/cf-advise
Manual onlyMedium1,500 – 3,000 tokens injected into promptnewDecision advisory through a structured interview that ends in a verdict-first recommendation.
Context footprint: (medium) — what does this mean?
Only the --save flow uses the memory MCP from coding-friend-cli for indexed storage. Without the CLI: falls back to a direct file write under docs/memory/decisions/. The advisory itself needs no CLI. See CLI requirements.
The /cf-advise skill helps you decide, not build. It turns a vague question into an actionable decision through a structured interview — surfacing the hidden requirements behind the surface question — then delivers a verdict-first recommendation with pitfalls and ranked alternatives. It sits before /cf-plan in the workflow: decide → plan → build.
It is advisory-only — it never writes or edits code, and never writes a plan file. To act on the advice, you run /cf-plan.
Usage
/cf-advise [question]
Flags
| Flag | Effect |
|---|---|
--quick (alias --fast) | Fewer interview questions — aim for the 2–3 highest-signal ones. Use when the decision is narrow. |
--save | After the verdict, persist the decision to docs/memory/decisions/ and index it in memory. Off by default — advice stays in chat. |
What It Does
- Analyzes the Input — Reads the decision, extracts what's stated, what's implied, and the hidden assumptions baked into the question. Classifies the decision as codebase-relevant or abstract.
- Scouts the Codebase (conditional) — For codebase-relevant decisions, launches the
cf-exploreragent to ground the advice in real constraints and prior art, and searches existing decisions in memory. - Interviews One Question at a Time — Never batches questions. Asks one, waits, and forms the next from your answer — walking the arc why → pros/cons → alternatives → constraints → convergence (4–8 questions, 2–3 with
--quick). - Confirms the Reframing — Restates the decision as concrete requirements and goals, not the original vague question, and gets your explicit confirmation before advising.
- Delivers the Verdict — Leads with the recommendation, then justifies.
- Emits Outputs — By default the advice stays in chat with a pointer that acting on it is
/cf-plan. With--save, persists and indexes the decision.
Verdict-First Output
The recommendation is delivered in this order — the answer first, never buried:
- Verdict — the recommendation in one or two sentences, no hedging preamble.
- Why — the reasoning, grounded in your interview answers and (if gathered) codebase evidence.
- Recommended actions — the concrete next steps if you follow the verdict.
- Pitfalls — what to watch for and what commonly goes wrong with this choice.
- Ranked alternatives — the other options, ordered, each with a one-line "choose this instead if…".
Examples
/cf-advise Should we adopt a monorepo or keep separate repos?
/cf-advise --quick Is it worth adding Redis caching to the API?
/cf-advise --save Which state management library should we standardize on?
When to Use
- Deciding whether or which — not how to build
- Weighing a trade-off between concrete options
- Pressure-testing a vague idea before committing to a plan
- Any question shaped like "should I", "is X worth it", "which approach", "am I overthinking this"
How It Differs
- Unlike
/cf-plan— decides whether / which, not how to build. No plan doc, no execution. - Unlike
/cf-ask— reasons about a decision or trade-off, rather than looking up how existing code works. - Unlike
/cf-research— a focused verdict for one decision, not a multi-doc research dump.