Job loop prep

Build the kit for one interview loop, the way a product team prepares a launch: a spec (the JD crux), the launch artifacts (pitch, one-pager, leave-behind), a beta with outside judges (the mock set, scored by interview-coach), and the post-launch follow-through. Career-critical work: do it fully, or say exactly what is missing.

Paths below are relative to the kit's root. Installed as a standalone skill, copy references/ and scripts/ next to this file.

Frame

You need the company, the role, the loop stage, the interviewers' roles and the date. Ask once, up front, for anything missing that changes the kit; proceed on the rest.

Gather before building

Research first, format second. Sources: the JD; the candidate's claims ledger (every number); the interviewers' public background (role, team, what they have written); the company's own product and vocabulary; references/playbook.md. Never rely on memory for a number.

Step 0: JD crux decode

Read the JD bullet by bullet. For each of the top 6, one line: the problem the team cannot solve yet, and the crux, the one input without which the others produce nothing ("ensure teams operate efficiently" is about cohesion, not more planning; "own executive communication for accountability" says accountability is not happening on its own and needs a visible system plus an executive sponsor). A word repeated three or more times is the hiring manager circling a pain; write the hypothesis. Then one assumption-busting line: what this team assumes about the problem that the rest of the industry does not, said back in plain language. Save as jd-crux.md. It is page one of the kit and the input to every later step. (Nancy Chu, 2026-06-02.)

Claim freeze

Every number in the kit matches the ledger wording. Every growth claim names the role it happened in. Never say a result was achieved when the truth is that it was set up or started. A claim not in the ledger is left out, or marked as needing a reference before it is said out loud. Run:

python3 scripts/claim_check.py pitch.md claims-ledger.json

Build four artifacts

1. The pitch. A 90-second opener in the playbook's "tell me about yourself" shape: one line of experience aimed at the JD crux, why this team now, and a question back (Nancy Chu, 2026-10-06). Plant decision rights early. Run the altitude test: lever sentences (the bet, the moat, the business line, the org outcome) at least 2x feature sentences (Nancy Chu, 2026-06-23):

   python3 scripts/altitude_count.py pitch.md --duties --gate

2. The one-pager. The strategy artifact in the company's own vocabulary. It opens with the investment case (does owning this move the top line or protect a moat, or does it only save a fee), then the two or three moves, then how the org would know it is working (Nancy Chu, 2026-06-30; Simon Witten, post on strategy into delivery). Wireframes in the body, frameworks in the appendix.
3. The leave-behind. One page they keep, built around the one question you want them arguing about after you leave, taken from the assumption-busting line in jd-crux.md.
4. The mock set. The 8 most likely questions for this loop and these interviewers. For each: its playbook row (what it really tests, the shape that clears the bar, the source), the candidate's story that fits, and the one method sentence for that story (Nancy Chu, 2026-04-18). Strategy and product sense prompts carry the four investment tests; conflict prompts carry the escalation shape; people-management prompts carry the portfolio model. Hand the set to interview-coach, which returns a scorecard with the thinking audit.

Rehearse, do not just write

Every story gets a 30-second and a 2-minute version; the core stories also get 10 minutes. Say them out loud; reading is not rehearsing. For VP loops, each core story narrates an outcome through the org, not only work that was the candidate's own (Nancy Chu, 2026-07-21).

Follow-through (adapters)

The follow-through is a list of proposed tasks, not actions. The skill never sends a message or writes to a task manager itself. It emits one JSON object per line for an optional adapter to apply:

{"adapter": "tasks", "loop": "<company-role>", "kind": "post-onsite-artifact", "title": "Send the 48-hour follow-up artifact to the hiring manager", "due": "YYYY-MM-DD", "source": "post-onsite-48h.md"}
{"adapter": "tasks", "loop": "<company-role>", "kind": "thank-you", "title": "Thank-you note: <interviewer role>", "due": "YYYY-MM-DD"}

• post-onsite-artifact, due within 48 hours of the onsite: one page that answers the question the panel pushed hardest on, better than the room allowed. It passes the claim freeze like any other sendable draft.
• thank-you, due the day after: one per interviewer, each naming something specific from that conversation.

Adapter interface: read the JSON lines from stdin or a file, create one task per line in the user's own tool, and print what was created. With no adapter configured, write the lines to followups.jsonl and show them to the user. Draft any note the candidate will send with a drafting skill that enforces their voice (for example drafting-harness), and run its gates.

Deliver

The artifacts as files (jd-crux.md, the pitch, the one-pager, the leave-behind, the mock set, followups.jsonl), plus a five-line summary with the gate counts. A notification adapter is optional and follows the same pattern: propose the message, never send it from this skill.

Outside check before delivery

Report each count in the summary:

• pitch: lever sentences at least 2x feature sentences, duties lines 0 (altitude_count.py)
• every sendable artifact: 0 claim_check failures
• jd-crux.md: one line per top-6 bullet, plus the repeated-word hypothesis and the assumption line
• mock set: a playbook row cited for every question
• no em dashes; no banned words

Sources

Nancy Chu: 2026-04-18 (method sentence), 2026-06-02 (JD decode, assumption busting), 2026-06-23 (altitude), 2026-06-30 (four investment tests), 2026-07-21 (L7 vs L8+), 2026-09-29 (escalation), 2026-10-06 (tell me about yourself), 2026-10-20 (people management). Simon Witten: posts on strategy into delivery, tell me about yourself, and questions for us. Paraphrased with citations in references/playbook.md.

Files

• examples/outputs/: a full kit for an invented candidate and an invented JD (jd-crux.md, pitch.md, pitch-v0-before.md, mock-set.md, post-onsite-48h.md, followups.jsonl).