IB Deck Quality Checker
Audits investment-banking decks for numerical, narrative, language, and visual errors.
The skill explicitly says it is read-and-report only and performs no edits, while the repository requires professional review of outputs, reducing external-impact risk. However, it does not specify sensitive financial-data handling, data flow, user confirmation, isolation, rollback, or dependency permissions, so it is well below full marks.
The workflow, helper script, references, and slide-level attribution requirements are reasonably clear, and the script reports a missing input file. However, the static materials contain no key-path test suite, visual QC lacks a reproducible procedure, and the script's categorization and 5% tolerance can create false positives or negatives; the static ceiling therefore applies and deductions are warranted.
The audience, core scenarios, trigger language, and four review areas are fairly clear. Boundaries for non-fit cases, input limitations, PowerPoint add-in capabilities, and Chinese-language IB support are not sufficiently defined, and mainland-China network fit is not addressed, so only a moderate score is justified.
The documentation has a readable structure with environment checks, staged workflow, references, and a report template; the repository also provides an Apache-2.0 license and repository-level CI signals. The skill lacks its own versioning, changelog, named maintenance responsibility, update path, installation/dependency notes, and troubleshooting FAQ, which warrants deductions.
The skill addresses number consistency, narrative alignment, language, and visual QC, and its report template is directly usable for the core review task. However, visual verification is underspecified and the script only processes extracted text, so static evidence cannot establish reliable coverage of charts, overflow, contrast, or other visual issues; substantial human review remains necessary.
The source files, helper script, and report format are auditable, and repository CI includes plugin validation and secret scanning. Those workflows do not demonstrate the selected skill's key-path results, and there are no committed test fixtures, third-party execution records, or corroborating validation evidence, so verifiability remains limited.
- Treat the number-extraction script as an aid, not proof of financial correctness; manually confirm metric definitions, periods, units, and context.
- The visual-QC step lacks a concrete tool, input contract, and failure-feedback procedure; text extraction alone should not be assumed to detect overlap, overflow, contrast, or missing chart sources.
- The skill does not define handling, retention, or transmission boundaries for sensitive client materials; the deploying organization should verify data governance and access controls before use.
What it does & when to use it
This skill reviews pitch decks, client-ready presentations, and related investment-banking materials. It reads the deck slide by slide and checks four dimensions: number consistency, data-narrative alignment, IB language standards, and visual and formatting quality control. Findings are categorized by severity and attributed to specific slides. The workflow is read-and-report only and does not edit the deck.
Reads an uploaded .pptx file or a live open deck while preserving slide-level attribution; writes extracted text to a temporary file; runs scripts/extract_numbers.py to detect conflicts across revenue, EBITDA, multiples, margins, and other numeric categories; checks calculations, units, and time periods; tests whether charts support stated claims; scans for casual language, terminology drift, missing sources, axis labels, typography inconsistencies, number and date formatting drift, missing footnotes or disclaimers, overlaps, overflow, and contrast issues; produces a report grouped as Critical, Important, and Minor.
- An investment-banking analyst checks a pitch deck before sending it to a client and needs cross-slide number reconciliation.
- A deal team updates a chart and wants to verify that the written narrative still matches the data.
- A senior banker performs a final delivery review for language, sourcing, and formatting issues.
- A presentation owner needs to assess whether client materials are ready for delivery.
Pros & cons
- Covers numerical, narrative, language, and visual QC in one workflow.
- Requires slide-level attribution for reported findings.
- Designed specifically for investment-banking pitch decks and client materials.
- Does not modify the original presentation.
- The source provides no test results, sample report, or supported PowerPoint version.
- The workflow depends on scripts/extract_numbers.py and reference files whose full implementations are not shown.
- Visual verification is required, but no specific visual inspection tool is named.
- It reports issues but does not fix the presentation.
How to install
The skill is supplied through the financial-analysis plugin in anthropics/financial-services. In Cowork, open Settings → Plugins → Add plugin, paste https://github.com/anthropics/financial-services, and select the relevant vertical. In Claude Code, run: claude plugin marketplace add anthropics/financial-services, then claude plugin install financial-analysis@claude-for-financial-services. The source does not provide a standalone installation command for this skill.
How to use
Provide or open a .pptx deck in a supported environment, then ask: “Review this pitch deck for number consistency, data-narrative alignment, language, and formatting, and tell me whether it is client-ready.” You can also ask it to reconcile all figures across slides. It returns a read-only QC report.