Potential Discovery Workflow

Purpose

Perform deep analysis of incoming content to:

1. Extract and understand its core purpose
2. Create extensive, structured summaries
3. Identify potential value and applications for the current project
4. Suggest concrete integration opportunities or inspired improvements

When to Use

• Evaluating a new library, tool, or framework
• Reviewing incoming code contributions or PRs
• Analyzing competitor products or similar solutions
• Assessing documentation, specs, or RFCs
• Exploring repositories for reusable patterns
• Reviewing articles, blog posts, or research papers

Input Types

analyze: ./incoming/proposal.md

analyze: ./incoming/new-library/

analyze: https://github.com/user/repo

Procedure

Phase 1: Content Ingestion

1. Identify content type (text/file/folder/URL)
2. Load content:
   • Text: Use directly
   • File: Read file contents
   • Folder: Scan structure, read key files (README, package.json, etc.)
   • URL: Fetch content (if MCP fetch available) or note for manual review
3. Assess scope: Estimate content size and complexity

Phase 2: Purpose Extraction

Analyze content to identify:

Output format:

### Purpose Analysis

**Core Purpose:** {one-sentence summary}

**Problem Solved:** {what pain point it addresses}

**Target Audience:** {who would use this}

**Key Features:**
- {feature 1}
- {feature 2}
- {feature 3}

**Architecture/Approach:** {how it works at a high level}

**Maturity:** {early/stable/mature} — {evidence}

Phase 3: Extensive Summary

Create a comprehensive summary covering:

### Detailed Summary

#### Overview
{2-3 paragraphs explaining what this is and how it works}

#### Structure
{File/folder organization, key components}

#### Technical Details
- **Language/Runtime:** {languages, versions}
- **Dependencies:** {key dependencies}
- **Build System:** {how to build/run}
- **Testing:** {test approach}

#### Strengths
- {strength 1}
- {strength 2}

#### Limitations
- {limitation 1}
- {limitation 2}

#### Notable Patterns
- {interesting pattern or technique}
- {reusable approach}

Phase 4: Potential Assessment

Evaluate relevance to current project:

1. Read project context:

   • Check

     .agent/constitution.md

     for project scope
   • Check

     .agent/map.md

     for codebase structure
   • Check current

     focus.md

     for active work

2. Identify alignment:

   • Technology overlap (same language, framework, etc.)
   • Problem overlap (solves similar issues)
   • Pattern overlap (uses approaches we could adopt)

3. Generate potential opportunities:

### Potential for Current Project

**Relevance Score:** {low/medium/high}

**Alignment:**
- ✅ {alignment point 1}
- ✅ {alignment point 2}
- ⚠️ {partial alignment}
- ❌ {misalignment}

**Opportunities:**

1. **{Opportunity Title}**
   - What: {description}
   - Effort: {low/medium/high}
   - Value: {low/medium/high}
   - How: {brief approach}

2. **{Opportunity Title}**
   - What: {description}
   - Effort: {low/medium/high}
   - Value: {low/medium/high}
   - How: {brief approach}

**Risks/Concerns:**
- {risk 1}
- {risk 2}

**Recommendation:** {adopt/adapt/inspire/skip}

Phase 5: Issue Creation

Convert opportunities into trackable IDEA issues using

agent-ops-idea

For each identified opportunity:

1. Create IDEA issue (minimal mode via agent-ops-idea):

   ## IDEA-{N}@{hash} — {Opportunity Title}
   
   **Status:** `idea`
   **Type:** IDEA
   **Created:** {date}
   **Source:** Potential discovery from {content identifier}
   
   ### Original Opportunity
   {opportunity description from Phase 4}
   
   ### Context
   - **Discovered in:** {source content}
   - **Relevance:** {low/medium/high}
   - **Effort:** {low/medium/high}
   - **Value:** {low/medium/high}
   
   ### Notes
   {any additional context or caveats}
   

2. Present summary to user:

   ## Potential Discovery Complete
   
   **Analyzed:** {content identifier}
   **Relevance:** {overall score}
   
   **Issues Created:**
   - IDEA-{N}@{hash}: {title} (effort: X, value: Y)
   - IDEA-{N}@{hash}: {title} (effort: X, value: Y)
   
   These are now in backlog.md for later triage.
   Use `/agent-task triage` to review and prioritize.
   

