annotate-source-files
About
This skill automatically adds PUT workflow annotations to source files using the correct comment syntax for over 30 programming languages. It handles annotation generation, multiline comments, internal variables, and validation, making it ideal for documenting workflows in data pipelines or multi-step computations. Use it after analyzing a codebase and creating an annotation plan to embed workflow documentation directly into your source files.
Quick Install
Claude Code
Recommendednpx skills add pjt222/agent-almanac -a claude-code/plugin add https://github.com/pjt222/agent-almanacgit clone https://github.com/pjt222/agent-almanac.git ~/.claude/skills/annotate-source-filesCopy and paste this command in Claude Code to install this skill
Documentation
Annotate Source Files
Add PUT workflow annotations to source files. Putior can extract structured workflow data and generate Mermaid diagrams.
When Use
- After analyzing codebase with
analyze-codebase-workflowand having annotation plan - Adding workflow documentation to new or existing source files
- Enriching auto-detected workflows with manual labels and connections
- Documenting data pipelines, ETL processes, or multi-step computations
Inputs
- Required: Source files to annotate
- Required: Annotation plan or knowledge of workflow steps
- Optional: Style preference: single-line or multiline (default: single-line)
- Optional: Whether to use
put_generate()for skeleton generation (default: yes)
Steps
Step 1: Determine Comment Prefix
Each language has specific comment prefix for PUT annotations. Use get_comment_prefix() to find correct one.
library(putior)
# Common prefixes
get_comment_prefix("R") # "#"
get_comment_prefix("py") # "#"
get_comment_prefix("sql") # "--"
get_comment_prefix("js") # "//"
get_comment_prefix("ts") # "//"
get_comment_prefix("go") # "//"
get_comment_prefix("rs") # "//"
get_comment_prefix("m") # "%"
get_comment_prefix("lua") # "--"
Got: String like "#", "--", "//", or "%".
Line and block comments: putior detects annotations in both line comments (
//,#,--) and C-style block comments (/* */,/** */). For JS/TS, both//and/* */blocks scanned. Python triple-quote strings (''' ''') not detected — use#for Python annotations.
If fail: Extension not recognized? File language may not be supported. Check get_supported_extensions() for full list. For unsupported languages, use # as conventional default.
Step 2: Generate Annotation Skeletons
Use put_generate() to create annotation templates based on auto-detected I/O.
# Print suggestions to console
put_generate("./src/etl/")
# Single-line style (default)
put_generate("./src/etl/", style = "single")
# Multiline style for complex annotations
put_generate("./src/etl/", style = "multiline")
# Copy to clipboard for pasting
put_generate("./src/etl/", output = "clipboard")
Example output for R file:
# put id:'extract_data', label:'Extract Customer Data', input:'customers.csv', output:'raw_data.internal'
Example output for SQL:
-- put id:'load_data', label:'Load Customer Table', output:'customers'
Got: One or more annotation comment lines per source file, pre-filled with detected function names and I/O.
If fail: No suggestions generated? File may not contain recognizable I/O patterns. Write annotations manually based on understanding of code.
Step 3: Refine Annotations
Edit generated skeletons to add accurate labels, connections, metadata.
Annotation syntax reference:
<prefix> put id:'unique_id', label:'Human Readable Label', input:'file1.csv, file2.rds', output:'result.parquet, summary.internal'
Fields:
id(required): Unique identifier, used for node connectionslabel(required): Human-readable description shown in diagraminput: Comma-separated list of input files or variablesoutput: Comma-separated list of output files or variables.internalextension: Marks in-memory variables (not persisted between scripts)node_type: Controls Mermaid node shape and class styling. Values:"input"— stadium shape([...])for data sources and configuration"output"— subroutine shape[[...]]for generated artifacts"process"— rectangle[...]for processing steps (default)"decision"— diamond{...}for conditional logic"start"/"end"— stadium shape([...])for entry/terminal nodes
Example with node_type:
# put id:'config', label:'Load Config', node_type:'input', output:'config.internal'
# put id:'transform', label:'Apply Rules', node_type:'process', input:'config.internal', output:'result.rds'
# put id:'report', label:'Generate Report', node_type:'output', input:'result.rds'
Multiline syntax (for complex annotations):
# put id:'complex_step', \
# label:'Multi-line Label', \
# input:'data.csv, config.yaml', \
# output:'result.parquet'
Block comment syntax (for //-prefix languages only: JS, TS, Go, Rust, C, C++, Java, etc.):
Languages that use // for line comments also support PUT annotations inside /* */ and /** */ block comments. Use * put as line prefix inside block body:
/* put id:'init', label:'Initialize Config', output:'config.internal' */
/**
* put id:'process', \
* label:'Process Records', \
* input:'config.internal, records.json', \
* output:'results.json'
*/
function processRecords(config, records) {
// ...
}
JSDoc-style annotations particularly useful when documenting workflow steps alongside API documentation:
/**
* Transform raw sensor data into normalized readings.
* put id:'normalize', label:'Normalize Sensor Data', input:'raw_readings.json', output:'normalized.parquet'
*/
export function normalizeSensorData(readings: SensorReading[]): NormalizedData {
// ...
