cache-docs
npx machina-cli add skill eidetics/claude-eidetic/cache-docs --openclaw/cache-docs
Cache external docs so future queries use search_documents (~20 tokens/result) instead of re-fetching (~5K+ tokens).
Usage:
/eidetic:cache-docs <library>/eidetic:cache-docs <library> <topic>
Step 1: Parse Arguments
Extract library (required) and topic (optional). If no argument, ask: "Which library's docs would you like to cache?"
Step 2: Check Existing Cache
search_documents(query="overview", library="<LIBRARY>")
- Results found and fresh: inform user, offer to refresh.
- Results found but stale: proceed to refresh.
- No results: proceed to Step 3.
Step 3: Resolve Library
resolve-library-id(libraryName="<LIBRARY>")
Pick the best matching ID.
Step 4: Fetch Docs
query-docs(libraryId="<LIBRARY_ID>", topic="<TOPIC or 'getting started'>")
Step 5: Cache
index_document(
content="<FETCHED_CONTENT>",
source="context7:<LIBRARY_ID>/<TOPIC>",
library="<LIBRARY>",
topic="<TOPIC>",
ttlDays=7
)
Step 6: Verify
search_documents(query="<TOPIC or library>", library="<LIBRARY>", limit=3)
Report: chunks cached, TTL, search command.
Note: repeat Steps 4-5 with different topics to cache multiple topics for the same library.
Source
git clone https://github.com/eidetics/claude-eidetic/blob/main/plugin/plugins/claude-eidetic/skills/cache-docs/SKILL.mdView on GitHub Overview
cache-docs caches fetched external documentation locally so future queries can use search_documents instead of re-fetching large content. This reduces token usage by serving small, indexed chunks (~20 tokens per result) instead of pulling thousands of tokens at query time.
How This Skill Works
It parses the library and optional topic, checks the cache with a search_documents call, resolves the library to a library ID, fetches docs with query-docs, and stores them via index_document with a 7-day TTL. It finishes by verifying the cached content with a final search_documents query to confirm the topic or library is cached.
When to Use It
- You frequently answer questions about a single library and want faster responses than live fetches.
- Network access is slow or expensive, making repeated fetches unattractive.
- You want to reduce token costs by serving cached content instead of re-fetching.
- You need consistent results for common topics without hitting external sources each time.
- You plan to cache multiple topics for the same library over time.
Quick Start
- Step 1: Issue /eidetic:cache-docs <library> <topic> (topic is optional).
- Step 2: The skill resolves the library, fetches docs via query-docs, and caches them with a 7-day TTL.
- Step 3: Verify cached content by running search_documents(query="<TOPIC>" or library, library="<LIBRARY>", limit=3).
Best Practices
- Always specify a library and, if possible, a focused topic to maximize cache utility.
- Prioritize core topics like 'getting started' or 'overview' to bootstrap useful caches.
- Refresh stale caches proactively and verify freshness with a search_documents check.
- Align TTL with your update cadence; TTL defaults to 7 days but can be adjusted.
- Periodically spot-check cached content against source docs to catch drift or changes.
Example Use Cases
- Cache Python standard library docs to speed up a Python helper assistant.
- Cache React or Vue docs for a frontend development assistant.
- Cache NumPy docs for data science help and numerical computing tasks.
- Cache PostgreSQL docs for a database administration assistant.
- Cache TensorFlow or PyTorch docs for ML model guidance.
