- Home
- Skills
- Research & Analysis
- Manuscript Fact-Check
Works with the AI Agents you already use
Manuscript Fact-Check
Know that every number in your paper can be reproduced from the raw data, and which ones cannot, before a reviewer asks.
Free
Manuscript Fact-Check
Example session with this skill installed
The paper says "Genomes from Site A carried more chitinase genes than those from Site B (5.6 versus 3.0)". My table is demo_data.tsv. Check this claim.
- Read your context and instructions
- Compiled the manuscript fact-check
Claim check: chitinase genes, Site A vs Site B
Summary: the sentence at line 24 is wrong twice. The value for Site A is
5.0, not 5.6, and the
unit is missing (these are averages per genome). All the other numbers of the manuscript (abstract, Table 1, Figure 1 legend) are reproduced exactly from demo_data.tsv.
Claim (line 24): "Genomes from Site A carried more chitinase genes than those from Site B (5.6 versus 3.0)."
Recomputation (commands and output unedited)
Command: ../../Skill_redaction_en/source-verification/scripts/aggregate_table.py demo_data.tsv --group site --value chitinase_genes --operation mean
Rows kept after filters: 20
Site A mean = 5 (n values = 12, n rows = 12)
Site B mean = 3 (n values = 8, n rows = 8)
Is 5.6 another rule? Two alternatives tested, neither gives 5.6:
Command: ../../Skill_redaction_en/source-verification/scripts/aggregate_table.py demo_data.tsv --group site --value chitinase_genes --operation median
Rows kept after filters: 20
Site A median = 4.5 (n values = 12, n rows = 12)
Site B median = 2.5 (n values = 8, n rows = 8)
Command: ../../Skill_redaction_en/source-verification/scripts/aggregate_table.py demo_data.tsv --group site --value chitinase_genes --operation mean --filter genus!=Imaginomonas
Rows kept after filters: 14
Site A mean = 5.75 (n values = 8, n rows = 8)
Site B mean = 3.5 (n values = 6, n rows = 6)
Verdict per claim
| Line | Claim | Source and command | Recomputed | Verdict |
|---|---|---|---|---|
| 24 | Site A "5.6" | demo_data.tsv, column chitinase_genes, mean by site | 5.0 (n = 12 genomes) | Corrected: 5.6 is not reproducible |
| 24 | Site B "3.0" | same command | 3.0 (n = 8 genomes) | Verified, unit missing |
| 5–6 | Abstract "5.0 … per genome", "3.0 per genome" | same command | 5.0 / 3.0 | Verified (now contradicts line 24) |
| 14–15 | "Twenty genomes (12 from Site A and 8 from Site B)" | --group site --operation count | 12 / 8 | Verified |
| 26–27, 31–32 | "the mean was 4.2 chitinase genes per genome" | --value chitinase_genes --operation mean | 4.2 (n = 20) | Verified |
| Table 1 | 7.0, 4.5, 3.5 (Site A); 5.0, 2.0, 1.5 (Site B) | --group species --operation mean --filter "site=Site A" (then Site B) | identical | Verified |
Suggested wording (for your approval)
Genomes from Site A carried on average more chitinase genes per genome (5.0, n = 12) than those from Site B (3.0, n = 8).
This also keeps the promise made to Reviewer 2 (comment 2.2: every average keeps its unit).
Table 1 recomputed (Site A shown; Site B gives 5 / 2 / 1.5). The printed command keeps its quotes, so it can be copied as is:
Command: ../../Skill_redaction_en/source-verification/scripts/aggregate_table.py demo_data.tsv --group species --value chitinase_genes --operation mean --filter "site=Site A"
Rows kept after filters: 12
Bacillus exemplaris mean = 7 (n values = 4, n rows = 4)
Fictibacter sp. mean = 4.5 (n values = 4, n rows = 4)
Imaginomonas sp. mean = 3.5 (n values = 4, n rows = 4)
Next steps: approve the wording before it replaces line 24; keep the commands above in your notes.
Connects securely to your tools. The creator never sees your data.
About this skill
Scientific manuscripts often suffer from "drift" where numbers in the text no longer match the underlying data due to manual edits, version changes, or forgotten filters. This skill recomputes every numerical claim and metric definition directly from your raw data files to ensure absolute reproducibility.
What it does
- Claims auditing identifies every number, percentage, and software version within a manuscript for verification.
- Source mapping traces specific values back to their origin in raw TSV, CSV, or XLSX files, including specific columns and filters.
- Automated recomputation uses analysis scripts to reproduce values and logs the exact commands used for the audit trail.
- Consistency checking verifies that values remain identical across the abstract, results, discussion, and supplemental tables.
- Discrepancy reporting flags incorrect units, taxonomic level mismatches, double-counting errors, and non-verifiable claims.
How it works
- Inventory claims by listing each number and its intended measurement from the manuscript text.
- Execute recomputation using
scripts/aggregate_table.pyor custom logic to derive values from raw data files. - Compare results against the text to check for exact matches, unit consistency, and label accuracy.
- Classify status for each claim as verified, corrected, or not verifiable with a detailed reason for the author.
Frameworks & tools
This skill utilizes scripts/aggregate_table.py for data processing and supports standard data formats including TSV, CSV, and XLSX. It is designed for research environments involving pipeline logs, version files, and bibliographic references.
Why this beats prompting it yourself
Generic prompts often hallucinate plausible definitions to make numbers "fit" or fail to catch subtle unit errors like "per genome" versus "per sample." This skill enforces a strict verification protocol that prioritizes "not verifiable" flags over guesswork, mirroring a rigorous peer-review process.
Use cases
- Auditing a manuscript draft before submission to a high-impact journal.
- Responding to reviewer requests for precise metric definitions or units.
- Updating software versions and parameters in the Methods section from logs.
- Synchronizing values across a paper, cover letter, and response to reviewers.
Known limitations
Verification requires access to the raw data files and analysis scripts referenced in the text. Hand-filled spreadsheet columns without clear entry criteria will be flagged as non-verifiable.
Requires access to raw data files (TSV, CSV, XLSX) and pipeline logs. Cannot verify numbers derived from hand-filled columns lacking reproducible criteria.
How to use
- 1
Connect your agent to the Agensi MCP
Copy the prompt and paste it into Claude, ChatGPT, Cursor or any agent that supports MCP. It sets up the connection for you. Manual setup
- 2
Sign in to Agensi
Your agent shows a sign-in link. Sign in and click Allow.
- 3
Ask for what you need
Your agent loads this skill and uses it on your task. It always asks before any payment.
Reviews
No reviews yet
Be one of the first to try it. Every listed skill goes through the trust checks below.
Security scanned
Checked by our 8-point scan before listing
Fresh listing
Recently published to Agensi
Free forever
No account required to browse
Trust & safety
Security scan
Security scan passed 6 days ago
- Free with an Agensi account
Requires