Research document lifecycle management — creating, reusing, updating, superseding, and archiving research across the platform. Ensures research is saved as research (NOT specs), discoverable by all agents, and kept current. Use when user says "research this", "look into", "investigate", "find out about", "what do we know about", "save this research", "update research", "check prior research", "research document", "research folder", or any agent produces research findings that should be persisted.
Standardmäßig ist der Prompt ausgewählt, der zuerst die Quelle prüft. Sie können zu einem direkten Befehl wechseln oder eine lokale Kopie herunterladen.
Quelldateien prüfen
Lesen Sie SKILL.md und alle von SkillsMP angezeigten Begleitdateien, bevor Sie sich für eine Installation entscheiden.
Mit Codex oder Claude installieren Kopieren Sie diesen Prompt, fügen Sie ihn in Codex, Claude oder einen anderen Assistant ein und lassen Sie die Skill-Seite prüfen und installieren.
Ein direkter Befehl überspringt den Prüf-Prompt. Prüfen Sie die Quelle, bevor Sie ihn ausführen.
Research document lifecycle management — creating, reusing, updating, superseding, and archiving research across the platform. Ensures research is saved as research (NOT specs), discoverable by all agents, and kept current. Use when user says "research this", "look into", "investigate", "find out about", "what do we know about", "save this research", "update research", "check prior research", "research document", "research folder", or any agent produces research findings that should be persisted.
Research Management Skill
Standard procedure for how ANY agent on the platform creates, stores, discovers, updates, and retires research documents. This skill exists because research is not a spec — they serve fundamentally different purposes and belong in different places.
Why This Exists
Research kept getting saved as specs (data/specs/), which is wrong:
Research
Spec
Purpose
Capture context for decision-making
Capture requirements for implementation
Audience
Any agent or human needing context
The agent/person building the thing
Lifecycle
Updated with new findings, superseded when obsolete
Versioned, frozen once approved
Location
data/research/
data/specs/
Mutability
Living document — update in place
Immutable once approved (create v2 instead)
Rule of thumb: If it answers "what did we learn?" → research. If it answers "what should we build?" → spec.
Critical Rules — Read These First
Research goes in data/research/, never in data/specs/. Specs are for implementation requirements. Research is for captured context, analysis, and findings.
Check before creating. Before starting ANY new research, search data/research/ for existing documents on the topic. Update existing docs — don't create duplicates.
One canonical document per topic. If research on "LinkedIn DM capabilities" already exists, update it. Don't create a second file.
Always include sources. Every claim needs attribution. No "studies show..." without citing which studies.
Use the standard format. Every research document follows the template below. No exceptions.
Update, don't duplicate. When new findings arrive for an existing topic, update the existing document's content and bump the last-updated timestamp.
Supersede, don't delete. When a research document is completely replaced by better research, mark the old one status: superseded with a superseded-by pointer. Never delete research — it's historical context.
Tags are mandatory. Every research doc must have at least 2 tags for discoverability.
If a topic doesn't fit: Use the closest category. When in doubt, use business/ for external-facing research and platform/ for internal platform research.
Naming Convention
{descriptive-topic-slug}.md
Lowercase, hyphen-separated
Descriptive enough to identify the topic from the filename alone
No dates in filenames (dates go in frontmatter)
No version numbers (research docs are updated in place, not versioned)
---
topic: "[Clear, descriptive topic title]"
created: YYYY-MM-DD
last-updated: YYYY-MM-DD
status: active | superseded | archived
author: "[agent-name or user]"
sources:
- "[URL or description of source 1]"
- "[URL or description of source 2]"
tags:
- "[tag1]"
- "[tag2]"
related-research:
- "[path/to/related-doc.md]"
superseded-by: "[path/to/newer-doc.md]" # only if status: superseded
---# [Topic Title]## Summary
[2-5 paragraph executive summary of key findings. This is the "compacted context" —
someone reading ONLY this section should understand the core conclusions.]
## Key Findings### [Finding 1 Title]
[Detailed explanation with evidence and citations]
### [Finding 2 Title]
[Detailed explanation with evidence and citations]
### [Finding N Title]
[Detailed explanation with evidence and citations]
## Raw Data & Evidence
[Supporting data, screenshots references, detailed analysis, benchmark numbers, etc.
This section can be longer — it's the backup for the summary above.]
## Sources & Citations1. [Source name](URL) — [what it contributed to this research]
2. [Source name](URL) — [what it contributed to this research]
## Open Questions- [Unanswered questions that need further research]
- [Gaps in the data]
## Related Documents- [Link to related research doc] — [how it connects]
- [Link to related spec, if applicable] — [relationship]
Required Sections
Section
Required?
