feature-brief
Frame a feature before it is built — the user problem, a job story, one success metric with baseline and target, the adoption hypothesis, a click budget for the happy path, scope cuts, and kill criteria. Use when starting a new feature, epic or ticket, when someone says "we should build X", or when a spec has no measurable goal.
Try your first task
A source-grounded starter task has not been verified for this skill yet.
SKILL.md
name: feature-briefdescription: Frame a feature before it is built — the user problem, a job story, one success metric with baseline and target, the adoption hypothesis, a click budget for the happy path, scope cuts, and kill criteria. Use when starting a new feature, epic or ticket, when someone says "we should build X", or when a spec has no measurable goal.Feature Brief
Turn a feature idea into a one-page brief that says who it is for, how we will know it worked, and when we will stop. Most failed features weren't wrong in the code — they had no measurable goal, so nobody noticed they weren't used.
The brief is the root document of the feature folder: docs/product/<feature-slug>/brief.md, or wherever the project already keeps product docs. The other skills in this set read it — happy-path-audit takes its task and click budget, instrument-feature its metrics, ab-test-design its hypothesis.
Phase 1: Read before asking
Read what already exists: the ticket or PRD, linked discussions, the code the feature touches, existing analytics events and dashboards, the feature-flag setup, and any earlier files in the feature folder. Note what the codebase tells you for free: who can reach this screen (roles, plans), what data already exists, what the current workaround looks like.
Phase 2: Ask what only the user knows
Ask one consolidated round of questions, limited to what changes the brief:
- Who — which segment, role or plan? First-time or returning users?
- Evidence — what shows the problem is real (support tickets, interviews, data, churn reasons)? "Nothing yet" is a valid answer; the brief will mark it as an assumption, and
user-interviewscan gather the evidence. - Success — what would make this worth it in six weeks? Push from outputs ("feature shipped") to outcomes ("teams send their first invoice in under two minutes").
- Constraints — deadline, platforms, what's off limits.
Offer choices with a "decide for me" option where you can propose a sensible default. Don't ask what the code or the ticket already answers.
Phase 3: Write the brief
Use this structure. Keep it to one screen and cut any section with nothing real in it.
# <Feature name>
**Problem.** <One sentence, in the user's words.>
**Evidence.** <Data, tickets, quotes — or "Assumption: not yet validated.">
**Job story.** When <situation>, I want to <motivation>, so I can <outcome>.
**Today's workaround.** <What users do now and what it costs them: time, clicks, errors, money.>
## Success
| Metric | Type | Baseline | Target | By |
|---|---|---|---|---|
| <% of active teams that send ≥ 1 invoice> | Primary | <value, or "unknown — measure first"> | <value> | <date> |
| <support tickets tagged "invoicing"> | Guardrail | | must not rise | |
| <first invoice within 7 days of signup> | Leading | | | |
## Adoption hypothesis
- **Eligible users:** <who should use it, and how many>
- **Discovery:** <where they'll find it, at the moment they need it>
- **Value moment:** <the action that proves value — not "opened the page">
## Happy path
<Start state → numbered steps → goal state.>
**Budget:** ≤ <N> clicks, ≤ <M> fields.
## Scope
- **In:** …
- **Out, and why:** …
- **Later:** …
## Rollout and kill criteria
- Behind flag `<name>`; ramp plan <…>; A/B test: yes/no, and why.
- **Kill or rework if** <primary metric> < <threshold> by <date>.
## Risks and open questions
- …
Phase 4: Challenge the brief
Check the draft against these failure modes and fix or flag each one:
- Unmeasurable metric. Can the primary metric be computed from events that exist today? If not,
instrument-featureis a prerequisite — say so. - Vanity metric. Page views, clicks on the entry point and "users who saw it" measure exposure, not value. Replace them with the value moment.
- Cheaper solution. Could a copy change, a better default, or surfacing an existing feature solve the problem? Say so. It's the most valuable sentence a brief can contain. When demand itself is the open question, propose a fake-door test first: the entry point ships alone, counts the clicks, and tells users honestly that the feature is coming.
- Solution posing as a problem. "Users need a dashboard" is a solution. Rewrite it as the need.
- No kill criterion. A feature without one is never removed. Pick a threshold the team would genuinely act on;
launch-reviewchecks it at the review date. - No budget. If the happy path has no click budget, estimate one now;
happy-path-auditwill refine it.
Output
Write brief.md to the feature folder. Summarize in three lines: the primary metric and its target, the biggest assumption, and the next skill to run — usually happy-path-audit when the flow exists, or instrument-feature when the baseline is unknown.