How to use the Analytical Method Validation agent skill

What the skill makes the agent do

The skill tells the agent to work in a fixed order. First it establishes which framework governs the work, such as ICH Q2(R2), USP <1225>, ICH M10, CLSI EP or ISO/IEC 17025, because these frameworks require different characteristics and study layouts. Then it insists that acceptance criteria are stated before any data are collected, since criteria chosen after seeing results are not acceptance criteria.

The scripts compute statistics on data you supply, including calibration response, accuracy and precision, detection and quantitation limits, and ICH M10 bioanalytical run checks. Each script reports computations and supported findings only.

• Plans the validation: which characteristics are required and what the protocol should contain
• Checks whether a calibration model holds across the range
• Reports recovery and how much variability is between days
• Calculates DL and QL by each allowed approach
• Flags supported ICH M10 numerical checks on a run

Install the skill from the ZIP or the Claude apps

Option A: download the ZIP from this page and unzip it into your personal skills folder, ~/.claude/skills/, or into a project's .claude/skills/ folder so that only that project uses it.

Option B: in the Claude apps, open Settings, go to Capabilities, and upload the skill under Skills.

The skill needs Python 3.11 or newer to run its scripts. The scripts use only the standard library, so no extra packages are listed in the source.

Trigger the skill with a natural request

You do not need special syntax. The skill's description lists trigger phrases such as "method validation", "validation protocol", "acceptance criteria", "linearity", "LOD", "LOQ", "Passing-Bablok", "method transfer", "ICH M10" and "ISR". Any request to show that an assay works also triggers it.

Name the technique, the attribute and the framework when you can. The more specific the request, the better the agent can choose the right characteristics.

• "Draft a validation protocol for an HPLC assay of impurities under ICH Q2(R2)."
• "Check whether my calibration data in calibration.csv are linear across the reportable range."
• "Set up an ICH M10 incurred sample reanalysis check for my LC-MS/MS run."

Run the planning command

The planning script can list the supported frameworks and generate a protocol with the required characteristics already filled in. Run these from the skill's scripts folder.

Every bracketed field in the generated protocol is a decision to record before data collection. Q2(R2) requirements depend on the measured attribute, such as assay, impurity, impurity-limit or identity, not on the technique alone.

Use the statistics scripts with your own data

The scripts expect controlled input data files, typically CSV, supplied by you. The source uses placeholder filenames, so substitute your own. All scripts accept --format table, tsv or json. Provenance and caveats go to stderr and data goes to stdout, so redirecting with > keeps them separate.

Exit code 0 means no findings, 1 means findings were raised and 2 means bad input, so the scripts can gate a workflow. Version 2.0 changed the input contract, so re-run saved analyses rather than comparing exit codes with version 1.x.

• check_response.py: calibration model across the range
• check_accuracy_precision.py: recovery and between-day variability
• check_detection_limits.py: DL and QL by allowed approach
• check_bioanalytical_run.py: supported ICH M10 numerical checks

Safety notes: scripts and network

This skill runs scripts. The source states it does not use the network. It also says the skill does not supply text from paywalled USP, CLSI or ISO documents and does not provide their numeric thresholds, so those numbers must come from an authorised copy you obtain yourself.

The skill is not a decision-maker. It explicitly does not decide that a procedure is validated, release a batch, accept or reject a run, or close an investigation. Only qualified reviewers and the quality unit can make those calls.

• Review any script before running it on controlled data, as with any downloaded code
• Do not ask the agent to retrieve or reconstruct paywalled standard text
• Keep controlled data files under your own quality system

Tips for getting useful output

State your acceptance criteria before you run any analysis, and give them to the agent in the request. The skill is built to flag decisions made after data collection.

Use the protocol and report templates as the structure for your documents. The report template states that any criterion that differs from the approved protocol is a deviation to document, not an edit to make.

• Give the intended purpose and the decision the result supports
• Supply complete labelled calibrator and QC records for ICH M10 runs
• For QL confirmation, supply explicit bias and CV limits