}
Note: Block comment annotations not supported for
#-prefix languages (R, Python, Shell) or---prefix languages (SQL, Lua). Use only line comments for those languages. Block-originated annotations do not support backslash continuation across lines.
Cross-file data flow (connecting scripts via file-based I/O):
# Script 1: extract.R
# put id:'extract', label:'Extract Data', output:'raw_data.internal, raw_data.rds'
data <- read.csv("source.csv")
saveRDS(data, "raw_data.rds")
# Script 2: transform.R
# put id:'transform', label:'Transform Data', input:'raw_data.rds', output:'clean_data.parquet'
data <- readRDS("raw_data.rds")
arrow::write_parquet(clean, "clean_data.parquet")
Got: Annotations refined with accurate IDs, labels, and I/O fields that reflect actual data flow.
If fail: Unsure about I/O? Use .internal extension for in-memory intermediates and explicit file names for persisted data.
Step 4: Insert Annotations into Files
Place annotations at top of each file or immediately above relevant code block.
Placement conventions:
- File-level annotation: Place at top of file, after any shebang line or file header comment
- Block-level annotation: Place immediately above code block it describes
- Multiple annotations per file: Use for files with distinct workflow phases
Example placement in R file:
#!/usr/bin/env Rscript
# ETL Extract Script
#
# put id:'read_source', label:'Read Source Data', input:'raw_data.csv', output:'df.internal'
df <- read.csv("raw_data.csv")
# put id:'clean_data', label:'Clean and Validate', input:'df.internal', output:'clean.rds'
df_clean <- df[complete.cases(df), ]
saveRDS(df_clean, "clean.rds")
Use Edit tool to insert annotations into existing files without disturbing surrounding code.
Got: Annotations inserted at appropriate locations in each source file.
If fail: Annotations break syntax highlighting in editor? Ensure comment prefix correct for language. PUT annotations are standard comments and should not affect code execution.
Step 5: Validate Annotations
Run putior's validation to check annotation syntax and connectivity.
# Scan annotated files
workflow <- put("./src/", validate = TRUE)
# Check for validation issues
print(workflow)
cat(sprintf("Total nodes: %d\n", nrow(workflow)))
# Verify connections by checking input/output overlap
inputs <- unlist(strsplit(workflow$input, ",\\s*"))
outputs <- unlist(strsplit(workflow$output, ",\\s*"))
connected <- intersect(inputs, outputs)
cat(sprintf("Connected data flows: %d\n", length(connected)))
# Generate diagram to visually inspect
cat(put_diagram(workflow, theme = "github", show_source_info = TRUE))
# Merge with auto-detected for maximum coverage
merged <- put_merge("./src/", merge_strategy = "supplement")
cat(put_diagram(merged, theme = "github"))
Got: All annotations parse without errors. Diagram shows connected workflow. put_merge() fills in any gaps from auto-detection.
If fail: Common validation issues:
- Missing closing quote:
id:'name→id:'name' - Using double quotes inside:
id:"name"→id:'name' - Duplicate IDs across files: each
idmust be unique across entire scanned directory - Backslash continuation on wrong line:
\must be last character before newline
Checks
- Every annotated file has syntactically valid PUT annotations
-
put("./src/")returns data frame with expected number of nodes - No duplicate
idvalues across scanned directory -
put_diagram()produces connected flowchart (not all isolated nodes) - Multiline annotations (if used) parse correctly with backslash continuation
-
.internalvariables appear only as outputs, never as cross-file inputs - Files excluded via
excludeparameter do not appear in workflow (e.g.,put("./src/", exclude = "test_")skips test helpers)
Pitfalls
- Quote nesting errors: PUT annotations use single quotes:
id:'name'. Double quotes cause parsing issues when annotation inside string context. - Duplicate IDs: Every
idmust be globally unique within scanned scope. Use naming convention like<script>_<step>(e.g.,extract_read,transform_clean). - .internal as cross-file input:
.internalvariables exist only during script execution. To pass data between scripts, use persisted file format (.rds,.csv,.parquet) as output of one script and input of next. - Missing connections: Diagram shows disconnected nodes? Check output filenames in one annotation exactly match input filenames in another (including extensions).
- Wrong comment prefix: Using
#in SQL file or//in Python causes annotation to be treated as code, not comment. Always verify withget_comment_prefix(). - Forgetting multiline continuation: When using multiline annotations, every continued line must end with
\and next line must start with comment prefix. - Python triple-quote strings: putior does not scan Python triple-quote strings (
''' ''',""" """). Always use#for Python PUT annotations. - Meta-pipeline annotations: Annotate build script that also scans for annotations (e.g., script that calls
put()andput_diagram())? Script's own annotations will appear in generated diagram. Either exclude file from scanning (seegenerate-workflow-diagramCommon Pitfalls) or avoid placing PUT annotations in build script itself.
See Also
analyze-codebase-workflow— prerequisite: produces annotation plan this skill followsgenerate-workflow-diagram— next step: generate final diagram from annotationsinstall-putior— putior must be installed before annotatingconfigure-putior-mcp— MCP tools provide interactive annotation assistance
GitHub Repository
Related Skills
content-collections
MetaThis 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.
polymarket
MetaThis 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.
creating-opencode-plugins
MetaThis 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.
sglang
MetaSGLang 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.
