20 HR Skills · build, don't install.

name: job-description-writer
description: Draft job descriptions that lead with outcomes (not duties), use inclusive language, and surface 3-5 specific success indicators. Activates when user asks to write, edit, or improve a job description, role spec, or job ad. Pulls from the team's existing JD library; otherwise uses the org's level taxonomy and comp band as constraints.
inputs: role title, level, team, key outcomes, must-have skills, comp band
references: inclusive-language-guidelines.md, level-taxonomy.md, voice-profile.md
## Examples

### Invocation
`/job-description-writer` for Senior Eng Manager, Platform team, L7,
must own platform v3 migration, comp band $230-280k

### Expected output
- Title + level tag (L7) + team
- 3-5 outcomes (not duties) tied to platform KPIs
- 3 success indicators with measurable signals
- Must-haves (4-6) + nice-to-haves (3-4), all level-calibrated
- Inclusive-language pass + posted-ready format
- NOT a wall of bullet points · a JD a hiring manager can actually post

name: interview-question-generator
description: Generate behavioural and situational interview questions tied to specific competencies in the job description. Produces 3 deep questions per competency with follow-up probes, each tagged to the competency it tests.
inputs: JD, candidate level, interview stage (screen/onsite/exec)
references: competency-library.md
## Examples

### Invocation
`/interview-question-generator` for L5 PM onsite, 4 panelists,
45 min each, JD attached, focus on product judgment

### Expected output
- 3 deep questions per competency (4 competencies = 12 questions)
- Each question tagged with the competency it tests
- 2 follow-up probes per question
- Calibration rubric (1-5) so panelists score consistently
- NOT "tell me about yourself" generics · sharp behavioural questions only

name: candidate-evaluation-synthesizer
description: Synthesize interviewer notes from a loop into structured evaluation: signals for and against per competency, calibrated against the JD bar, with an explicit recommendation and confidence. Flags unconscious-bias language for review.
inputs: JD, interviewer notes, level expectations, rubric
references: competency-rubric.md, bias-language-checklist.md
## Examples

### Invocation
`/candidate-evaluation-synthesizer` 5 interviewer notes
attached, Director loop, JD bar attached

### Expected output
- Signals-for + signals-against per competency
- Hire / no-hire recommendation with confidence (e.g. 4/5)
- 2-3 calibration notes vs the level bar
- Bias-language flags surfaced for human review
- NOT a long narrative · a debrief-ready structured doc

name: outreach-message-drafter
description: Draft personalised outreach by reading candidate profile data and identifying one specific hook (a project, company, transition) before pitching the role.
inputs: candidate profile URL, role context
## Examples

### Invocation
`/outreach-message-drafter` candidate LinkedIn URL attached,
role: Sr Product Designer at Highlevel

### Expected output
- 4-line message max
- One specific hook (project, transition, company) from their profile
- One specific reason this role intersects their trajectory
- Refuses to send if no hook found · prompts for more context
- NOT "Hi {first_name}" generics

name: offer-letter-drafter
description: Draft offer letters consistent with the org's comp philosophy, level expectations, and legal language. For negotiation responses, propose 3 options (hold the line / partial accept / counter-with-justification) with rationale.
inputs: candidate name, role, offer terms, negotiation context
references: offer-letter-template.md, negotiation-playbook.md
## Examples

### Invocation
`/offer-letter-drafter` L6 Eng, $215k base + 0.05% equity + 15% bonus,
candidate countering at $230k

### Expected output
- Full offer letter (legal-language-checked)
- 3 negotiation responses: hold / partial / counter to $225k
- Comp-philosophy rationale per option
- Trade-offs flagged (e.g. cash vs equity vs sign-on)
- LEGAL REVIEW callout for any non-standard clauses

name: talent-pipeline-review
description: Read ATS export, summarize pipeline state by role and stage, flag bottlenecks (long-stuck candidates, drop-off cliffs), and suggest 3 specific actions. Only reports on what's in the data · never invents candidates.
inputs: ATS export, target hire dates per role
## Examples

### Invocation
`/talent-pipeline-review` 12 open eng roles, 8 weeks of ATS data,
quarter close in 6 weeks

### Expected output
- Funnel snapshot per role (top of funnel · screen · onsite · offer)
- 3 roles flagged as stalling with specific bottleneck
- Suggested intervention per stalled role (sourcing · screening · offer)
- 1-line forecast: which roles likely close in time

name: performance-review-prep
description: Help managers structure performance reviews by reading source material (1:1 notes, project outcomes, peer feedback) and producing a draft review aligned to the org's competency rubric. Surfaces specific behaviours, never generic adjectives.
inputs: employee name, role, level, source material, rubric
references: competency-rubric.md
## Examples

