Team Hiring

Team hiring

Full-funnel hiring skill. Mode A covers sourcing (JD + interview process + LinkedIn post + optional careers page). Mode B covers batch operations (interview hub per role per sprint) and per-candidate Stage-1 outputs (scorecard from Granola transcript + advance email or decline note). One skill, two operational halves, calibrated per client.

Doctrine inherited (Step 7 — 0626 rollout, locked 2026-06-04)

Output complies with output-tenets.md, output-simplicity.md, ai-speak-anti-patterns.md. Step 6 calibration: see [[feedback_execution_doctrine_refinements_step6]].

Refinements applied: R1 (JD + careers page → end-customer-facing → no sources; interview hub + scorecard → client-team review → cleaned cites in appendix), R2 (Mode A multi-asset pack ships as one doc with toggles per asset: JD + process + LinkedIn post + careers page), R3 (JD opener capability-led, never "we are thrilled to grow our team"), R6 (JD close → application primary CTA), R9 (verb-led role + section names).

Modes

When to run → which mode

Don't run when the user wants: organisational design (headcount planning, org chart), updating an existing JD after feedback (edit directly), coaching session or interview-prep for the hiring panel, or a reference-check script or offer-letter template.

Inputs

Mode A required (no defaults): org context (client folder), role title, artifact scope (3-pack vs 4-pack).

Mode B required (no defaults): org context, role (must already have an existing interview hub OR be creating one in Step 1), and per-step specifics:

• Step 1 (hub): role family (SDR / GM / PMM / engineer / etc.), candidate list for this batch (names + LinkedIn URLs + comp expectation + visa flag if relevant).
• Step 2 (scorecard): Granola meeting ID (or transcript) + the role's interview hub (for the canonical scorecard table) + user-stated pre-verdict (advance-clean / advance-with-caveat / reject).
• Step 3 (comms): the candidate's scorecard verdict + any caveats (office availability / comp recalibration / availability timing / format choice) for advance variants.

Recommended for both modes (pulled from client artefacts): voice rules (brand-name convention, em-dash policy, English variant, banned buzzwords, externally-safe vs internal-only proof), positioning + traction, ICP / buyer, team structure, mission + values. Source: client CLAUDE.md, positioning/, messaging/, icp/.

If any required input is missing, ask the user. Don't fabricate role specifics or candidate context.

Steps

Mode A — Sourcing artefacts (Steps A1-A8)

1. Resolve context. Read client CLAUDE.md, latest.md, canonical references. Pull voice rules, externally-safe proof points, team-member list, existing team-hiring/ precedents.
2. Confirm scope. Role title + reporting line + comp band (with role-family defaults). Artifact scope (3-pack or 4-pack).
3. Route to role-family patterns. Per the premium reference, identify 3-5 category-leading companies whose JDs and careers pages inform this hire.
4. Draft the JD. Use the 10-section template in the premium reference. Length 1,500-2,200 words.
5. Draft the interview process. Use the 3-stage template in the premium reference. Calibrate Stage 2 to role family. Length 1,200-1,500 words.
6. Draft the LinkedIn hiring post. Use one of the 4 hook archetypes from the premium reference (Challenge / Contrarian-Craft / Mission-Led / Story). Length 350-500 words.
7. (If 4-pack) Draft the careers page. WebFetch live About page first. Synthesise 5-7 operating principles from About + brand kit + voice + expert-pov. Length 1,200-1,800 words.
8. Voice-rule self-review + Gate 3 pause + push handoff. Per the premium reference voice-rule grep + push protocol.

Mode B — Batch operations (Steps B1-B3)

1. Build the interview hub (per role per sprint). Per the premium reference: this-week's-candidates table + Stage-1 screening script (60-min outline with per-section timing) + live exercise template (per role family) + scorecard template (7 dims for SDR, 10 for GM, role-family adaptations for PMM / engineer / CS) + per-candidate prep appendices (snapshot + three-risks-to-verify + customised opening hook + two probe questions + comp framing + Stage-2 handoff template) + canonical 3-stage process reference. Output: MMYY-{role}-interview-hub.md in client's team-hiring/ folder. Notion-publish as a child page; manifest line wired per .claude/rules/notion-protocol.md.

2. Score the candidate (per candidate per Stage-1 interview). Per the premium reference: pull Granola transcript via mcp__granola__get_meeting_transcript. Map transcript evidence to the role's canonical scorecard table (from the hub). Apply hiring bar ("no dimension below 3, at least 3 dimensions at 4+"). Verdict-up-top, verbatim transcript quotes per dimension, Stage-2 handoff template (or rejection rationale), "next move" footer. Three verdict shapes: advance-clean, advance-with-caveat (sub-variants per the premium reference), reject (rubric-anchored). Output: MMYY-{firstname-lastname}-stage1-scorecard.md. Notion-publish as child of the hub page (parent inheritance via the hub's pageId).

3. Draft candidate-facing comms (per candidate post-verdict). Per the premium reference: advance email (positive signal → next-stage framing → any-caveat-conversation framing → close) OR decline note (shape-of-role framing, not performance → named specific strength → genuine stay-in-touch close). Tone gate: "would I be comfortable if the candidate posted this on LinkedIn?" Output: MMYY-{firstname-lastname}-stage1-{advance-email|decline-note}.md. Local-only (operational draft, not stakeholder artefact). Voice rules per client CLAUDE.md.

What good looks like

Mode A quality bar (existing v1.0, unchanged):

• Voice match: reads like the client's existing brand voice. 100-Posts Test.
• Specificity over generality: KPIs quantified, team members named, reporting line explicit, comp a real band.
• Externally-safe proof only: internal-only metrics never appear.
• Role-family pattern visible without copying.
• Application format consistent ("No CV. Three things." per Genesys convention).

Mode B quality bar (new in v2.0):

• Hub: scorecard dimensions match the role family (SDR 7-dim, GM 10-dim, etc.). Stage-1 script timed to 30 minutes. Per-candidate prep appendix grounded in real LinkedIn + JD context (no fabricated risks).
• Scorecard: verdict at line 1 (not buried). Every dimension score backed by a verbatim transcript quote or explicit "not assessed in Stage 1" marker. Hiring-bar logic applied correctly. Stage-2 handoff template populated. "Next move" footer specifies who-does-what.
• Comms: advance email frames any caveat (office / comp / availability) honestly without making the candidate feel they're on probation. Decline note frames on shape-of-role, not performance; names a specific strength; warm but unambiguous.
• Voice rules per client: no em dashes if banned (e.g., ClientCo), brand-name spelling correct, no affirmative use of internally-banned metrics.

Self-evaluation (pre-delivery checklist):

Mode A (unchanged): voice match, KPIs quantified, team members named, comp band specific, brand-name convention, em-dash policy, English variant, no banned buzzwords, no internal-only proof, application format, interview Stage 2 role-appropriate, (if 4-pack) careers page operating principles client-specific + works at openings = 0.

Mode B (new):

• Hub: scorecard dimensions match role family, Stage-1 script timed to 30 min, per-candidate appendices grounded in real profile data.
• Scorecard: verdict at line 1, every dimension backed by verbatim quote, hiring bar applied, Stage-2 handoff populated, "next move" footer specific.
• Comms: advance email caveat framed cleanly, decline note shape-of-role + named strength + warm close.
• Voice rules: pulled from active client's CLAUDE.md, applied per-output.

Final ship gate

Run /premortem --output before ship. See /premortem skill for the 5 execution domains.

Trivial-case escape: ## Premortem\nNo failure modes — trivial change satisfies the contract.