Back to Skills

implement-audit-trail

pjt222
Updated Yesterday
4 views
17
2
17
View on GitHub
Metaaidesigndata

About

This skill helps developers implement audit trail functionality in R projects for regulated environments like pharmaceuticals. It provides tools for logging, provenance tracking, electronic signatures, and data integrity checks to meet 21 CFR Part 11 compliance. Use it when you need tamper-evident analysis logs and to track who did what and when for regulatory submissions.

Quick Install

Claude Code

Recommended
Primary
npx skills add pjt222/agent-almanac -a claude-code
Plugin CommandAlternative
/plugin add https://github.com/pjt222/agent-almanac
Git CloneAlternative
git clone https://github.com/pjt222/agent-almanac.git ~/.claude/skills/implement-audit-trail

Copy and paste this command in Claude Code to install this skill

Documentation

Implement Audit Trail

Add audit trail capabilities to R projects for regulatory compliance.

When to Use

  • R analysis requires electronic records compliance (21 CFR Part 11)
  • Need to track who did what, when, and why in an analysis
  • Implementing data provenance tracking
  • Creating tamper-evident analysis logs

Inputs

  • Required: R project with data processing or analysis scripts
  • Required: Regulatory requirements (which audit trail elements are mandatory)
  • Optional: Existing logging infrastructure
  • Optional: Electronic signature requirements

Procedure

Step 1: Set Up Structured Logging

Create R/audit_log.R:

#' Initialize audit log for a session
#'
#' @param log_dir Directory for audit log files
#' @param analyst Name of the analyst
#' @return Path to the created log file
init_audit_log <- function(log_dir = "audit_logs", analyst = Sys.info()["user"]) {
  dir.create(log_dir, showWarnings = FALSE, recursive = TRUE)

  log_file <- file.path(log_dir, sprintf(
    "audit_%s_%s.jsonl",
    format(Sys.time(), "%Y%m%d_%H%M%S"),
    analyst
  ))

  entry <- list(
    timestamp = format(Sys.time(), "%Y-%m-%dT%H:%M:%S%z"),
    event = "SESSION_START",
    analyst = analyst,
    r_version = R.version.string,
    platform = .Platform$OS.type,
    working_directory = getwd(),
    session_id = paste0(Sys.getpid(), "-", format(Sys.time(), "%Y%m%d%H%M%S"))
  )

  write(jsonlite::toJSON(entry, auto_unbox = TRUE), log_file, append = TRUE)
  options(audit_log_file = log_file, audit_session_id = entry$session_id)

  log_file
}

#' Log an audit event
#'
#' @param event Event type (DATA_IMPORT, TRANSFORM, ANALYSIS, EXPORT, etc.)
#' @param description Human-readable description
#' @param details Named list of additional details
log_audit_event <- function(event, description, details = list()) {
  log_file <- getOption("audit_log_file")
  if (is.null(log_file)) stop("Audit log not initialized. Call init_audit_log() first.")

  entry <- list(
    timestamp = format(Sys.time(), "%Y-%m-%dT%H:%M:%S%z"),
    event = event,
    description = description,
    session_id = getOption("audit_session_id"),
    details = details
  )

  write(jsonlite::toJSON(entry, auto_unbox = TRUE), log_file, append = TRUE)
}

Got: R/audit_log.R created with init_audit_log() and log_audit_event() functions. Calling init_audit_log() creates the audit_logs/ directory and a timestamped JSONL file. Each log entry is a single JSON line with timestamp, event, analyst, and session_id fields.

If fail: If jsonlite::toJSON() fails, ensure the jsonlite package is installed. If the log directory cannot be created, check file system permissions. If timestamps lack timezone, verify %z is supported on the platform.