### Invocation
`/performance-review-prep` for [team member], 6 months of 1:1
notes + project artifacts + peer feedback attached

### Expected output
- 3 evidence-backed strengths (with specific moments)
- 2 specific growth areas (with examples · not abstract)
- Calibration note vs level expectations
- 4-bullet talking-point card for the 1:1
- NOT a generic review template · this person's actual signal

name: coaching-conversation-prep
description: Pre-conversation prep that identifies the leader's hypothesis, the 3 best questions to test it, and 2 likely defensive responses with re-framing. Outputs a 1-page brief, not a script · leaves the conversation itself to the human.
inputs: situation, employee context, desired outcome
## Examples

### Invocation
`/coaching-conversation-prep` Q3 OKR at 40% complete,
direct report, conversation Friday

### Expected output
- Conversation map: open · understand · co-diagnose · co-plan
- 6 specific questions to ask (open-ended, not leading)
- 3 statements to AVOID (e.g. "I'm disappointed")
- What a good outcome looks like (3 signs)
- Tone calibration: care + clarity, not either alone

name: pip-letter-drafter
description: Draft PIPs that meet the legal bar (specific measurable expectations, timeline, support, consequences) while remaining humane. Flags language that could be discriminatory.
inputs: employee role, performance gaps, manager support plan, timeline
references: pip-template.md, legal-language-checklist.md
## Examples

### Invocation
`/pip-letter-drafter` for Alex Chen, Senior Analyst,
3 consecutive missed deadlines, 60-day plan, direct but supportive

### Expected output
- 2-sentence opener naming the pattern without blame
- 3 specific behavior changes, each with a 30-day checkpoint
- Weekly check-in cadence + measurable goals
- Closing transition to remediation timeline
- LEGAL REVIEW callout for termination-trigger language
- NOT a verbatim script · bullets the manager speaks naturally from

name: career-conversation-framework
description: Generate a tailored career conversation guide based on the employee's current level, time in role, and stated aspirations. Outputs 4 question clusters covering current state, desired state, gap, and first move.
inputs: employee name, level, tenure, aspiration
## Examples

### Invocation
`/career-conversation-framework` 3-year IC, top performer,
eyeing Sr → Staff promo, conversation next week

### Expected output
- 5-question conversation flow:
  1. What's motivating you now?
  2. Where are you strongest today?
  3. What's the gap to Staff?
  4. What do you want to be known for?
  5. What do we agree to next?
- 90-day experiment plan with 2-3 concrete bets
- NOT a career-ladder doc · a real talk template

name: engagement-survey-analyzer
description: Analyse open-text and quantitative survey data: cluster open-text responses by theme, surface 3 hot-spot themes vs prior wave, recommend 3 listening-tour questions. Preserves anonymity · never quotes responses tied to individuals.
inputs: survey CSV/JSON, prior wave data
## Examples

### Invocation
`/engagement-survey-analyzer` 2,400 free-text comments
+ scores from Q3, PII already stripped

### Expected output
- 5 dominant themes ranked by frequency
- 3 sample anonymized quotes per theme
- 2 unexpected signals (the weak signals)
- 3 team-level hotspots flagged for action
- DM-ready summary per manager (1 paragraph)
- Refuses to call a theme with fewer than 5 supporting comments

name: compensation-benchmark
description: Take a role spec and produce a market-anchored comp range using only the data provided (Radford, Mercer, Pave, Levels.fyi, or company bands). Surfaces the assumption: if you only have headline data, it'll say so.
inputs: role, level, location, comp data sources
references: comp-philosophy.md, level-taxonomy.md
## Examples

### Invocation
`/compensation-benchmark` Sr Eng, NYC, L6, benchmark
base + equity vs Levels.fyi p75

### Expected output
- 1-page band recommendation (p50 / p65 / p75)
- Every number cited to source (Levels.fyi · Radford · internal peers)
- Retention risk note if we stay at current
- 2-3 mitigations (refresh equity · sign-on · band adjust)
- NOT a single number · a defensible range with rationale

name: org-design-helper
description: Help People leaders pressure-test an org design proposal. Inputs the current org and the proposed change; outputs 5 risks, 3 alternative structures, and the people-impact map. Recommends only structural options · not specific terminations.
inputs: current org chart, proposed change, business rationale
## Examples

### Invocation
`/org-design-helper` 24-person eng org, current 1 manager,
need 3 squads of 7-9 with mixed levels

### Expected output
- 3 candidate structures (e.g. by product · by tech layer · by customer)
- Trade-offs per option (focus vs flex vs depth)
- Staffing matrix per option: who reports where, by level
- 5 risks (e.g. single-point-of-failure) + mitigation
- NOT one "right answer" · 3 options with the call left to you

name: skills-gap-analysis
description: Given a target capability map and current team skills (from a survey or exported assessment), identify the top 5 skills gaps with severity ratings and build/buy/borrow recommendations.
inputs: capability map, current skills inventory
## Examples

