From ephemeris
Answer a question from the local wiki using the current session's model. Read-only.
How this command is triggered — by the user, by Claude, or both
Slash command
/ephemeris:query <question>This command is limited to the following tools:
The summary Claude sees in its command listing — used to decide when to auto-load this command
# /ephemeris:query Answer a natural-language question about the user's project from the local ephemeris wiki. All reasoning happens in this session — no subprocess, no API key, no outbound calls. This skill is read-only: it never writes, edits, or moves a file. ## Instructions ### 1. Parse the question - If `$ARGUMENTS` is empty, emit `Usage: /ephemeris:query "<question>"` and stop. - Otherwise, treat `$ARGUMENTS` as the question. Extract 3–6 key terms: nouns, named entities, distinctive verbs. Ignore stop-words and generic filler. ### 2. Resolve the wiki root - `<wiki_root>` is `...
Answer a natural-language question about the user's project from the local ephemeris wiki. All reasoning happens in this session — no subprocess, no API key, no outbound calls. This skill is read-only: it never writes, edits, or moves a file.
$ARGUMENTS is empty, emit Usage: /ephemeris:query "<question>" and
stop.$ARGUMENTS as the question. Extract 3–6 key terms: nouns,
named entities, distinctive verbs. Ignore stop-words and generic filler.<wiki_root> is $EPHEMERIS_WIKI_ROOT if set and non-empty, else
~/.claude/ephemeris/wiki.<wiki_root> does not exist, emit Wiki not found at <path>. Run /ephemeris:ingest first. and stop.Glob: <wiki_root>/**/*.md.
Wiki is empty — no pages have been built yet. and
stop.For each key term from step 1, run Grep with:
pattern: the key term (case-insensitive)path: <wiki_root>output_mode: files_with_matchesglob: *.mdUnion the file paths returned across all terms. Rank by number of terms matched (most terms = most relevant). Take the top 5.
Cannot answer this from the wiki — no relevant pages found. followed by an empty ## Sources block and stop.For each of the top 5 files: Read the full file contents. Hold the content in
working memory.
Compose an answer to the question using only the content of the pages you
just Read. Do not draw on training-data knowledge about the project or
codebase — only what the wiki explicitly says.
Grounding rule: Every claim in the answer body must be traceable to a
specific wiki page you read. If the pages don't support a claim, drop the claim
or mark it (not found in wiki). If none of the pages actually answer the
question, emit Cannot answer this from the wiki — no relevant pages found.
and skip to the ## Sources block.
End the response with a ## Sources block listing every wiki page referenced
in the answer, as relative paths under <wiki_root>. Example:
## Sources
- topics/authentication.md
- DECISIONS.md
- entities/AuthMiddleware.md
The ## Sources block appears even when the answer is the cannot-answer
sentinel (possibly empty). It is the contract that tells the user what the
skill looked at.
npx claudepluginhub trevoredris/ephemeris --plugin ephemeris/queryQueries compiled wiki for answers using quick (indexes), standard (articles), or deep (full+raw+sibling) depths, citing sources only from wiki content.
/queryExecutes SQL queries against Claude Code/Codex CLI history and current Git repo data (commits, branches, diffs). Displays results in formatted tables.
/querySearches Weaviate collections using natural language queries. Specify query text, collections, and limit. Returns matching objects for further processing.
/queryQueries the research vault by searching papers and findings for question terms, synthesizes an answer with wikilink citations.
/queryQueries the LLM Wiki to answer a question from accumulated knowledge, providing citations to source pages and using search/graph tools as needed.
/queryOptimizes and analyzes queries across SQL databases, MongoDB, Redis, and ORMs: interprets EXPLAIN plans, recommends indexes, rewrites for performance, detects N+1 issues, profiles slow queries. Supports flags like --explain, --indexes, --n-plus-one.