cf-ask

/cf-ask

Answer the question: $ARGUMENTS

Purpose

Quick, focused Q&A about the codebase. Proactively explores code to find the answer, then saves the Q&A to project memory so it can be referenced later.

• Unlike /cf-research: single focused answer, no multi-doc output
• Unlike /cf-remember: proactively explores the codebase to answer vs extracting knowledge already in conversation

Output: {CF_DOCS_ROOT}/memory/{category}/{name}.md. Never use relative paths in write specs (nested git repos); always {CF_DOCS_ROOT} / absolute.

Workflow

Step 0: Custom Guide

bash "<plugin-root>/lib/load-custom-guide.sh" cf-ask

If the block above printed anything, apply only the ## Before, ## Rules, and ## After sections; if it shows the raw command instead of output, re-run that exact load-custom-guide.sh fence now.

Step 1: Parse the Question

1. Read $ARGUMENTS as the question

2. If no question provided, ask the user what they want to know

3. Identify keywords and likely relevant areas (modules, features, patterns)

4. Classify the question type — check if the question is a flow question. A flow question asks about how something works end-to-end, how components interact, or what happens when a process runs. Trigger words: "how does X work", "flow of", "lifecycle", "sequence", "process", "when X happens", "walk me through", "how are X connected", "what triggers", "what happens when", "pipeline", "chain". Non-flow questions (lookup/definition/pattern/why) do NOT trigger this path.

   Set IS_FLOW_QUESTION = true/false for use in Steps 3, 4, 5, and 6.

Step 2: Check Existing Memory (Memory Recall)

Before exploring the codebase, search existing memory docs.

Recall memory (see Verbs) with 2–3 keywords from the question: { "query": "<keywords>", "limit": 5 }; grep scope {CF_DOCS_ROOT}/memory/**/*.md, third tier = file content.

After finding matches:

Read the top 2–3 most relevant matched files before treating a hit as a direct answer.