World-Class Benchmark Orchestrator

/lp-do-worldclass is the goal-validation, benchmarking, and ideas-emission layer for startup loop world-class analysis.

This skill does three things:

1. Validate the operator-authored goal artifact at <artifact_dir>/worldclass-goal.md and generate or refresh a deep-research prompt
2. When a benchmark result exists and is current, scan strategy artifacts against world-class standards via modules/scan-phase.md
3. Present improvement ideas to the operator for selection via modules/ideas-phase.md, then invoke /lp-do-ideas for each selected idea

Keep this file thin. Do not embed scan rules, benchmark rubrics, or ideas templates here.

Invocation

/lp-do-worldclass (--biz <BIZ> | --app <APP>) [--as-of-date <YYYY-MM-DD>] [--dry-run]

--biz and --app are mutually exclusive; exactly one is required.

Parameters

Global Invariants

Operating mode

SCAN + BENCHMARK

Allowed actions

• Read strategy artifact files, frontmatter, and goal/benchmark artifacts (read-only scan)
• Read <artifact_dir>/worldclass-goal.md and worldclass-benchmark.md
• Edit <artifact_dir>/worldclass-goal.md — benchmark-status field only (goal-phase Step 4 updates this after writing the research prompt)
• Write research prompt to <artifact_dir>/worldclass-research-prompt.md
• Write scan output to <artifact_dir>/worldclass-scan-<YYYY-MM-DD>.md
• Append to the scan output file — ideas-phase appends the ## Dispatches (Dry Run) block and ## Ideas Summary section
• Read docs/business-os/startup-loop/ideas/trial/queue-state.json (read-only; used for already-queued pre-filter in ideas-phase)
• Invoke /lp-do-ideas for each operator-selected idea (ideas-phase delegates all queue writes to /lp-do-ideas)
• Create <artifact_dir> if it does not exist — goal-phase may create the directory when --app mode resolves to a new app-level path

Prohibited actions

• Code changes, refactors, or production data modifications
• Destructive shell or git commands
• Writing directly to queue-state.json — all queue writes are delegated to /lp-do-ideas
• Running scan-phase or ideas-phase in any state other than State 3 (goal-phase runs in States 2, 3, and 4)
• Skipping the dry-run flag when the operator passes it — must respect --dry-run and never write to queue-state.json in dry-run mode
• Reusing a pre-existing scan file. If worldclass-scan-<YYYY-MM-DD>.md already exists for the current --as-of-date, it must be overwritten by a fresh scan — never read and reused as the output of this invocation. Scan-phase Steps 1–7 execute from scratch on every invocation without exception.

State Machine

The skill auto-routes to one of four states based on file presence and version alignment. Evaluate states in order; take the first match.

State 1 — No goal artifact

Stop with:

No world-class goal found for <BIZ> (app: <APP> | biz mode).
Expected path: <artifact_dir>/worldclass-goal.md
Create a goal artifact from the template at docs/plans/lp-do-worldclass/worldclass-goal.template.md before running lp-do-worldclass.

State 2 — Goal exists; no benchmark

Run goal-phase (validate goal artifact and generate research prompt). Then stop with:

Research prompt written to: <artifact_dir>/worldclass-research-prompt.md

Next step: run the prompt in your deep-research tool of choice and paste the result as:
  <artifact_dir>/worldclass-benchmark.md

Ensure the benchmark frontmatter includes:
  goal_version: <value from worldclass-goal.md>

Then re-run /lp-do-worldclass <original-invocation-form> to continue.

Where <original-invocation-form> is --app <APP> if invoked with --app, or --biz <BIZ> if invoked with --biz.

State 3 — Goal + current benchmark

Run all three phases in sequence: goal-phase → scan-phase → ideas-phase. Emit a summary to the operator on completion.

State 4 — Benchmark stale

Run goal-phase first (to refresh the research prompt if goal_version has changed), then stop with:

Benchmark is out of date for <BIZ> (app: <APP> | biz mode).
  goal_version in goal:      <goal.goal_version>
  goal_version in benchmark: <benchmark.goal_version>

The goal has changed since the benchmark was gathered. Rerun deep research using the refreshed prompt at:
  <artifact_dir>/worldclass-research-prompt.md

Paste the updated result as <artifact_dir>/worldclass-benchmark.md (replacing the existing file),
ensure frontmatter goal_version matches <goal.goal_version>, then re-run /lp-do-worldclass <original-invocation-form>.

