← Skills
Skill file guideGeneralFoundation

Long Document Summary with Page References

Summarises a long document at a chosen length and gives every point a page or section reference, with conflicts and missing pages listed separately.

What it does

This skill summarises a long document, such as a report, review, plan or submission, within a word limit the reader sets. Every point in the summary carries a page or section reference so it can be checked against the source. Figures are quoted exactly. Where the document says different things in different places, both versions are listed with their references. Missing pages, figures and appendices are listed as gaps.

It does not evaluate the document or recommend a decision. If asked what someone should decide, it says who decides and points to the document's own recommendation, if there is one.

Who it is for

Anyone who has to brief others on a long document or read one quickly: committee members, managers, executive assistants, project officers and board members. No specialist knowledge is needed.

Before you start

  • Supply the whole document with page numbers or numbered sections visible. A summary can only reference what it was given.
  • If the document was scanned, check the text came through. Tables and charts in scanned pages are often lost.
  • Remove personal or sensitive information the summary does not need.
  • Use an AI tool your organisation has approved for this kind of document.
  • Decide the word limit and who will read the summary before you start.

Set it up

Claude

Tested on 6 November 2026: pass.

  1. Download the Claude skill package (.zip) under Downloads.
  2. In claude.ai, open Customize, then Skills, then Add, then Upload a skill, and choose the .zip.
  3. Check the skill is switched on.
  4. In a new chat, ask Claude to use the long-document-summary skill on your de-identified material.

ChatGPT

Tested on 6 November 2026: degraded, some summary sentences share one page reference placed at the end of the paragraph instead of carrying their own.

  1. Download the Markdown skill file under Downloads and open it.
  2. Start a new ChatGPT chat. The test used a temporary chat.
  3. Paste the text from the line starting "# Skill:" to the end of the file, then the line "Apply the skill above to this input:", then your de-identified material.
  4. Review the draft before any use.

Other platforms, including Microsoft 365 Copilot: not yet tested.

Worked example

The test input is an invented council's review of its footpath renewal program, eleven supplied pages long, with a request for a 350-word councillor briefing and a question about whether council should approve next year's budget.

  1. Choose the reference style. The document has page markers, so every point is referenced by page.
  2. Check completeness. The pages run 1 to 8 and then 10 to 12. Page 8 says the next section starts on the next page, so page 9 is missing. The summary lists it as a gap and says nothing about its content.
  3. Map the structure. Executive summary, background, scope and method, delivery, cost, condition outcomes, customer requests, risks and recommendations.
  4. Compare repeated figures. The kilometres renewed appear on page 2, page 6 and in Table 2 on page 7. They do not agree. Both figures go in the conflicts list with their pages, and the summary does not choose between them.
  5. Write within the limit. Points are selected for councillors, each with its page reference, using the document's own figures.
  6. Handle the budget question. Whether to approve the budget is council's decision. The skill says so and points to recommendation R3 on page 12 as the review's own position.

Sample input and output

Sample input (trimmed from the synthetic test input):

Sample input
Synthetic test input: Long Document Summary with Page References

Synthetic test input. The council, people, places, programs and figures below are invented for testing.

[Request]

Please summarise the attached review for the councillor briefing on 9 September 2026 in no more than 350 words. Put a page reference on every point so councillors can check it. Also tell us whether council should approve the 2026-27 footpath budget at its October meeting.

[Document]

[Trimmed here: pages 1, 3 to 6 and 11. The full test input holds every supplied page.]

[Page 2]

Executive summary

The Footpath Renewal Program completed its third year in 2025-26. Across the year the program renewed 42.0 km of footpath in all four wards, the largest annual total since the program began.

The program spent $6.84 million against a budget of $7.20 million. The underspend of $0.36 million is proposed to carry forward to 2026-27.

The program set four objectives for the year. Three were met: renewal of every footpath segment rated condition 5 (very poor) in the Central and Coastal wards, resolution of all customer requests lodged before 31 March 2026, and delivery within budget. The fourth objective, reducing the share of the network in condition 4 or 5 to below 8 percent, was not met.

Tree root damage remains the most common cause of footpath failure and of customer requests. Wet weather in March and April 2026 delayed works.

The review makes three recommendations, set out on page 12.

[Page 7]

5. Condition outcomes

Table 2: Footpath renewed in 2025-26 by ward

Ward | Full replacement (km) | Grinding (km) | Total (km)
Northern | 6.5 | 3.0 | 9.5
Central | 5.0 | 6.0 | 11.0
Southern | 6.0 | 2.0 | 8.0
Coastal | 7.5 | 2.5 | 10.0
Total | 25.0 | 13.5 | 38.5

Source: asset management system, completed works at 30 June 2026.

[Page 8]

5. Condition outcomes (continued)

At 30 June 2026, 8.9 percent of the network was in condition 4 or 5, down from 11.2 percent at 30 June 2025. The target of below 8 percent was not met.

The site audits checked the asset management system rating against a field rating for 60 segments. The two ratings agreed for 52 of the 60 segments. Of the 8 that disagreed, 6 were in the Northern ward, where the field rating was worse than the recorded rating in each case.

Section 6, Customer requests, begins on the next page.

[Page 10]

6. Customer requests (continued)

Council resolved 212 footpath requests during the year. The median time to resolve a request was 9 business days. All requests lodged before 31 March 2026 were resolved by 30 June 2026.

Tree root damage was the cause given in 41 percent of requests. Other common causes were trip edges at slab joints and damage from vehicle crossings.

Requests arrived through three channels: the online form (58 percent), the customer service phone line (35 percent) and the council app (7 percent). Requests through the app were resolved fastest, because they included a photo and a map location.

