Knowledge Management

Knowledge Management

This skill manages persistent knowledge in data/knowledge/ in the consumer repo. Knowledge is organized into four category subfolders:

• people (data/knowledge/people/) — stakeholder profiles, communication styles, working preferences
• reference (data/knowledge/reference/) — company info, product overviews, team structure, OKR history
• research (data/knowledge/research/) — domain research, literature reviews, one-shot insights
• market-landscape (data/knowledge/market-landscape/) — living documents of the competitive landscape. Written exclusively by /market-scan as append-only dated sections. Read by this skill but not edited directly.

Scope boundary: Knowledge stores durable, synthesized understanding. Time-stamped observations (competitor launched X yesterday, user said Y, we discovered Z) belong in Signals — use /log-signal for those. Customer personas have their own location (data/personas/); use /define-persona. See the DB Routing Rubric in .claude/context/data-schemas.md.

File Layout

Each entry is one file per topic. Frontmatter:

---
title: Jane Doe — CEO at Acme
category: people             # people | reference | research | market-landscape
tags: [stakeholder, fund]
last_updated: YYYY-MM-DD
status: active               # active | archived
---

No index.md per subfolder — glob + frontmatter is the discovery mechanism.

Modes

1. Fetch (default)

Triggers: /knowledge people "name", /knowledge research "topic", /knowledge reference, or just /knowledge "search term"

Steps:

1. Identify the category from the command (people, reference, research, market-landscape). If no category specified, search across all four.
2. Glob data/knowledge/{category}/*.md (or all categories).
3. Filter by frontmatter: status: active, plus any tag or title match from the search term.
4. For multiple results: show a list with file paths and one-line summaries (read just the frontmatter title + first body line).
5. For a single result: read the full file and show the body.

For People entries, format the output as:

## {Name} — {Role}

**Quick Facts:** {tenure, background, location}

**Communication Style:**
- Prefers: {preferences}
- Dislikes: {anti-patterns}
- Response times: {by channel}

**How to Work With Them:**
- {decision-making style}
- {meeting preferences}
- {update format preferences}

**What They Care About:** {priorities, metrics, pet topics}

**Personal Notes:** {icebreakers, interests, relationship context}

2. Store

Triggers: /knowledge add, /knowledge store

Steps:

1. Ask the user for:
   • Category: people, reference, or research (Market Landscape is reserved for /market-scan — do not create these manually).
   • Title: name or topic.
   • Tags: optional freeform tags.
2. Compute the slug from the title and the file path: data/knowledge/{category}/{slug}.md. If the file exists, ask whether to overwrite or pick a new slug.
3. Based on the category, prompt for structured content:
   • people — use the stakeholder profile template (Quick Facts, Communication Style, How to Work With Them, What They Care About, Personal Notes)
   • reference — freeform content covering the topic
   • research — structured with Key Findings, Implications, Sources, Date of Research
4. Write the file with frontmatter (title, category, tags, last_updated: today, status: active) followed by the body.
5. Confirm creation with the file path and a one-line summary.

Redirects: If the user tries to store content that is clearly a time-stamped observation (dated event, recent finding, competitor move), redirect them: "That sounds like a Signal, not a Knowledge entry. Want to run /log-signal instead?" If it's a customer persona, redirect to /define-persona.

3. Review

Triggers: /knowledge review

Steps:

1. Glob data/knowledge/**/*.md. Read frontmatter only (not bodies) for each file.
2. Analyze for:
   • Staleness (category-dependent):
     • people / reference / research: last_updated older than 90 days
     • market-landscape: last_updated older than 30 days (market moves faster)
   • Gaps: Categories with few or no entries (e.g., "0 People entries on file")
   • Volume: Flag if any category exceeds 30 entries (suggest pruning)
   • Archived noise: Files with status: archived older than 180 days that could be deleted instead of kept.
3. Present a review summary:

## Knowledge Review — {date}

**Total entries:** {count} (people: {n}, reference: {n}, research: {n}, market-landscape: {n})

### Stale Entries
- {file path} — Category: {cat}, Last updated: {date}, Threshold: {30d or 90d}

### Gaps
- {category} has no entries

### Suggestions
- Update: {list of stale entries worth refreshing}
- Archive: {list of entries that may no longer be relevant}
- Add: {suggested new entries based on gaps}
- Re-scan: {market-landscape entries past the 30-day threshold — suggest /market-scan}

4. Ask the user if they want to update, archive, or add any entries.

People Profile Template

When storing a new People entry, use this template as the body structure (after the YAML frontmatter):

# {Name} — {Role}

## Quick Facts
- **Role:** {title}
- **Reports to:** {manager}
- **Tenure:** {time at company}
- **Background:** {previous roles, education}
- **Location:** {office/remote, timezone}

## Communication Style

**Prefers:**
- {preference 1}
- {preference 2}

**Dislikes:**
- {anti-pattern 1}
- {anti-pattern 2}

**Response times:**
- {channel}: {typical response time}

## How to Work With Them

### In Meetings
- {meeting preferences}

### Getting Decisions
- {decision-making style and process}

### Giving Updates
- {preferred format and cadence}

## What They Care About

**Top priorities:**
1. {priority}

**Key metrics they track:**
- {metric}

**Pet topics:**
- {topic they're passionate about}

## Personal Notes
- {icebreakers, interests, family, relationship context}

Market Landscape Entry Structure

Market Landscape entries are maintained by /market-scan — this skill reads them but does not edit them. Expect this machine-readable structure so heading-based parsing works:

---
title: {Product} — Market Landscape — {Market Category}
category: market-landscape
tags: [...]
last_updated: YYYY-MM-DD
status: active
---

# {Product} — Market Landscape — {Market Category}

_Living document. New scans append dated sections below._

## Scan — {YYYY-MM-DD}

### Competitor Radar
### Product & Feature Moves
### Funding & Business Moves
### Customer Sentiment
### Strategic Implications
### What's New vs. Prior Scans
### Sources

## Scan — {YYYY-MM-DD}
...

When surfacing a Market Landscape entry in Fetch mode, show only the most recent ## Scan — sub-section unless the user asks for history.

Suggested Follow-ups

After fetching People knowledge, suggest: "Want me to prep for a meeting with {name}?"

After storing Research knowledge, suggest: "Want me to check how this connects to your current backlog? Run /fetch-context."

After fetching a Market Landscape entry older than 30 days, suggest: "This scan is {N} days old. Want to run /market-scan to refresh it?"

After a review, suggest: "Want me to help fill the gaps? I can draft entries for missing categories."