Notes
Frontmatter
✅ Yes
All fields except superseded-by and related-research
Summary
✅ Yes
The compacted context — most agents will only read this
Key Findings
✅ Yes
At least one finding
Raw Data & Evidence
Optional
Include when there's substantial supporting data
Sources & Citations
✅ Yes
At least one source
Open Questions
Optional
Include when gaps exist
Related Documents
Optional
Include when cross-references exist
Research Lifecycle
Phase 1: Discovery — Check Before You Research
Before starting ANY research task:
1. glob("data/research/**/*.md") — see what exists
2. grep for topic keywords in data/research/ — find related docs
3. If a matching document exists with status: active → UPDATE it, don't create new
4. If a matching document exists with status: superseded → check what superseded it
5. If nothing exists → proceed to Phase 2
This is not optional. The #1 anti-pattern is creating duplicate research because an agent didn't check first.
Phase 2: Create — Save New Research
When creating a new research document:
Choose the right category folder from the directory structure above
Create the category folder if it doesn't exist yet (New-Item -ItemType Directory)
Use the standard template — copy the template above, fill in all required sections
Set status to active
Include at least one source — if the research came from web searches, cite the URLs. If from agent analysis, cite the agent and methodology.
Add tags — at least 2, chosen for discoverability (other agents will search by these)
Cross-reference — if related research exists, add it to related-research in frontmatter AND update the related doc's related-research to point back
Phase 3: Update — Augment Existing Research
When new findings relate to an existing research document:
Update the Summary if the new findings change the conclusions
If findings CONTRADICT existing content — update the contradicted section with the correction, noting what changed and why. Don't silently overwrite.
Phase 4: Supersede — Replace Outdated Research
When research is completely outdated or replaced:
Create the new research document following Phase 2
Update the old document's frontmatter:
Set status: superseded
Set superseded-by: path/to/new-doc.md
Add a note at the top of the old document:
> ⚠️ **SUPERSEDED** — This research has been replaced by [new-doc.md](path). See that document for current findings.
Do NOT delete the old document — it's historical context that may still be referenced
Phase 5: Archive — Retire Irrelevant Research
When research is no longer relevant (topic is moot, project cancelled, etc.):
Update frontmatter: Set status: archived
Add archive note at the top:
> 📦 **ARCHIVED** — This research is no longer actively maintained. Archived on YYYY-MM-DD. Reason: [why].
Don't delete — leave in place for historical reference
Agent Integration
For ANY Agent Doing Research
When your task involves research (web searches, analysis, investigation):
BEFORE researching: Check data/research/ for existing documents on the topic (Phase 1)
AFTER researching: Save findings using the template (Phase 2) or update existing (Phase 3)
In your response: Reference the research document path so other agents can find it
In your memory: Note the research doc path in your working memory for future reference
For Agents Needing Context
When you need background on a topic before making a decision:
Search data/research/ using glob/grep with topic keywords and tags
Read the Summary section — this gives you compacted context without reading the full doc
Check status — only trust active documents. superseded means there's a newer version.
Check last-updated — if it's months old, the research may need refreshing before relying on it
Research-to-Spec Handoff
When research informs a spec:
Research lives in data/research/ — it's the "why" and "what we learned"
Spec lives in data/specs/ — it's the "what to build" derived from the research
The spec's frontmatter should reference the research doc(s) that informed it
The research doc's related-research should link to the spec
data/research/technical/linkedin-dm-capability.md → informs → data/specs/linkedin-dm-v1.md
(what's possible) (what to build)
Anti-Patterns
Anti-Pattern
Why It's Wrong
Do This Instead
❌ Saving research in data/specs/
Specs are for implementation requirements, not context
Save in data/research/{category}/
❌ Researching without checking existing docs
Creates duplicates, wastes effort
Always run Phase 1 discovery first
❌ Creating a new doc when one exists on the topic
Fragments knowledge, causes conflicting sources
Update the existing document (Phase 3)
❌ Research without sources/citations
Unverifiable claims, might be hallucinated
Always cite at least one source
❌ Letting research go stale
Agents rely on outdated context for decisions
Update or archive when information changes
❌ Deleting superseded research
Loses historical context
Mark as superseded with pointer to replacement
❌ Saving research only in agent working memory
Lost when memory is trimmed, not discoverable by other agents
Persist to data/research/
❌ Embedding research findings in chat responses only
Context lost after session ends
Always persist to file
Examples
Example 1: Agent receives "research LinkedIn DM capabilities"
1. glob("data/research/**/*linkedin*") → found: data/research/technical/linkedin-dm-capability.md
2. Read it → status: active, last-updated: 2026-05-04
3. It's recent and active → UPDATE this doc with new findings rather than creating new
4. Bump last-updated, add new sources, update summary if conclusions changed
Example 2: Agent finishes researching a brand-new topic
1. glob("data/research/**/*pricing*") → no results
2. grep("pricing strategy", "data/research/") → no results
3. No existing research → CREATE new doc at data/research/business/saas-pricing-models.md
4. Use the standard template, fill all required sections
5. Commit the new file
Example 3: Agent finds existing research is completely wrong
1. Read data/research/financial/aws-cost-analysis.md → findings are from 3 months ago, AWS changed pricing
2. Create new doc: data/research/financial/aws-cost-analysis.md (same path — update in place if fixable)
3. OR if it's a completely different analysis: create new doc AND mark old as superseded
4. Update old doc: status: superseded, superseded-by: path/to/new-doc.md
