quick

name	quick
description	Fast answer to a question with vault context (the spoken answer is ungated; only captured notes are verified before writing). Usage: /learning-loop:quick "question" or /learning-loop:quick (infers from context). One-shot: web research + vault awareness + auto-capture if novel.

Quick: Fast Verified Answers

Overview

Answer a question quickly with web research, vault awareness, and source verification. One shot, no interactive rounds. Auto-captures noteworthy answers to the vault.

When to Use

• /learning-loop:quick "how much have jaguar prices dropped recently?": direct question
• /learning-loop:quick: infer question from conversation context

Process

Step 1: Parse the Question

If args provided: Use as the question.

If no args: Read recent conversation. Identify the question being discussed. If no clear question, ask the user with AskUserQuestion.

Step 2: Parallel Research

Spawn both subagents in the same turn (a single message with two Agent tool calls):

Vault Scout (discovery-vault-scout):

Search the vault for what the user already knows about this topic.

Topic: <question keywords>
Vault path: {{VAULT}}/
Angle: <the specific question being asked>

Return relevant notes with content, and identify gaps.

Researcher (discovery-researcher):

Answer this specific question with web research.

Topic: <the question>
Existing knowledge: (empty: vault results not available yet)

Focus on answering the question directly, not mapping the landscape.
Source-resolve any academic claims.
Return: direct answer, supporting evidence, sources with metadata, confidence level.

Step 3: Synthesize Answer

Merge vault-scout and researcher results into a direct answer.

Structure:

• If vault has relevant notes, lead with what the user already knows and how the new info extends or updates it
• If vault contradicts web findings, flag it explicitly: "Your note X says Y, but recent evidence shows Z"
• If vault has nothing relevant, lead with the web findings
• Keep it to 3-10 sentences. Cite sources inline.
• End with confidence: high (multiple concordant sources), medium (single source or mixed), low (sparse evidence, flag uncertainty)

Tone: Conversational. Plain language. No hedging paragraphs: if uncertain, one line says so.

Step 4: Auto-Capture Gate

Evaluate silently in the main thread. Do not ask the user.

Novelty check: Compare researcher findings against vault-scout results. If the vault already covers >80% of the answer's substance, skip capture.

Substance check: Is the core insight a durable pattern, mechanism, or decision-relevant fact? If it's transient (today's weather, a live score, a price that will change next week), skip capture.

If both pass, verify before writing. Run ONE round of the note-verifier subagent (subagent_type: learning-loop:note-verifier) with a one-note batch: a list with a single {path, content} entry, path the intended destination (e.g. 0-inbox/<slug>.md; the file does not exist yet, the path just names the output section) and content the draft note (placeholders resolved per agents-shared/vault-io.md). The agent returns one ## Verification: <note title> section per note; branch on its ### Status::

• PASS → write as-is.
• PARTIAL → write, carrying the verifier's [partial] claim flags into the note.
• ISSUES FOUND → apply the verifier's ### Corrections (adopt corrected URLs/claims). Contradicted (level-0) claims and fabricated references are REMOVED from the draft — never written, not even as source: unverified (same rule as /discovery's fabricated-reference handling). Any other claim with no surviving source moves out, or the note writes with source: unverified.

ONE round only — no revise loop. This is the lightweight tier; the spoken answer in Step 3 is NOT gated, only the captured note.

Then spawn a note-writer subagent (subagent_type: learning-loop:note-writer); the Insight/Research content in the prompt is the POST-correction draft (after the verifier's Corrections and [partial] flags are applied), not the original:

Write an inbox note for the Obsidian vault.

Insight: <the core durable insight from the answer>
Research: <researcher's findings with sources>
Related notes: <vault-scout's relevant note paths>
Source project: "conversation"
Date: <today YYYY-MM-DD>
Destination: 0-inbox/

Write the note to its destination using the Write tool, then report the exact written path.
Return the filename, title, and the written path when done.

This is a capture, not a promotion: leave destination_locked unset, so a gate-worthy answer can still be promoted out of 0-inbox/. The note may land in 1-fleeting/ or 3-permanent/ — read the folder from the agent's reported path, don't assume 0-inbox/.

If either fails: No capture. Move to report.

Step 4.5: Replay Post-Write Hooks

If Step 4 dispatched the note-writer subagent, replay the post-write hook chain on the new note. Subagent Write/Edit calls bypass PostToolUse, so without this the note misses the hooks/post-tool.js dispatcher (autolink + edge-infer modules).

Use the exact path the agent reported (the gate may have routed the note to a folder other than 0-inbox/; a reconstructed 0-inbox/<filename> path would point at a nonexistent file and the replay would silently fail). $NOTE_PATH is the absolute path built from the Written: line in the agent's response — never reconstructed from the requested destination:

printf '%s\n' "$NOTE_PATH" \
  | node "${CLAUDE_PLUGIN_ROOT}/scripts/sweep-hook-replay.mjs" --stdin

Skip this step if Step 4 didn't capture (novelty/substance gate failed). See skills-shared/hook-replay.md for the full pattern.

Step 5: Report

One line, using the actual folder the agent wrote to (read it from the agent's reported path):

Quick: "question" | Captured: "Note Title" → <actual-folder>/filename.md

or

Quick: "question" | No capture (already known)

or

Quick: "question" | No capture (transient)

Then done. No follow-up suggestions. No source handoff. Back to what we were doing.

Subagent Usage

Agent	Model	Role
discovery-vault-scout	Haiku	Keyword search + similar-note fan-out
discovery-researcher	Sonnet	2-3 web searches + source verification
note-verifier (conditional)	Sonnet	Step 4: ONE verification round, gates the captured note only
note-writer (conditional)	Sonnet	Atomic note in persona voice

Key Principles

• One shot. No interactive rounds, no steering, no "what angle interests you?"
• Answer the question. Not "here's the landscape." Direct answer with evidence.
• Vault-integrated. Weave existing knowledge into the answer naturally.
• Auto-capture. Novel durable insights get captured silently. No prompting.
• Fast. Shallow research depth. Two parallel agents. Minimal main-thread work.
• Honest. If evidence is thin, say so in one line. Don't hedge for three paragraphs.