Interview
Pre-plan scope builder that interviews the user one question at a time to flesh out scope before a build starts. Trigger naturally on vague build asks such as "flesh out scope", "help me spec", "not sure what I want", "spec this out", or "what should I build" — no explicit invocation required. Scores the emerging spec on a 4-dimension clarity scorecard and refuses handoff to plan-create below threshold.From its SKILL.md
npx -y skills add ProxiBlue/pb-hcf --skill interviewAssembled from the repository path, not quoted from the project. Check it against their README if it does not work.
One thing to look at
- 2 stars2 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.
SKILL.md
4.1 KB, as published. Nobody here has run it
/pb-hcf:interview
Pre-plan scope builder. Turns a vague build ask into a clarity-scored spec that /hcf:plan-create consumes verbatim. Implements the Harper Reed one-question-at-a-time interview pattern.
Auto-trigger, no gating. This skill carries no disable-model-invocation flag — it must fire naturally whenever the user's ask is vague about scope (see trigger phrases in the frontmatter description above). Do not require an explicit /pb-hcf:interview invocation; treat any vague build ask as the trigger.
Flow
- Intake ask — read the user's raw request as the interview seed.
- Silent codebase recon — glob/read any artefacts the user named (files, modules, configs) before asking anything. Never ask the user for information recon can obtain itself.
- Question loop — apply the one-question-per-turn rule: ask exactly ONE question per turn, never bundle two questions into a single turn. Prefer the
AskUserQuestiontool when available; fall back to plain prose when it is not.- Each turn, pick the lowest-scoring dimension on the current scorecard and craft the single question that would most improve it.
- Never ask more than 10 questions total in one interview. Each of the (at most) 10 questions targets only the current lowest-scoring dimension.
- Scorecard — after every answer, update and show the running scorecard (see below) before asking the next question.
- Threshold reached — once the total is 16/20 or above, stop asking questions and emit the Interview Spec block.
- Threshold not reached — if still under 16/20 after 10 questions, apply the refusal rule below.
Scorecard: 4-dimension scorecard (Goal/Constraints/Success Criteria/Context, 0-5 each)
Maintain a visible 4-dimension scorecard, updated every turn, across exactly these four dimensions:
| Dimension | Score |
|---|---|
| Goal | 0-5 |
| Constraints | 0-5 |
| Success Criteria | 0-5 |
| Context | 0-5 |
Show it after every turn, e.g.:
Scorecard: Goal 4/5 · Constraints 2/5 · Success Criteria 3/5 · Context 3/5 — total 12/20
Refusal rule: 16/20 threshold
Below the 16/20 threshold, refuse to hand off to /hcf:plan-create. Tell the user the current total out of 20 and which dimensions are weak, and offer two options: answer more questions, or force proceed.
If the user forces proceed while still below 16/20, emit the Interview Spec block anyway, stamped with:
CLARITY DEBT: <dimensions scored below 4/5>
List every dimension currently scoring below 4/5 in the CLARITY DEBT stamp.
Output: ## Interview Spec block
At handoff — whether reached at threshold or forced with a CLARITY DEBT stamp — emit exactly one markdown block headed ## Interview Spec, formatted for plan-create args consumption:
## Interview Spec
**Goal:** ...
**Constraints:** ...
**Success Criteria:** ...
**Context:** ...
**Open-but-defaulted items:** ...
(If forced below threshold, the CLARITY DEBT: <dimensions below 4/5> stamp is prepended immediately above this block.)
plan-create handoff
After emitting the ## Interview Spec block, suggest running /hcf:plan-create with the block pasted as its args. This is the plan-create handoff: plan-create consumes the ## Interview Spec block verbatim as its scope input — do not paraphrase or reformat it before suggesting the handoff.
What this skill does NOT do
- Does not gate behind
disable-model-invocation— it stays eligible for model-invoked auto-trigger on vague build asks. - Does not write files, run code, or call
/hcf:plan-createitself — output ends at the## Interview Specblock plus the handoff suggestion. - Does not exceed 10 questions per interview, regardless of clarity score.