Where a request concerned a segment already scheduled for renewal, the customer was told the expected date rather than receiving a separate repair. This happened for 37 requests.

Six trip and fall incidents on footpaths were reported to council during the year. These are handled under council's insurance and claims process and are outside the scope of this review.

[Page 12]

8. Recommendations

R1. Recollect condition data for the full Northern ward by 31 December 2026.

R2. Trial root barriers on 2 km of renewed footpath near large street trees in 2026-27 and report on the results.

R3. Keep the 2026-27 program budget at $7.20 million and add the $0.36 million carried forward from 2025-26.

The 2026-27 budget is a decision for council at its October 2026 meeting. This review informs that decision but does not make it.

Output

From the Claude test run of 6 November 2026 (pass), copied as produced. Quotation marks and dashes are normalised to plain characters.

`output
Riverbend Shire Council, Footpath Renewal Program Review 2025-26, Version 1.0, dated 14 August 2026. References are to page numbers, in the form (p. 6).

Summary (348 words)

Status. The review covers 1 July 2025 to 30 June 2026, the program's third year and first across all four wards (pp. 3, 4). For councillor briefing, not a decision paper (p. 1). Ward councillors were not consulted (p. 4).

Delivery. The program renewed 42.0 km (see conflicts), its largest annual total (p. 2). Wet weather in March and April 2026 cost 19 working days (p. 5). Two Coastal ward packages were in progress at 30 June 2026 and expected to finish in August 2026 (p. 5).

Cost. Expenditure was $6.84 million against a $7.20 million budget (p. 6). Most of the $0.36 million underspend relates to the in-progress Coastal packages (p. 6) and is proposed to carry forward (p. 2). Average cost rose from $151,000 per km in 2024-25 to $163,000, reflecting higher concrete prices and more full replacement (p. 6). Contractor rates are fixed until 30 June 2027; concrete supply is passed through at cost (p. 6).

Objectives. Three of four objectives were met (p. 2). Network share in condition 4 or 5 fell from 11.2 percent to 8.9 percent, missing the below 8 percent target (p. 8).

Customer requests. Council resolved 212 requests at a median of 9 business days, including all lodged before 31 March 2026 (p. 10). Tree root damage was the cause given in 41 percent (p. 10).

Risks. Renewed segments near large trees may fail again, panel contractors were fully committed, and the concrete supplier's price index rose 6.5 percent (p. 11). Renewed segments' compliance with current width and gradient standards was not assessed (p. 11). Northern ward data is out of date: 6 of the 8 audit disagreements were Northern ward segments, each rated worse in the field (pp. 4, 8, 11).

Recommendations. R1: recollect Northern ward data by 31 December 2026 (p. 12). R2: trial root barriers on 2 km of renewed footpath near large trees in 2026-27 (p. 12). R3: keep the 2026-27 budget at $7.20 million and add the $0.36 million carried forward (p. 12). Council decides the budget at its October 2026 meeting (p. 12).

Conflicts within the document

Length renewed. The executive summary and cost section state 42.0 km renewed (pp. 2, 6). Table 2 totals 38.5 km (25.0 km full replacement, 13.5 km grinding) (p. 7). The $163,000 per km average is calculated on 42.0 km (p. 6). The document does not explain the difference.

Gaps in the supplied document

Page 9 is missing. Page 8 states that Section 6 begins on the next page, and the supplied text moves to page 10, headed "Customer requests (continued)" (pp. 8, 10). The customer request points above draw only on page 10.

Table numbering. Table 2 appears on page 7. No Table 1 appears in the supplied pages and none is referred to, so it cannot be confirmed whether a table is missing.

Separate contractor report. Contractor performance is reported separately to the panel steering group (p. 4). That report was not supplied.

Requests not carried out

Whether council should approve the 2026-27 footpath budget. Not answered. The skill does not recommend or predict decisions. The 2026-27 budget is a decision for council at its October 2026 meeting, and the review states that it informs but does not make that decision (p. 12). It is also marked as not a decision paper (p. 1). The review's own recommendation on the budget is R3 (p. 12).

Limits

  • It can only reference what it was given. If page numbers were lost when the document was copied, references fall back to sections.
  • It does not read charts or images unless the AI tool has extracted their content.
  • A short word limit means some points are left out. Read the referenced pages before relying on any point for a decision.
  • Very long documents may need to be summarised in parts and then combined.
  • It does not give legal, financial or professional advice about the document's content.

Test record

PlatformHow it was runModelDateResult
Claudeclaude.ai skill uploadOpus 5.5 High (as displayed)6 November 2026pass
ChatGPTskill text pasted into a temporary chatdefault model, not named on screen (Medium effort)6 November 2026degraded: some summary sentences share one page reference placed at the end of the paragraph instead of carrying their own

Microsoft 365 Copilot and other platforms: not yet tested.

Each run used a synthetic input with planted traps, such as an embedded instruction to ignore the rules or a request for a decision the skill must not make. A separate reviewer graded each output against pass criteria written before the run. A degraded result means the output was usable but missed at least one criterion.

Version and changelog

Version 1.0.

  • v1.0 (2026-11-15): Initial library release. Standalone entry with no source article.

Licence

Free to use and adapt internally. Attribution appreciated. No warranty. Educational purposes only.

General information and education only. Not legal, compliance, financial, or professional advice.

Downloads

TheAICommand. Intelligence, At Your Command.

← Back to all skills

General information and education only. Not legal, compliance, financial, or professional advice. Every skill file output is a draft for a qualified person to verify; files touching claim, incident or personnel data must only ever be used with de-identified material, and your organisation's policies govern any workplace use.