GDLD Quick Reference

Available Diagrams

!bash -c 'find docs/gdl -name "*.gdld" -maxdepth 3 2>/dev/null | while read f; do title=$(grep -m1 "^@diagram" "$f" 2>/dev/null | grep -o "title:[^|]*" | sed "s/title://"); echo "- $f${title:+ ($title)}"; done'

Format

@type|key:value|key:value|key:value

Every record is one line. Every field is self-describing (key:value). No schema lookup needed.

Core Record Types

Type	Purpose	Key Fields
@diagram	File metadata	id, type, purpose
@group	Subgraph	id, label, file, parent, pattern
@node	Graph node	id, label, group, shape, status, tags
@edge	Relationship	from, to, label, type, status, tags
@use-when	When to use	condition, threshold
@use-not	When NOT to use	condition, reason
@component	Key component	name, file, does
@config	Configuration	param, value, source
@gotcha	Lesson learned	issue, detail, fix, severity
@recovery	Failure handling	issue, means, fix, severity
@decision	Architectural decision	id, title, status, reason
@pattern	Related pattern	name, for, file
@entry	Entry point	use-case, command
@note	Freeform prose	context, text
@participant	Sequence actor	id, label, role
@msg	Ordered message	from, to, label, type
@block	Conditional/loop start	id, type, label
@endblock	Block end	id
@seq-note	Sequence annotation	over, text
@scenario	Diagram variant	id, label, inherits
@override	Element override in scenario	scenario, target, field, value
@exclude	Remove element from scenario	scenario, target
@view	Named perspective/filter	id, label, filter, includes, excludes, level
@include	File inclusion	file, records, prefix
@deploy-env	Deployment environment	id, label, provider
@deploy-node	Infrastructure node	id, label, env, parent, technology
@deploy-instance	Component-to-node map	component, node, instances, config
@infra-node	Non-app infrastructure	id, label, node, technology

Grep Patterns

# What is this diagram?
grep "^@diagram" file.gdld

# When to use / not use?
grep "^@use-when" file.gdld
grep "^@use-not" file.gdld

# What are the gotchas?
grep "^@gotcha" file.gdld

# How do I run it?
grep "^@entry" file.gdld

# What components are involved?
grep "^@component" file.gdld

# All nodes / edges
grep "^@node" file.gdld
grep "^@edge" file.gdld

# What connects to node X?
grep "@edge|to:X" file.gdld

# What does X connect to?
grep "@edge|from:X" file.gdld

# Decision points
grep "@node.*shape:diamond" file.gdld

# Cross-file: all gotchas
grep "^@gotcha" *.gdld

# Recovery procedures
grep "^@recovery" file.gdld

# Configuration values
grep "^@config" file.gdld

# Pattern references
grep "^@pattern" file.gdld

# Notes and caveats
grep "^@note" file.gdld

# Sequence: participants, messages
grep "^@participant" file.gdld
grep "^@msg" file.gdld

# Messages from/to a participant
grep "@msg|from:X" file.gdld
grep "@msg|to:X" file.gdld

# Decisions
grep "^@decision" file.gdld
grep "@decision.*status:accepted" file.gdld

# Elements by tag
grep "tags:.*pii" file.gdld
grep "tags:" file.gdld

# Deprecated or planned elements
grep "status:deprecated" file.gdld
grep "status:planned" file.gdld

# Cross-file edge references
grep "diagram-id#" *.gdld

# Severity filtering
grep "severity:critical" file.gdld
grep "severity:" file.gdld

# Scenarios and overrides
grep "^@scenario" file.gdld
grep "@override.*scenario:production" file.gdld

# Views
grep "^@view" file.gdld

# Includes
grep "^@include" file.gdld

# Deployment
grep "^@deploy-env" file.gdld
grep "^@deploy-node.*env:production" file.gdld
grep "^@deploy-instance" file.gdld
grep "^@infra-node" file.gdld

Combined Queries (Fewer Tool Calls)

Use regex unions to answer multi-type queries in a single grep:

# Full applicability picture (use + don't use)
grep -E "^@use-when|^@use-not" file.gdld

# Gotchas + recovery procedures together
grep -E "^@gotcha|^@recovery" file.gdld

# Complete graph (all nodes and edges)
grep -E "^@node|^@edge" file.gdld

# Implementation context (components + config)
grep -E "^@component|^@config" file.gdld

# All decision points + their branches
grep "shape:diamond" file.gdld && grep "type:conditional" file.gdld

# Cross-file: all gotchas and recovery across all diagrams
grep -E "^@gotcha|^@recovery" *.gdld

# Cross-file: every entry point across all systems
grep "^@entry" *.gdld

# Full deployment picture
grep -E "^@deploy-env|^@deploy-node|^@deploy-instance|^@infra-node" file.gdld

# Scenario overview
grep -E "^@scenario|^@override|^@exclude" file.gdld

# All critical issues across diagrams
grep "severity:critical" *.gdld

Prefer combined queries when you need multiple record types to answer a single question.

