Back to skills
SKILL.md
Memory Documentary Rjmurillo Ai Agents
ASecurityGenerate evidence-based documentary reports by searching across all 4
- 666 stars
- 0 votes
- 0 copies
- 1 view
- Added September 2, 2026
Works with
Security analysis
100/100npx -y skills add majiayu000/claude-skill-registry --skill memory-documentary-rjmurillo-ai-agents --agent claude-codeAre you the author of Memory Documentary Rjmurillo Ai Agents?
Add the live security badge to your README. It updates with every re-scan.
[](https://www.skillsdirectory.com/skills/majiayu000-memory-documentary-rjmurillo-ai-agents)---
name: memory-documentary
version: 1.0.0
description: Generate evidence-based documentary reports by searching across all 4
memory systems (Claude-Mem, Forgetful, Serena, DeepWiki), .agents/ artifacts, and
GitHub issues. Produces investigative journalism-style analysis with full citation
chains.
license: MIT
model: claude-sonnet-4-5
metadata:
timelessness: 8/10
category: analysis
---
# Memory Documentary Skill
Generate comprehensive documentary-style reports from your memory systems with full evidence chains.
---
## Quick Start
```text
/memory-documentary [topic]
```
Example topics:
- "recurring frustrations"
- "coding patterns not codified"
- "evolution of thinking on testing"
- "decisions I second-guessed"
- "what I complain about most"
---
## Triggers
Use this skill when:
- "Search my memories for patterns about..."
- "What does my history say about..."
- "Generate a documentary on my [topic]"
- "Cross-reference all systems for..."
- "Evidence-based analysis of my [topic]"
---
## Quick Reference
| Phase | Action | Output |
|-------|--------|--------|
| 1 | Topic Comprehension | Search variants, scope boundaries |
| 2 | Investigation Planning | Explicit queries per system |
| 3 | Data Collection | Evidence with IDs, timestamps |
| 4 | Report Generation | Documentary with citations |
| 5 | Memory Updates | Store meta-pattern discovered |
---
## How It Works
The skill searches ALL available data sources systematically:
**Memory Systems (4 MCP servers)**:
- Claude-Mem: Timeline observations via 3-layer workflow
- Forgetful: Semantic memory with linked entities
- Serena: Project-specific lexical memory
- DeepWiki: Documentation resources
**Project Artifacts**:
- `.agents/retrospective/` - Learning extractions
- `.agents/sessions/` - Session logs
- `.agents/analysis/` - Research reports
- `.agents/architecture/` - ADRs
- `.agents/planning/` - Plans and PRDs
**GitHub Issues**:
- Open and closed issues
- Comments and related PRs
---
## Commands
```bash
# Standard invocation
/memory-documentary "recurring frustrations"
# With explicit time range
/memory-documentary "testing patterns" --since 2025-12-01
```
---
## Report Structure
### Executive Summary
- Key finding in one sentence
- Timeline span (earliest to most recent)
- Evidence count (memories + observations + issues + files)
- Major pattern categories
### Evidence Trail
Full citation for each finding:
- Memory ID/Observation ID with retrieval command
- Source system (Forgetful/Claude-Mem/Serena)
- Timestamp/Creation date
- Direct quote from source
- Links to related evidence
### Pattern Evolution
Timeline showing how thinking changed:
```text
YYYY-MM-DD: [Observation #ID] - Initial state
YYYY-MM-DD: [Memory #ID] - First iteration
YYYY-MM-DD: [Issue #NNN] - Technical response
```
### Unexpected Patterns
Cross-system synthesis revealing:
- Frequency patterns (time clustering)
- Correlation patterns (X happens before Y)
- Avoidance patterns (conspicuous absence)
- Contradiction patterns (saying vs doing)
- Evolution patterns (recursive loops)
- Emotional patterns (frustration markers)
### Synthesis
- What all systems agree on
- What's recent (Claude-Mem) vs crystallized (Forgetful)
- Actionable recommendations
---
## Evidence Standards
| Standard | Requirement |
|----------|-------------|
| Citation | Every claim has ID, timestamp, quote |
| Quotes | Direct quotes, not paraphrases |
| Verification | Retrieval commands for all evidence |
| Cross-links | Related evidence connected |
---
## Anti-Patterns
| Avoid | Why | Instead |
|-------|-----|---------|
| Partial searches | Miss critical evidence | Search ALL systems |
| Paraphrasing | Loses verifiability | Direct quotes only |
| Single query | Miss variations | 3+ query variants per system |
| Skipping systems | Incomplete picture | Check all 4 MCP servers |
---
## Output Location
Reports saved to: `.agents/analysis/[topic]-documentary-[date].md`
---
## Related Skills
| Skill | Relationship |
|-------|--------------|
| memory | Operations (search, update) |
| exploring-knowledge-graph | Forgetful traversal |
| retrospective | Learning extraction |
| skillbook | Pattern → skill conversion |
---
## On Invocation
**Read**: `references/execution-protocol.md`
This contains the full 5-phase execution protocol with detailed instructions for each data source.
Attribution
Comments
Loading comments…
Newsletter
Get the weekly list of safe new skills, and the ones we pulled.