6.2 KiB
description, tags
| description | tags | ||
|---|---|---|---|
| Identify potential ADRs for specific modules based on the codebase mapping (Phase 2) |
|
Launches the adr-analyzer agent to identify potential ADRs in specified modules.
When multiple modules are specified, launches agents in parallel for faster analysis.
Prerequisites: Run /adr-map first if mapping.md doesn't exist in the output directory.
What it does:
- Analyzes specified module(s) with strict architectural filtering
- Executes multiple modules in parallel when 2+ modules are specified
- Uses git history internally to enrich potential ADRs with temporal context
- Creates individual ADR files in priority-based folders (NO numbers in filenames)
- Weaves git insights naturally into content (dates, keywords, evolution)
- Updates the index with findings
Usage:
/adr-identify [module-ids] [--output-dir=PATH] [--adrs-dir=PATH]
Examples:
/adr-identify
# Prompts which modules to analyze (uses docs/adrs by default)
/adr-identify AUTH
# Analyze only AUTH module
/adr-identify AUTH DATA API
# Analyze multiple modules
/adr-identify --output-dir=output/adrs AUTH
# Analyze AUTH module with custom output directory
/adr-identify --adrs-dir=docs/adrs/generated AUTH
# Uses existing ADRs from custom directory for context and duplicate detection
Output Structure:
{OUTPUT_DIR}/potential-adrs/ # Default: docs/adrs/potential-adrs
├── must-document/ # High priority (score ≥100 out of 150)
│ └── MODULE-ID/
│ ├── decision-title-kebab-case.md
│ └── another-architectural-decision.md
└── consider/ # Medium priority (score 75-99 out of 150)
└── MODULE-ID/
└── medium-priority-decision.md
Important: Filenames use descriptive kebab-case WITHOUT numbers. Numbering happens in Phase 3 when formal ADRs are generated.
Existing ADR Integration: Automatically scans {OUTPUT_DIR}/generated/ (or --adrs-dir path) for existing ADRs to:
- Avoid duplicate identification (>70% similarity warning)
- Detect relationships with existing decisions (40-70% similarity)
- Provide timeline context (decision evolution, supersession patterns)
- Suggest consolidation opportunities (implementation details of existing ADRs)
How It Works:
- Scoring: 3 dimensions (Scope+Impact, Cost to Change, Team Knowledge), 150 points total (75 base score from Step 0 + 75 from dimensions)
- Thresholds: ≥100 pts (67%, high priority), 75-99 pts (50-66%, medium priority), <75 pts (discard)
- Filtering: Step 0 (Positive Identification) + Red Flags 1-5
- Git Integration: Enriches content with temporal context (dates, evolution, intent keywords)
Key Features:
- Step 0 (Positive Identification): Base technology decisions (infrastructure services, framework, ORM, API, auth, payment, AI/ML) automatically trigger ADR creation with base score
- Red Flag 5: Consolidates overly granular decisions into larger strategic ADRs (prevents ADR sprawl)
- Operational Impact: Captures observability, resilience, and production reliability decisions
Expected Results:
- Only ~5% of findings become ADRs (strict filtering)
- Potential ADRs enriched with git history (dates, evolution, intent keywords)
- No separate "Git History" section - insights woven naturally into content
Note: This identifies potential ADRs with evidence. You decide which to formally document in Phase 3.
Implementation Instructions
When the user invokes /adr-identify with module IDs:
Parse Arguments
Extract from command:
- Module IDs (e.g., AUTH, DATA, API)
--output-dir=<path>: Optional output directory (default:docs/adrs)--adrs-dir=<path>: Optional existing ADRs directory (default:{OUTPUT_DIR}/generated/)
Single Module (e.g., /adr-identify AUTH)
Launch a single adr-analyzer agent with the Task tool:
Without --output-dir:
Task tool:
- subagent_type: adr-analyzer
- prompt: "Identify potential ADRs for the AUTH module"
With --output-dir:
Task tool:
- subagent_type: adr-analyzer
- prompt: "Identify potential ADRs for the AUTH module with --output-dir=custom/path"
With --adrs-dir:
Task tool:
- subagent_type: adr-analyzer
- prompt: "Identify potential ADRs for the AUTH module with --adrs-dir=docs/adrs/generated"
Multiple Modules (e.g., /adr-identify AUTH DATA API or /adr-identify --output-dir=output/adrs AUTH DATA)
Launch multiple adr-analyzer agents in parallel using a single message with multiple Task tool calls:
Without --output-dir:
Single message with multiple Task tool calls:
Task tool call 1:
- subagent_type: adr-analyzer
- prompt: "Identify potential ADRs for the AUTH module"
Task tool call 2:
- subagent_type: adr-analyzer
- prompt: "Identify potential ADRs for the DATA module"
Task tool call 3:
- subagent_type: adr-analyzer
- prompt: "Identify potential ADRs for the API module"
With --output-dir:
Single message with multiple Task tool calls:
Task tool call 1:
- subagent_type: adr-analyzer
- prompt: "Identify potential ADRs for the AUTH module with --output-dir=custom/path"
Task tool call 2:
- subagent_type: adr-analyzer
- prompt: "Identify potential ADRs for the DATA module with --output-dir=custom/path"
Task tool call 3:
- subagent_type: adr-analyzer
- prompt: "Identify potential ADRs for the API module with --output-dir=custom/path"
IMPORTANT:
- When 2+ modules are specified, you MUST send a single message with multiple Task tool calls
- Each agent should analyze only ONE specific module
- Do NOT run agents sequentially - run them in parallel for better performance
- Each agent will create its own potential ADR files in the appropriate priority folders
- The index file will be updated by each agent independently
- All agents must use the same --output-dir if specified
No Modules Specified (e.g., /adr-identify or /adr-identify --output-dir=custom/path)
- Determine output directory (from --output-dir or default
docs/adrs) - Read
{OUTPUT_DIR}/mapping.mdto get available modules - Present the module list to the user
- Ask which module(s) they want to analyze
- Once specified, follow the logic above (single or multiple)