No opportunities found:

## Potential Discovery Complete

**Analyzed:** {content identifier}
**Relevance:** Low

No actionable opportunities identified for the current project.
The content may still be useful as reference material.

Output Locations

.agent/issues/backlog.md

IMPORTANT: This skill only writes to

backlog.md

.counter

• Create report files in

  .agent/discoveries/

• Modify

  focus.md

• Make any code changes

All findings become IDEA issues for later review and triage.

Integration with agent-ops-idea

This skill uses

agent-ops-idea

1. Analysis complete → Opportunities identified
2. For each opportunity → Invoke

   agent-ops-idea

   (minimal mode)
3. Create IDEA issue → Added to

   backlog.md

   with:
   • Original opportunity description
   • Source reference (what content it came from)
   • Relevance assessment
4. User reviews later → Triage to promote or drop

Integration with Other Skills

agent-ops-idea

agent-ops-project-sections

agent-ops-context-map

agent-ops-tasks

Customization Options

User can specify intent to focus analysis:

`agent-discover` ./incoming/auth-library/ --intent "improve our authentication"
`agent-discover` https://github.com/user/tool --intent "CLI patterns"
`agent-discover` spec.md --intent "API design inspiration"

When intent is provided:

• Purpose analysis weighted toward intent
• Potential assessment focused on stated goal
• Opportunities filtered for relevance

Completion Criteria

• Content successfully ingested
• Purpose clearly identified and documented
• Comprehensive summary created
• Potential assessed against project context
• Opportunities converted to IDEA issues in backlog.md
• User presented with summary of created issues

Anti-patterns (avoid)

• ❌ Shallow analysis (just reading README)
• ❌ Missing project context (analyzing in isolation)
• ❌ Vague opportunities ("could be useful")
• ❌ Ignoring limitations or risks
• ❌ Over-promising value without evidence
• ❌ Creating files outside of backlog.md and .counter
• ❌ Modifying focus.md or other state files
• ❌ Making code changes

Examples

Example 1: Library Analysis

Input:

analyze: ./incoming/result-monad/

Output:

# Potential Discovery Report

**Content:** ./incoming/result-monad/
**Analyzed:** 2026-01-17
**Context:** agent-ops-cli (Python CLI tool)

### Purpose Analysis

**Core Purpose:** Rust-style Result type for Python error handling

**Problem Solved:** Replaces exceptions with explicit error returns

**Key Features:**
- Result[T, E] generic type
- Pattern matching support
- Railway-oriented programming helpers

### Potential for Current Project

**Relevance Score:** Medium

**Alignment:**
- ✅ Same language (Python)
- ✅ We have error handling inconsistencies
- ⚠️ Would require refactoring existing code

**Opportunities:**

1. **Adopt for new modules**
   - What: Use Result type in kg module
   - Effort: Low
   - Value: Medium
   - How: Add as optional dependency, use in new code

**Recommendation:** Adapt — use pattern for new code, don't refactor existing

Example 2: URL Analysis

Input:

analyze: https://github.com/anthropics/anthropic-cookbook --intent "prompt engineering patterns"

Output:

# Potential Discovery Report

**Content:** anthropic-cookbook repository
**Analyzed:** 2026-01-17
**Context:** AgentOps prompt/skill design

### Purpose Analysis

**Core Purpose:** Collection of Anthropic API usage patterns and examples

### Potential for Current Project

**Relevance Score:** High

**Opportunities:**

1. **Prompt structure patterns**
   - What: XML-tagged prompt sections
   - Value: High
   - How: Review and adopt in skill prompts

2. **Tool use examples**
   - What: Function calling patterns
   - Value: High  
   - How: Reference for harness tool design