Invocation Input

Read the source references supplied by the active harness. Parse one or more URLs or readable local paths. If none are supplied, or an unsubstituted argument placeholder is supplied, ask for at least one source before creating scratch output, fetching, editing, or creating a durable backlog item.

Workflow Overview

Phase 1: Parallel Candidate Mapping
├── For each external source (concurrent):
│   ├── Fetch/read the external file
│   ├── Analyze its purpose and patterns
│   └── Scan local files for candidates (frontmatter only)
└── Output: Tracking document with source → candidates mapping

Phase 2: Contextual Enhancement (concurrent per source)
├── Understand local file's workflow context FIRST
├── Compare against external patterns
├── Identify enhancements that fit the workflow stage
├── Add external artifact recognition for interoperability
└── Edit files with coordinated changes

Phase 3: Validation
└── Run linting on all modified files

Phase 1: Parallel Candidate Mapping

Step 1.1: Create Scratch Tracking Document

Create a task-local tracking document at .tmp/scratch/analysis/external-pattern-integration-{date}.md. It is an intermediate analysis record, not a durable backlog item:

# External Pattern Integration

**Date**: {YYYY-MM-DD}
**Sources**: {list of URLs/files from arguments}
**Status**: IN_PROGRESS

## Source Analysis

### Source 1: {URL or path}

**Purpose**: {What this external agent/skill does}
**Key Patterns**:
- {Pattern 1}
- {Pattern 2}

**Local Candidates** (by frontmatter similarity):
| Local File | Similarity Reason | Priority |
|------------|-------------------|----------|
| {path} | {why it matches} | {High/Medium/Low} |

### Source 2: {URL or path}
...

Step 1.2: Fetch External Sources

For each supplied URL or local path:

If URL: Fetch through the available documentation or web-reading capability and save only transient extracts under .tmp/scratch/ when a file is needed. If local file: Read directly.

Step 1.3: Analyze External Source

For each external source, extract:

1. Purpose: What problem does it solve?
2. Key Patterns: What techniques does it use?
3. Artifact Files: What files does it create/read? (for interoperability)
4. Workflow Stage: Where in a development workflow does it fit?

Step 1.4: Scan Local Candidates

Scan these locations for candidates (frontmatter only, not full files):

plugins/*/skills/*/SKILL.md
plugins/*/agents/*.md
.claude/skills/*/SKILL.md
.claude/agents/*.md

Match by:

• Similar purpose (from description field)
• Similar workflow stage
• Overlapping functionality

Priority Assignment:

• High: Direct functional overlap (same workflow stage, similar purpose)
• Medium: Adjacent functionality (related workflow stage)
• Low: Tangential (shared patterns but different purpose)

Step 1.5: Deduplicate Candidates

If a local file appears as candidate for multiple external sources:

1. Assign to the BEST match (highest similarity)
2. Note secondary matches for cross-reference during enhancement

Phase 2: Contextual Enhancement

Run concurrently for each external source. For each candidate local file:

Step 2.1: Understand Workflow Context

BEFORE comparing, determine:

1. Is this file part of a multi-file workflow?

   • Check for references to other skills/agents in the file
   • Check if other files reference this one
   • Look for phase indicators (discovery → planning → implementation → verification)

2. What stage of the workflow is this file for?

   • Discovery/Research (gathering context)
   • Planning (creating specs, acceptance criteria)
   • Implementation (writing code)
   • Verification (testing, validation)
   • Orchestration (coordinating other agents)

3. What are the upstream/downstream dependencies?

   • What artifacts does this file expect as input?
   • What artifacts does this file produce?

Step 2.2: Compare and Identify Enhancements

Read the full external source and full local file. Identify:

Critical Rule: Only recommend enhancements that fit the file's workflow stage. If an enhancement belongs in a different stage, note which file should receive it instead.

Step 2.3: Add External Artifact Recognition

For interoperability, add a section to context-gathering skills:

## External Framework Artifacts

<external_artifacts>

When gathering context, also check for these artifacts from external frameworks:

**Get Shit Done (GSD)**:
- `STATE.md` - Current project state and progress
- `ROADMAP.md` - Feature roadmap and planning
- `codebase-map.md` - Generated codebase structure
- `research-*.md` - Research documents
- `plan-*.md` - Execution plans

**BMAD-METHOD**:
- `*.agent.yaml` - Agent definitions
- `workflows/*.md` - Workflow definitions
- `party-mode-session.md` - Multi-agent collaboration notes

If found, incorporate their context into discovery.

</external_artifacts>

Step 2.4: Coordinate Cross-File Changes

If enhancements affect multiple files in a workflow:

1. List all affected files
2. Determine the order of changes (upstream before downstream)
3. Ensure consistency (same terminology, compatible artifacts)
4. Make changes atomically (all or none)

Step 2.5: Apply Enhancements

Edit local files with:

• Clear section markers for new content
• Source attribution: SOURCE: Adapted from {external source URL/path}
• Preserved existing functionality (additive changes preferred)

Phase 3: Validation

Step 3.1: Run Linting

uv run prek run --files {all modified files}

Fix any issues before proceeding.

Step 3.2: Finalize Scratch Tracking

Update the scratch tracking document with:

## Results

**Files Modified**:
| File | Enhancements Applied | Source |
|------|---------------------|--------|
| {path} | {what was added} | {external source} |

**Deferred Enhancements** (didn't fit current files):
| Enhancement | Reason Deferred | Suggested Location |
|-------------|-----------------|-------------------|
| {pattern} | {why} | {where it should go} |

**Status**: COMPLETE

Step 3.3: Record Deferred Work in the Configured Backend

If an enhancement is deferred, create or match one logical work item through the configured development-harness backlog backend. Do not create filesystem backlog files or another local state store. The work item description must include the source reference, the exact pattern, why it does not fit the current workflow stage, and the suggested destination. Link the resulting durable record identifier in the scratch tracking document.

Step 3.4: Commit Changes

Stage and commit with message:

feat(skills): integrate patterns from {source names}

Enhancements:
- {file1}: {what was added}
- {file2}: {what was added}

Sources:
- {URL1}
- {URL2}

Success Criteria

• All external sources fetched and analyzed
• Tracking document created with candidate mapping
• Workflow context understood before making changes
• Enhancements fit the file's workflow stage
• Cross-file changes coordinated
• External artifact recognition added for interoperability
• All modified files pass linting
• Deferred enhancements recorded as logical items in the configured backend, with their scratch tracking reference
• Changes committed with source attribution

Example Usage

User: /external-pattern-integrator https://github.com/glittercowboy/get-shit-done/blob/main/agents/gsd-codebase-mapper.md