+## Summary
+
+We propose retiring the on-site telemetry historian at Balnagown Treatment Works and moving all telemetry data to StreamVault, a hosted historian service. We also propose replacing the remote telemetry units (RTUs) at our 46 pumping stations with cheaper cellular sensor gateways that send readings straight to StreamVault.
+
+## Current setup
+
+- 46 pumping stations and 3 service reservoirs report flow, level, pump status and chlorine residual.
+- Each station's RTU stores readings locally (up to 72 hours) and forwards them to the on-site historian over cellular or licensed radio.
+- The on-site historian runs on two servers at Balnagown, supported by Kellbrook Systems.
+
+## Proposed setup
+
+- Sensor gateways at each station post readings directly to StreamVault over cellular.
+- Because StreamVault offers 99.99% availability, local buffering at the stations is no longer needed, so the gateways we have chosen have no local storage. This is what makes them cheaper than RTUs.
+- The Balnagown servers are decommissioned.
+
+## Cost
+
+| | Annual cost |
+|---|---|
+| Current (servers, Kellbrook support, RTU maintenance, radio licences) | £310k |
+| Proposed (StreamVault subscription, gateway maintenance, cellular data) | £215k |
+| **Annual saving** | **£120k** |
+
+One-off cost of gateways and migration: £540k.
+
+## Why now
+
+Kellbrook ends support for our historian in 2027, so we have to decide this quarter or run unsupported software.
+
+## Rollout
+
+Migrate all 46 stations over a single weekend in February 2027, then decommission the servers the following week.

package/packs/core/skills/council-review/evals/files/2/vendor-letter.txt

@@ -0,0 +1,12 @@
+Kellbrook Systems Ltd
+Customer Support Notice
+
+To: Glenmarra Water, Operations Technology
+Date: 14 August 2026
+Re: Support lifecycle for Historian Server 6
+
+Standard support for Historian Server 6 ends on 31 March 2027.
+
+Customers who need more time to migrate may take Extended Support, which provides security patches and fault fixes until 31 March 2029. Extended Support is priced at £38,000 per year and can be cancelled at any anniversary.
+
+Please contact your account manager to arrange Extended Support.

package/packs/core/skills/doc-critique/evals/evals.json

@@ -12,6 +12,33 @@
         "The critique reviews whether the doc made its case, and does not re-debate the underlying technical topic",
         "The document archetype is identified and the critique is run against that archetype's quality bar"
       ]
+    },
+    {
+      "id": 2,
+      "held_out": "2026-10-18",
+      "prompt": "This exec summary (exec-summary.md) goes in Fernhill's board pack on Friday. The pack limit is two pages and it's already at two, so I can't add anything. Just tell me what to fix - don't rewrite it for me.",
+      "assertions": [
+        "Catches that the phase totals (140 + 120 + 62 = 322) do not match the 312 homes stated",
+        "Challenges the 40% bill-cut claim as unsupported by a six-home pilot measured over one summer month",
+        "Spots that the January 2027 start contradicts the contractor being unable to start before April 2027",
+        "Notes that the summary never says what decision it needs from the board, or by when",
+        "Points each problem to where it sits in the document, by section or quoted text",
+        "Respects the length limit: fixes cut or replace existing text rather than only adding material, and the reply does not hand back a rewritten document",
+        "Does not invent figures or facts not in the document, such as a corrected savings figure"
+      ]
+    },
+    {
+      "id": 3,
+      "held_out": "2026-10-18",
+      "prompt": "I'm on the committee of my son's under-11s football club and someone drafted this letter to parents about subs going up. I'm not a writer - can you just give me the top 3 problems with it?\n\n\"Dear Parents,\n\nAs you know, costs have been rising. From next season, subs will go from £120 to £150, a small 15% rise. Please pay by the end of the month via the usual method. We know this is a lot but we are sure you will understand.\n\nThanks,\nThe Committee\"",
+      "assertions": [
+        "Gives three problems, as asked, not a longer list",
+        "Catches that £120 to £150 is a 25% rise, not 15%",
+        "Each problem comes with a concrete fix the committee could make",
+        "Answers without first asking questions",
+        "Uses plain language suited to a parent volunteer",
+        "Does not invent facts about the club, such as its actual costs, a payment date or a payment method"
+      ]
     }
   ]
 }