Step 2: Add Data Integrity Checks

#' Compute and log data hash for integrity verification
#'
#' @param data Data frame to hash
#' @param label Descriptive label for the dataset
#' @return SHA-256 hash string
hash_data <- function(data, label = "dataset") {
  hash_value <- digest::digest(data, algo = "sha256")

  log_audit_event("DATA_HASH", sprintf("Hash computed for %s", label), list(
    hash_algorithm = "sha256",
    hash_value = hash_value,
    nrow = nrow(data),
    ncol = ncol(data),
    columns = names(data)
  ))

  hash_value
}

#' Verify data integrity against a recorded hash
#'
#' @param data Data frame to verify
#' @param expected_hash Previously recorded hash
#' @return Logical indicating whether data matches
verify_data_integrity <- function(data, expected_hash) {
  current_hash <- digest::digest(data, algo = "sha256")
  match <- identical(current_hash, expected_hash)

  log_audit_event("DATA_VERIFY",
    sprintf("Data integrity check: %s", ifelse(match, "PASS", "FAIL")),
    list(expected = expected_hash, actual = current_hash))

  if (!match) warning("Data integrity check FAILED")
  match
}

Got: hash_data() returns a SHA-256 hash string and logs a DATA_HASH event. verify_data_integrity() compares current data against a stored hash and logs a DATA_VERIFY event with PASS or FAIL status.

If fail: If digest::digest() is not found, install the digest package. If hashes don't match for identical data, check that column order and data types are consistent between hashing and verification.

Step 3: Track Data Transformations

#' Wrap a data transformation with audit logging
#'
#' @param data Input data frame
#' @param transform_fn Function to apply
#' @param description Description of the transformation
#' @return Transformed data frame
audited_transform <- function(data, transform_fn, description) {
  input_hash <- digest::digest(data, algo = "sha256")
  input_dim <- dim(data)

  result <- transform_fn(data)

  output_hash <- digest::digest(result, algo = "sha256")
  output_dim <- dim(result)

  log_audit_event("DATA_TRANSFORM", description, list(
    input_hash = input_hash,
    input_rows = input_dim[1],
    input_cols = input_dim[2],
    output_hash = output_hash,
    output_rows = output_dim[1],
    output_cols = output_dim[2]
  ))

  result
}

Got: audited_transform() wraps any transformation function, logging input dimensions and hash, output dimensions and hash, and the transformation description as a DATA_TRANSFORM event.

If fail: If the transform function errors, the audit event is not logged. Wrap the transform in tryCatch() to log both successes and failures. Ensure the transform function accepts and returns a data frame.

Step 4: Log Session Environment

#' Log complete session information for reproducibility
log_session_info <- function() {
  si <- sessionInfo()

  log_audit_event("SESSION_INFO", "Complete session environment recorded", list(
    r_version = si$R.version$version.string,
    platform = si$platform,
    locale = Sys.getlocale(),
    base_packages = si$basePkgs,
    attached_packages = sapply(si$otherPkgs, function(p) paste(p$Package, p$Version)),
    renv_lockfile_hash = if (file.exists("renv.lock")) {
      digest::digest(file = "renv.lock", algo = "sha256")
    } else NA
  ))
}

Got: A SESSION_INFO event logged with R version, platform, locale, attached packages with versions, and the renv lockfile hash (if applicable).

If fail: If sessionInfo() returns incomplete package information, ensure all packages are loaded via library() before calling log_session_info(). The renv lockfile hash will be NA if the project does not use renv.

Step 5: Implement in Analysis Scripts

# 01_analysis.R
library(jsonlite)
library(digest)

# Start audit trail
log_file <- init_audit_log(analyst = "Philipp Thoss")

# Import data with audit
raw_data <- read.csv("data/raw/study_data.csv")
raw_hash <- hash_data(raw_data, "raw study data")

# Transform with audit
clean_data <- audited_transform(raw_data, function(d) {
  d |>
    dplyr::filter(!is.na(primary_endpoint)) |>
    dplyr::mutate(bmi = weight / (height/100)^2)
}, "Remove missing endpoints, calculate BMI")

# Run analysis
log_audit_event("ANALYSIS_START", "Primary efficacy analysis")
model <- lm(primary_endpoint ~ treatment + age + sex, data = clean_data)
log_audit_event("ANALYSIS_COMPLETE", "Primary efficacy analysis", list(
  model_class = class(model),
  formula = deparse(formula(model)),
  n_observations = nobs(model)
))

# Log session
log_session_info()

Got: Analysis scripts initialize the audit log at the start, log each data import, transformation, and analysis step, and record session info at the end. The JSONL log file captures the complete provenance chain.