### Invocation
`/skills-gap-analysis` data team roster (12 people) +
2026 roadmap (ML pipelines, real-time, governance)

### Expected output
- Skills heatmap: current vs needed, per person
- 3 capability gaps ranked by roadmap risk
- Build-vs-hire-vs-train recommendation per gap
- Timeline + cost rough-cut per option
- NOT a generic skills matrix · this team, this roadmap

name: internal-comms-drafter
description: Generate internal comms in the company's voice. For sensitive comms (layoffs, restructures), produces 3 stakeholder versions: leaders / managers / individual contributors. Always includes "what employees should do next" CTA.
inputs: announcement type, audience, key facts, voice profile
references: voice-profile.md, comms-checklist.md
## Examples

### Invocation
`/internal-comms-drafter` org-wide announcement, return-to-office
3 days/wk starting Sept, ED + legal cleared, sensitive

### Expected output
- 3 drafts at different empathy/firmness levels
- Suggested 8-12 question FAQ
- 1:1 talking points for managers
- Likely all-hands Q&A questions + answers
- NOT a single canonical draft · 3 options the ED can pick from

name: policy-drafter
description: Draft HR policies grounded in the org's values, applicable employment law, and existing related policies. Outputs in plain English with a separate "manager guidance" section. Flags areas requiring legal review with [LEGAL REVIEW] markers.
inputs: policy topic, location/jurisdiction, related existing policies
## Examples

### Invocation
`/policy-drafter` remote-from-anywhere policy, 6 weeks/yr,
employees in US · UK · India · Germany

### Expected output
- Draft policy doc with sections (scope · process · approval)
- 8 jurisdictional flags (tax · employment law · benefits)
- In-scope vs out-of-scope examples
- Enforcement model + escalation
- Employee FAQ (8-10 questions)
- LEGAL REVIEW callout on tax residency rules

name: dei-language-reviewer
description: Read text and surface gendered, ableist, age-coded, or culturally exclusive language with specific replacements. Tags each suggestion with the bias type. Doesn't rewrite the document · leaves the user in control.
inputs: any text artifact
references: inclusive-language-guidelines.md
## Examples

### Invocation
`/dei-language-reviewer` JD + offer letter + onboarding deck attached

### Expected output
- Marked-up doc with flagged terms (gendered · ableist · age-coded)
- Suggested rewrite for each flag (keeps your voice)
- One-line "why this matters" per flag
- Tone-preservation note: refuses to over-correct
- NOT a generic find/replace · context-aware edits

name: onboarding-sequence-designer
description: Build a Day 1 / Week 1 / Month 1 / Month 3 plan with named owners, artifacts, and check-ins, tailored to the role and level. Pulls in role-specific reading lists if a team-onboarding/ folder exists. Always includes a 30-day feedback loop.
inputs: role, level, team, start date, manager name
## Examples

### Invocation
`/onboarding-sequence-designer` new VP Eng, starts Aug 18,
inherits 4 direct reports + quarter mid-flight

### Expected output
- 30-60-90 day plan with weekly milestones
- Curated 1:1 list (people · what to learn · what to avoid)
- 3 "early wins" to look for
- "Don't touch yet" list (what to leave alone in first 60 days)
- First stakeholder map (who they need to know by Day 30)

name: meeting-notes-to-actions
description: Read meeting notes (raw or from Granola/Otter/Fireflies) and extract decisions, owned actions with due dates, and open questions. Flags ambiguous ownership rather than fabricating owners.
inputs: meeting transcript or notes
## Examples

### Invocation
`/meeting-notes-to-actions` 60-min leadership-team
Granola transcript attached

### Expected output
- Decisions (3-5, each with rationale)
- Owned actions with due dates (each tagged to a person)
- Open questions flagged (not answered today)
- "Ambiguous owner" callouts (don't fabricate ownership)
- Follow-up meeting suggestions if needed
- NOT a meeting summary · an action doc

name: people-analytics-narrator
description: Take HR metrics and write a 1-page narrative for an executive audience: what's the story, what's the risk, what's the recommendation. Cites every number to its source. Refuses to draw conclusions without enough data points.
inputs: HR metrics dashboard export, period, prior-period comparison
references: metric-definitions.md
## Examples

### Invocation
`/people-analytics-narrator` HRIS Q3 export (headcount · attrition ·
hires · comp ratio · eNPS · time-to-fill)

### Expected output
- 1-page board narrative · 3 sections: story / risk / recommendation
- Every number cited to source + time period
- 2-3 specific risks with severity
- 2-3 specific recommendations with owner + timeline
- Refuses to call a trend with fewer than 3 data points