Encoding Guide

When converting architecture knowledge to GDLD, use this to choose the right record type:

Which record type?

Source Concept	Use	NOT
Hard-won lesson, "don't do X", key principle	@gotcha	@note
What to do when something breaks	@recovery	@gotcha
"When to use" with a measurable threshold	@use-when	@note
"When NOT to use" or anti-pattern	@use-not	(don't omit — see below)
Config value, threshold, parameter	@config	@note
Design decision with rationale	@decision	@note
Narrative context, rationale, caveats	@note	@gotcha
Reusable pattern referenced by other diagrams	@pattern	@note
Cross-cutting classification (pii, tier1, internal)	tags: field	separate @note
Dev/prod/error variant of same diagram	@scenario + @override	separate files
Named filter for specific audience	@view	ad-hoc grep
Shared records from another diagram	@include	copy-paste
Infrastructure/deployment target	@deploy-node	@node
Component-to-server mapping	@deploy-instance	@note
Non-app infra (LB, WAF, CDN)	@infra-node	@node
Severity of a gotcha or recovery	severity: field	separate @note

The @use-not rule

Every diagram that has @use-when records SHOULD also have @use-not records. Anti-patterns prevent misuse and are the most common omission. Ask: "When would someone reach for this pattern but be wrong to?"

# GOOD: explicit anti-patterns
@use-when|condition:processing 5+ documents|threshold:5+
@use-not|condition:single document processing|reason:state management overhead not justified
@use-not|condition:fast extraction under 2 minutes|reason:context compaction unlikely

# BAD: @use-when with no @use-not
@use-when|condition:processing 5+ documents|threshold:5+
# (missing — agents will hallucinate anti-patterns or say "none defined")

Gotcha vs Note vs Recovery

The most common encoding mistake. Use this test:

• Would another developer repeat this mistake? → @gotcha
• Did something break and here's how to fix it? → @recovery
• Is this just context or explanation? → @note

Source headings don't matter. "Key Principles", "Lessons Learned", "Important Notes", "Caveats" — if they describe mistakes to avoid, encode as @gotcha.

Example

# Document Processing Pipeline

@diagram|id:doc-pipeline|type:flow|purpose:multi-stage processing with state

# === WHEN TO USE ===
@use-when|condition:large document sets|threshold:10+ files
@use-when|condition:resumability needed
@use-not|condition:single-item processing|reason:overhead not justified

# === THE FLOW ===
@group|id:entry|label:Entry Stage|file:scripts/run.py
@group|id:agent|label:Processing Agent|file:src/agents/processor.py
@node|id:Input|label:Input Folder|group:entry
@node|id:CheckState|label:State exists?|shape:diamond|group:agent|role:decision
@node|id:LoadState|label:Load existing state|group:agent
@node|id:InitState|label:Initialize state|group:agent

@edge|from:Input|to:CheckState
@edge|from:CheckState|to:LoadState|label:Yes|type:conditional
@edge|from:CheckState|to:InitState|label:No|type:conditional

# === KEY COMPONENTS ===
@component|name:Entry point|file:scripts/run.py|does:CLI and preprocessing
@component|name:Agent|file:src/agents/processor.py|does:Multi-turn processing

# === GOTCHAS ===
@gotcha|issue:Output size limits|detail:Large outputs may be truncated|fix:Write to files
@gotcha|issue:State ordering|detail:Write result before updating state

# === ENTRY POINTS ===
@entry|use-case:Full processing|command:python scripts/run.py /path/to/input

Node Shapes

Shape	Use For
box	Process, action (default)
diamond	Decision, condition
circle	Start/end point
stadium	Input/output

Edge Types

Type	Use For
flow	Normal flow (default)
conditional	Yes/No branches
data	Data transfer
triggers	Event trigger

Tool Functions

Source the helpers:

source "${CLAUDE_PLUGIN_ROOT}/scripts/gdld-tools.sh"

Function	Usage	What it does
gdld_gotchas	gdld_gotchas FILE	List gotchas sorted by severity (SEVERITY|ISSUE|DETAIL|FIX)
gdld_nodes	gdld_nodes FILE [--group=G]	List node ID|LABEL pairs, optionally filtered by group
gdld_components	gdld_components FILE	List NAME|FILE|DOES
gdld_subgraph	gdld_subgraph FILE GROUP	Extract group + nodes + internal edges
gdld_filter	gdld_filter FILE --scenario=N	Apply scenario, output filtered GDLD
gdld_view	gdld_view FILE --view=N	Apply view filter, output filtered GDLD

Key Rules

• Each line is a complete, self-describing record
• grep "^@type" filters by record type
• grep "from:X" or grep "to:X" finds relationships
• Use # === SECTION === comments for organization
• Quantify conditions with threshold: not vague prose