Context Research Skill

Purpose

Eliminate the "let me look that up" cycle. Research a topic once, persist the findings, and let Claude reference them automatically in future sessions.

Configuration

Check for .context/config.md in the project root. If it exists, read the YAML frontmatter for source priority and preferences.

For the default config template, see references/source-config-template.md.

Workflow

Step 1: Parse Input

Extract from $ARGUMENTS:

• Topic: The subject to research (e.g., "Stripe subscriptions with Next.js")
• Refresh flag: Whether --refresh was passed

Step 2: Load Config

Read .context/config.md if it exists, parse YAML frontmatter for source priority. Fall back to defaults if missing.

Step 3: Check Existing Context

# Slugify the topic: lowercase, spaces→hyphens, strip special chars
SLUG=$(echo "$TOPIC" | tr '[:upper:]' '[:lower:]' | sed 's/[^a-z0-9]/-/g' | sed 's/--*/-/g' | sed 's/^-//;s/-$//')

Check if .context/${SLUG}.md already exists:

• If exists AND no --refresh → show the existing brief, note its date, ask if user wants to refresh
• If exists AND --refresh → proceed with re-research
• If not exists → proceed with research

Step 4: Research (Using Configured Source Priority)

Iterate through the sources list from config. For each source:

context7 (if in list and available):

• resolve-library-id for the library/framework mentioned in topic
• query-docs with relevant queries
• Extract: API patterns, code examples, configuration

firecrawl (if in list and available):

• Scrape official documentation pages
• Use preferred-sites from config to prioritize specific doc sites
• Extract: comprehensive guides, migration notes, gotchas

websearch (if in list and available):

• Search for "${TOPIC}" best practices
• Search for "${TOPIC}" gotchas common mistakes
• If preferred-sites configured: site:${site} ${TOPIC}

webfetch (if in list and available):

• Fetch specific documentation URLs found via earlier sources
• Good for grabbing API references, changelogs

Source availability handling (based on skip-unavailable config):

• skip-unavailable: true (default): Silently skip sources that error or aren't connected. Still produce the brief using whatever sources succeeded. Note in the brief header which sources were actually used: Sources: {used_list} (of {configured_count} configured).
• skip-unavailable: false: Warn the user when a configured source isn't available, but still continue with remaining sources. Don't halt the entire research.

Gather across all sources:

• Core concepts: What is this? Key mental models
• API patterns: Common usage patterns with code examples
• Gotchas: Known pitfalls, breaking changes, version-specific issues
• Code examples: Real-world snippets (not toy examples)
• Official docs links: URLs for deeper reading

Step 5: Generate Context Brief

Create .context/${SLUG}.md (respect max-lines from config):

# ${TOPIC}

> Researched: ${DATE} | Sources: ${SOURCE_LIST} | Lines: ${LINE_COUNT}/${MAX_LINES}

## Core Concepts

[Key mental models and architectural decisions]

## API Patterns

[Common usage patterns with code examples]

```[language]
// Example code

Gotchas & Pitfalls

• [Known issues, version-specific problems, common mistakes]

Code Examples

// Production-ready patterns

References

• [Official doc URL 1]
• [Official doc URL 2]

### Step 6: Confirm

✅ Context saved: .context/${SLUG}.md Sources: ${SOURCES_USED} (of ${SOURCES_CONFIGURED}) Topics: [list key sections]

This context will be automatically available when working on related code.

## Directory Structure

.context/ ├── config.md ← Source priority + preferences ├── stripe-subscriptions-with-nextjs.md ├── supabase-rls-policies.md └── react-server-components.md

## PRD Integration

When called from the **Discuss phase** (auto-research mode):
- The discuss workflow extracts technology keywords from the feature brief
- Calls this skill automatically for each identified technology
- Context briefs are then available to the discoverer agent

When called standalone:
- User runs `/cks:context "topic"` directly
- Full interactive mode with confirmations

## Customization

This skill ships with opinionated defaults. Review and adapt to your needs:

- **Source priority**: Which research tools to try first — edit `.context/config.md` or see `references/source-config-template.md`
- **Max lines**: Brief length cap (default 200) — edit `.context/config.md`
- **Preferred sites**: Prioritized documentation sites — edit `.context/config.md`
- **Auto-research**: Whether to run during `/cks:discover` (default true) — edit `.context/config.md`
- **allowed-tools**: Currently `Read, Grep, Glob, Write, WebSearch, WebFetch`. Add tools if your workflow needs more.
- **model**: Currently `sonnet`. Remove to use your default model.

## Rules

1. **Always create `.context/` directory** if it doesn't exist
2. **Include the research date** — context goes stale
3. **Prefer official docs** over blog posts or tutorials
4. **Include real code examples** — not just descriptions
5. **Keep it concise** — respect `max-lines` config (default 200)
6. **Slugify consistently** — lowercase, hyphens, no special chars
7. **Respect source priority** — use config order, skip unavailable
8. **Don't duplicate** — check existing briefs before researching