Extract Futureworks

Future-works analysis (audit + corpus mining)

A reusable future-works capability invoked by the academic agents. It mirrors the extract-statistic skill: two modes, one input contract each, and a strict boundary — it produces findings, the host agent judges them through its own deliberation step. The full pipeline and flag catalogue live in [references/futureworks-protocol.md](references/futureworks-protocol.md); the heading cues used to locate the relevant passages live in [references/section-cues.md](references/section-cues.md). Read both before using this skill. This file is the entry point and contract.

When to use

• Mode audit — a host auditor ([paper-auditor](../../agents/paper-auditor.md),

[scopus-auditor](../../agents/scopus-auditor.md), [thesis-auditor](../../agents/thesis-auditor.md), [thesis-proposal-auditor](../../agents/thesis-proposal-auditor.md)) reaches its future-works / hypothesis step. The skill audits the work's own stated future works and is then used to validate the work's hypotheses and to propose stronger ones drawn from the cited corpus.

• Mode mine — [scopus-researcher](../../agents/scopus-researcher.md) has downloaded the

corpus full text and wants every paper's stated future works extracted and synthesized, so the gap map, the Pareto matrix, and the hypotheses target real, author-declared open problems for the next research project.

When NOT to use

• As a standalone command typed by the user. The skill is invoked by the agents at their future-works

step, not directly. (It still runs if a user points it at a file, but its outputs feed a host pipeline.)

• For broad ideation unanchored to a manuscript or a corpus.

Cross-review boundary (do not run deliberation here)

This skill does not run a deliberation panel and does not call gemini_reviewer.py or github_reviewer.py. The host agents already run the mandatory [deliberation](../deliberation/SKILL.md) step once on their near-final output; the findings produced here are merged into the host plan and critiqued there. This keeps one panel per run.

Modes

Mode audit — one manuscript's own future works

• Input: a .tex or .md manuscript path. Resolve \input{}/\include{} recursively (up to 3

levels) and audit the merged document, exactly as the host auditors merge their input.

• Pipeline (see the protocol reference): locate the future-works / conclusion / limitations

section -> presence check -> per-statement testability and specificity audit -> link-to-limitation audit -> novelty check against Scopus -> hypothesis cross-check (validate the work's stated hypotheses against the corpus future works and propose stronger ones).

• Output:
• _futurework_report.md next to the manuscript (or

futurework_report_.md in the working directory for pasted text).

• A list of [FW …] flags the host folds into its plan (paper-auditor Section E, scopus-auditor

Section F1, thesis / thesis-proposal Section B).

• Never modify the manuscript directly. Corrections are applied by the host plan / latex-writer.

Mode mine — corpus future works from full text

• Input: a .bib (preferred) or an existing refs/ directory.
• Steps:

1. Ensure each retained paper's full text is present in any format - call

.claude/skills/scopus/scripts/download_pdf.py (presence-gated; PDF, then the HTML/Markdown any-format tiers).

1. Parse every present file with scripts/extract_text.py (--section-scan on) to get the

future-work / conclusion / limitations excerpts per paper.

1. Synthesize a corpus future-works table (one row per stated future-work item: paper,

statement, category, fit to the review theme/gap, effort 1-5, impact 1-5) and a research-opportunity list (recurring "X remains future work" across papers = high-value, author-declared gap). The host ranks the rows by a Pareto 80/20 score (lowest effort x highest impact first).

• Presence-gated: a paper whose full text could not be retrieved (status pdf-missing)

contributes title/abstract-level future works only and is flagged [FW FULLTEXT-MISSING]; it never blocks the pipeline.

• Output: _corpus_futurework.md (human-readable ranked table + opportunity list) and

_corpus_futurework.json (machine-readable, consumed by scopus-researcher). These route into scopus-researcher Step 9b (gap map), Step 9d (Pareto matrix), and Step 10 (hypotheses).

Prerequisites

• A Markdown/PDF backend importable for parsing (Docling preferred, else pymupdf4llm + pymupdf, else

the HTML tag-strip) - see .claude/skills/extract-statistic/scripts/requirements.txt. A missing backend degrades to abstract-level future works and flags it.

• For full-text retrieval in mine mode: .claude/skills/scopus/scripts/download_pdf.py reachable,

SCOPUS_API_KEY for the Elsevier source, and UNPAYWALL_EMAIL for the Unpaywall tier (the arXiv / PMC / Semantic Scholar tiers work without a key).

• For novelty / hypothesis lookups during the audit: scopus_api.py reachable. A network error is

flagged [SCOPUS UNAVAILABLE] and the audit proceeds without references.

Invocation

# Mode audit (host auditor):
python ".claude/skills/extract-statistic/scripts/extract_text.py" text "" --section-scan
# then apply the pipeline in references/futureworks-protocol.md and write _futurework_report.md

# Mode mine (scopus-researcher / auditor cited-corpus mining):
python ".claude/skills/scopus/scripts/download_pdf.py" bib "" --latex ""
python ".claude/skills/extract-statistic/scripts/extract_text.py" bib "" --latex "" --section-scan
# then synthesize _corpus_futurework.md + .json

The parser (extract_text.py) is shared with extract-statistic; this skill reuses its --section-scan output rather than shipping its own script. Pass --stats-scan --section-scan together when an agent mines statistics and future works in the same pass.

Resources

• references/futureworks-protocol.md - the audit + mine pipelines, the [FW …] flag catalogue, the

output format, and the Pareto ranking rule.

• references/section-cues.md - the English and French heading cues used to locate the future-work /

conclusion / limitations / open-problems passages.

• The shared parser .claude/skills/extract-statistic/scripts/extract_text.py (--section-scan); it

reuses download_pdf.py for any-format retrieval (it does not reimplement downloading).

Key rules

• Respond in the manuscript's language (French if the manuscript is French, English if English).
• Anti-AI-style hygiene follows .claude/skills/scientific-writing/references/writing_principles.md

(canonical): no em dashes, straight quotes only, no zero-width characters, no AI transition phrases, no overly perfect lists. Target an AI-style risk score below 10%.

• Never fabricate a future-work statement or a hypothesis; if the source does not state it, mark it

missing.

• Every proposed hypothesis must be testable by a named method and validated for novelty against

Scopus before it is offered (reuse the scopus_api.py search pattern).

• The Pareto 80/20 ranking is the host's: this skill supplies the per-paper future-work rows; the host

scores effort vs impact and maps each row to the review's themes and gaps.

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: LARi-UQAC
• Source: LARi-UQAC/ResearchTools
• License: MIT

Install and usage instructions live in the source repository linked above.