Module Routing

Execute phases in the order shown. Only run scan-phase and ideas-phase in State 3.

1. Preflight — resolve target (--biz or --app), validate strategy directory, set BIZ, APP, artifact_dir, app_dir; fail-closed if checks fail (see Preflight Gate below)
2. State routing — determine current state from file presence and version alignment (see State Machine above)
3. modules/goal-phase.md — validate goal artifact structure; generate or refresh the deep-research prompt; write worldclass-research-prompt.md (runs in States 2, 3, and 4)
4. modules/scan-phase.md — read benchmark; scan strategy artifacts; compare against world-class standards; produce gap table (runs in State 3 only)
5. modules/ideas-phase.md — derive improvement ideas from scan gaps; present selection list to operator; invoke /lp-do-ideas for each selected idea (dry-run: skip selection, write dry-run annotations to scan file) (runs in State 3 only)

Preflight Gate

Run checks in this order; stop immediately on the first failure.

Check 1 — Missing target: if neither --biz nor --app is provided, stop with:

Error: one of --biz or --app is required.
Usage: /lp-do-worldclass (--biz <BIZ> | --app <APP>) [--as-of-date <YYYY-MM-DD>] [--dry-run]

Check 2 — Mutual exclusivity: if both --biz and --app are provided, stop with:

Error: --biz and --app are mutually exclusive. Provide one or the other.

Check 3 — App resolution (--app mode only): read docs/business-os/strategy/businesses.json. Find the entry whose apps array contains the --app value. Set resolved_biz = entry.id. If not found, stop with:

Error: App '<APP>' not found in businesses.json.
Verify the app identifier matches an entry in the apps array of a registered business.
Path: docs/business-os/strategy/businesses.json

Check 4 — Strategy directory: validate the parent business strategy directory exists:

• --biz mode: docs/business-os/strategy/<BIZ>/ — if absent, stop with:
  Copy

  Error: No strategy directory found for business <BIZ>.
  Expected path: docs/business-os/strategy/<BIZ>/
  Run from repo root and verify the business identifier is correct.
  

• --app mode: docs/business-os/strategy/<resolved_biz>/ — if absent, stop with:
  Copy

  Error: No strategy directory found for parent business <resolved_biz> (resolved from app '<APP>').
  Expected path: docs/business-os/strategy/<resolved_biz>/
  Run from repo root and verify the businesses.json entry for '<APP>' is correct.
  

After all checks pass — set resolved variables (passed to all modules):

BIZ          = --biz value  OR  resolved_biz from businesses.json lookup
APP          = --app value  OR  null (biz mode)
artifact_dir = docs/business-os/strategy/<BIZ>/                      (biz mode)
             = docs/business-os/strategy/<BIZ>/apps/<APP>/            (app mode)
app_dir      = apps/<first-app-in-BIZ-apps-array>/                    (biz mode — from businesses.json)
             = apps/<APP>/                                             (app mode — exact)

In biz mode, app_dir is resolved to the first entry in the business's apps array in businesses.json that maps to an existing directory under apps/. If no apps are listed or none exist, app_dir is null (scan-phase will skip app-directory probing).

Inputs

Where <artifact_dir> is resolved in the Preflight Gate:

• --biz mode: docs/business-os/strategy/<BIZ>/
• --app mode: docs/business-os/strategy/<resolved-BIZ>/apps/<APP>/

Output Paths

• Research prompt: <artifact_dir>/worldclass-research-prompt.md — written by goal-phase (States 2, 3, and 4)
• Scan output: <artifact_dir>/worldclass-scan-<YYYY-MM-DD>.md — written by scan-phase (State 3 only; written on both live and dry-run)
• Queue dispatches: docs/business-os/startup-loop/ideas/trial/queue-state.json — written by /lp-do-ideas for each operator-selected idea (State 3 live runs only; ideas-phase never writes directly to this file)