If fail: If init_audit_log() is missing, ensure R/audit_log.R is sourced or the package is loaded. If events are missing from the log, verify that log_audit_event() is called after every significant operation.

Step 6: Git-Based Change Control

Complement the application-level audit trail with git:

# Use signed commits for non-repudiation
git config commit.gpgsign true

# Descriptive commit messages referencing change control
git commit -m "CHG-042: Add BMI calculation to data processing

Per change request CHG-042, approved by [Name] on [Date].
Validation impact assessment: Low risk - additional derived variable."

Got: Git commits are signed (GPG) and use descriptive messages referencing change control IDs. The combination of application-level JSONL audit trail and git history provides a complete change control record.

If fail: If GPG signing fails, configure the signing key with git config --global user.signingkey KEY_ID. If the key is not set up, follow gpg --gen-key to create one.

Validation

  • Audit log captures all required events (start, data access, transforms, analysis, export)
  • Timestamps use ISO 8601 format with timezone
  • Data hashes enable integrity verification
  • Session information is recorded
  • Logs are append-only (no deletion or modification)
  • Analyst identity is captured for each session
  • Log format is machine-readable (JSONL)

Pitfalls

  • Logging too much: Focus on regulated events. Don't log every variable assignment.
  • Mutable logs: Audit logs must be append-only. Use JSONL (one JSON object per line).
  • Missing timestamps: Every event needs a timestamp with timezone.
  • No session context: Each log entry should reference the session for correlation.
  • Forgetting to initialize: Scripts must call init_audit_log() before any analysis.

Related Skills

  • setup-gxp-r-project - project structure for validated environments
  • write-validation-documentation - validation protocols and reports
  • validate-statistical-output - output verification methodology
  • configure-git-repository - version control as part of change control

GitHub Repository

pjt222/agent-almanac
Path: i18n/caveman-lite/skills/implement-audit-trail
0
agentsagentskillsai-assisted-developmentclaude-codeskillsteams

Related Skills

content-collections

Meta

This skill provides a production-tested setup for Content Collections, a TypeScript-first tool that transforms Markdown/MDX files into type-safe data collections with Zod validation. Use it when building blogs, documentation sites, or content-heavy Vite + React applications to ensure type safety and automatic content validation. It covers everything from Vite plugin configuration and MDX compilation to deployment optimization and schema validation.

View skill

polymarket

Meta

This skill enables developers to build applications with the Polymarket prediction markets platform, including API integration for trading and market data. It also provides real-time data streaming via WebSocket to monitor live trades and market activity. Use it for implementing trading strategies or creating tools that process live market updates.

View skill

creating-opencode-plugins

Meta

This skill helps developers create OpenCode plugins that hook into 25+ event types like commands, files, and LSP operations. It provides the plugin structure, event API specifications, and implementation patterns for JavaScript/TypeScript modules. Use it when you need to intercept, monitor, or extend the OpenCode AI assistant's lifecycle with custom event-driven logic.

View skill

sglang

Meta

SGLang is a high-performance LLM serving framework that specializes in fast, structured generation for JSON, regex, and agentic workflows using its RadixAttention prefix caching. It delivers significantly faster inference, especially for tasks with repeated prefixes, making it ideal for complex, structured outputs and multi-turn conversations. Choose SGLang over alternatives like vLLM when you need constrained decoding or are building applications with extensive prefix sharing.

View skill