/query

description: Ask a question of the research vault. Searches papers and findings, synthesizes an answer with wikilink citations.
argument-hint: "<question>"

/paperloom:query

`$ARGUMENTS` is a natural-language question about the vault's contents.

Vault path

Read `vault_path` from `${CLAUDE_PLUGIN_CONFIG:vault_path}` (default `~/PaperLoom`), expand `~` to an absolute path, and use that for every Read/Grep/Glob.

Steps

1. **Plan the search**. Identify likely fields/authors/claim types from the question (e.g. "transformer complexity" → field `nlp`, `complexity-analysis`). Note key entities. 2. **Scan**:

• grep `<vault>/papers/` and `<vault>/findings/` for the key terms (titles, statements, frontmatter values).
• Read the top candidates (papers: Key Takeaways + relevant sections; findings: statement + evidence).

3. **Synthesize**. Answer the question. **Cite** every factual statement with a wikilink — e.g. `[[2017-06-attention-is-all-you-need]]` or `[[finding-self-attention-is-O-n2]]`. If findings conflict, say so and cite both sides. 4. **Log** via the shared script:

   "${CLAUDE_PLUGIN_ROOT}/.venv/bin/python3" "${CLAUDE_PLUGIN_ROOT}/scripts/log.py" "<vault-path>" query "<one-line question>" "cited <n> files"

Guardrails

• Never fabricate a citation. If you can't find evidence, say so and suggest which paper to ingest.
• Prefer finding-level wikilinks over paper-level when a specific finding is relevant.
• Keep answers tight. The user drills into the vault via the citations.