Audit websites for accessibility issues and WCAG compliance. Use when checking accessibility, fixing a11y issues, or ensuring WCAG compliance.
日本語の概要は準備中です。原文の説明を表示しています。
Builds estimates, quotes, and bids for trades and field-service businesses (HVAC, plumbing, electrical, roofing, landscaping, painting, cleaning, remodeling, general contracting, IT/AV installs). Turns a scope into line items with waste factors, burdened labor, materials, equipment, subs, permits, mobilization, contingency, overhead, profit, and tax, then writes a client-ready xlsx (live formulas) and PDF plus an internal cost sheet, with good/better/best options, add-ons, a payment schedule, exclusions, and assumptions. Use whenever someone asks for a quote, estimate, bid, proposal price, or job price, asks "how much should I charge for this job", wants to check a price or margin on a job, or is confused about markup vs margin. Computes deterministically with a script instead of mental math, because models routinely treat a 30% markup as a 30% margin and drop labor burden, waste, and mobilization.
インストールする前に、エージェントに与えられる指示の中身を確認できます。
A quote is a promise priced in advance. The expensive mistakes are not in the unit costs; they are the items nobody priced (permits, disposal, mobilization, burden) and the arithmetic done in someone's head (a 30% markup is only a 23% margin). This skill fixes both: a checklist pass for missing scope, then scripts/build_quote.py for every number.
Never do the pricing arithmetic in prose. Write the spec, run the script, and quote its output. The script uses Decimal math and the same rounding as the Excel formulas it writes, so the PDF, the client xlsx, and the internal sheet always agree to the penny.
references/unit-cost-starter.md and label every figure in the output as a placeholder to replace. Never present an invented rate as a market price.references/scope-checklists.md for the trade and ask about each commonly forgotten item. Every item ends up either priced as a line or written into exclusions. The script also scans the spec and flags common items it cannot find, so an unaddressed permit shows up as a [CHECK] line instead of a surprise.base_items, tier-specific scope in tiers, and optional extras in add_ons.python scripts/build_quote.py spec.yaml --check # validate + totals, writes nothing
python scripts/build_quote.py spec.yaml --out ./quotes # xlsx + pdf + internal + summary.json
python scripts/build_quote.py spec.yaml --explain LINE # show every step for one line
Needs openpyxl; pyyaml for YAML specs; reportlab for the PDF (skipped cleanly if missing).[WARN]:
min_effective_hourly; if it is low, the hours are too high or the price is too low.--explain and show the math to the user so they can trust the rest.-internal.xlsx for the business only, since it shows costs and profit. List the assumptions and exclusions in your reply, plus every placeholder the user still has to replace.State which one you are using every time. The script refuses a spec that sets both.
python scripts/build_quote.py --margin-to-markup 0.30.In this script the target applies to loaded cost (direct + contingency + overhead), so it is a net profit margin. If the business sets overhead to 0, the target must be large enough to cover overhead too (a gross margin), and the script warns about it.
trade: hvac # selects the missing-item scan
pricing:
target_margin: 0.25 # OR target_markup, never both
by_component: {subs: {markup: 0.15}, fees: {at_cost: true}}
labor:
burden_pct: 0.32 # payroll tax, workers comp, insurance, benefits
roles: {lead_tech: {wage: 34.00}, helper: {wage: 19.00, burden_pct: 0.30}}
overhead: {pct_of_cost: 0.15, per_labor_hour: 0}
contingency_pct: 0.03
tax: {mode: charge_on_sell, rate: 0.07, taxable: [materials]} # or contractor_pays_on_cost, or none
min_effective_hourly: 120
base_items:
- id: LINE
section: Installation
description: Replace refrigerant line set, braze, pressure test, evacuate
qty: 25
unit: ft
materials:
- {item: Line set 25 ft, qty: 1, unit_cost: 165, waste_pct: 0}
- {item: Insulation tape, qty: 1.6, unit: roll, unit_cost: 9, waste_pct: 0.10, pack_size: 1} # buys 2 rolls
- {item: Fittings, qty_per_unit: 1.4, unit_cost: 1, waste_pct: 0.10} # scales with line qty
labor: [{role: installer, hours: 2.5}] # or hours_per_unit
subs: [{item: Electrical sub, qty: 1, unit_cost: 425}]
tiers: [{name: Good, description: ..., items: [...]}] # omit tiers for a single scope
add_ons: [{name: UV light, items: [...]}]
payment_terms: {schedule: [{milestone: Deposit, pct: 0.30}, {milestone: Completion, pct: 0.70}]}
exclusions: [...]
assumptions: [...]
Shorthand for simple lines: unit_cost (+ cost_type, waste_pct) and hours or hours_per_unit (+ role) directly on the line. Line flags: at_cost: true (pass-through), lump_sum: true (hide the unit price). client.view: lump_sum hides unit prices on every line.
Tax modes differ by jurisdiction and by job type, so ask rather than assume. charge_on_sell charges the client tax on the taxable share of each line. contractor_pays_on_cost adds tax to material cost, marks it up with the material, and notes "prices include sales tax on materials". Tell the user to confirm the treatment with their accountant.
<number>-client.xlsx: Summary (options and add-ons with totals referencing each sheet), one sheet per option, Add-ons, Terms (payment schedule by option, exclusions, assumptions, signature block). Amounts, subtotals, tax, totals, and schedule are live formulas; the last milestone takes the remainder so the schedule sums to the penny.<number>-client.pdf: the same content as a sendable document.<number>-internal.xlsx: every component with base qty, waste, order qty, burdened rate, contingency, overhead, loaded cost, pricing rule, sell; per-line profit and margin; per-option summary with gross margin, net margin, markup achieved, labor hours, effective hourly, and all checks.<number>-summary.json: the same totals and warnings in machine-readable form.examples/interior-painting.json (markup 50%, 6.25% tax paid on materials) produces a $6,180.95 base scope, 33.4% net margin, 68.68 labor hours, $74.26 effective hourly. Hand-check of the walls line (--explain WALLS):
Paint: 10.57 gal x 1.10 waste = 11.63 -> 12 gal (full cans) x $42 = $504.00, + 6.25% tax = $535.50
contingency 5% = $26.78; overhead 10% of $562.28 = $56.23; loaded $618.50; x 1.50 = $927.75
Patch kit: $18.00 + $1.13 tax -> loaded $22.10 -> sell $33.14
Labor: 1,850 sf x 0.0125 hr = 23.125 hr x $24.00 x 1.28 burden = $710.40
contingency $35.52; overhead 10% of $745.92 + 23.125 hr x $6.00 = $213.34; loaded $959.26; x 1.50 = $1,438.89
Line sell $2,399.79 / 1,850 sf = $1.2972 -> $1.30/sf x 1,850 = $2,405.00 (cent rounding added $5.21)
examples/hvac-system-replacement.yaml shows good/better/best tiers ($11,221.86 / $14,280.56 / $16,999.65 with 7% tax on materials), a sub at 15% markup, and a permit passed through at cost, which is why the blended margin lands near 24% against a 25% target.
references/terms-and-exclusions.md and tell the user to have their terms reviewed by their own attorney. Do not give legal advice.references/scope-checklists.md: commonly forgotten items per trade.references/unit-cost-starter.md: EXAMPLE placeholder rates and how to build real ones.references/terms-and-exclusions.md: plain-language terms, change orders, deposits, exclusion wording.まだレビューはありません。使ってみた感想をお寄せください。
概要と使いどころ
Audit websites for accessibility issues and WCAG compliance. Use when checking accessibility, fixing a11y issues, or ensuring WCAG compliance.
日本語の概要は準備中です。原文の説明を表示しています。
Deploy a 2-layer parallel agent hierarchy for large, parallelizable work — big refactors, multi-file migrations, codebase-wide audits, bulk generation. A top-tier commander (Fable or Opus) orchestrates the swarms; the user picks a power level (Max Power / Heavy / Balanced / Economy) that sets the Opus/Sonnet/Haiku model mix per layer. Layer 1 is 3-50+ specialist agents, each with its own full context window; Layer 2 is 2+ sub-agents per member. Includes git safety, tiered sizing, a pre-deploy gate, phantom-completion checks, and multi-wave follow-up.
日本語の概要は準備中です。原文の説明を表示しています。
Use to test whether an AI agent, prompt, or skill actually works before or after it ships — build a small eval set, score runs, and catch regressions when the prompt or model changes. Trigger when the user asks whether a prompt is good, wants to compare models or versions, sees inconsistent agent output, or is about to put an agent in front of customers.
日本語の概要は準備中です。原文の説明を表示しています。
Deploys swarms of sub-agents for massive parallel data processing tasks. Unlike agent-army (which is for code changes), this is for DATA tasks -- processing 1000 documents, analyzing datasets, bulk content generation. Configurable swarm size, task distribution, result aggregation, progress tracking, and error recovery.
日本語の概要は準備中です。原文の説明を表示しています。
Designs and deploys custom agent teams for specific business workflows. Interactive discovery of business processes, then generates complete team configurations with specialized agent roles, tool access, communication protocols, and handoff rules.
日本語の概要は準備中です。原文の説明を表示しています。
Agent-to-Agent (A2A) communication protocol. Connect two or more Claude agents that pass messages, share context, delegate tasks, and collaborate. Implements structured handoffs, shared memory, and multi-agent conversations.
日本語の概要は準備中です。原文の説明を表示しています。