Your own agent builds, tests and maintains your measurement workflow.
A small kit of plain instruction files. Your agent reads them, interviews you, learns from reports you already trust, and builds a recurring workflow for capacity, cost, revenue and projection estimates from your spreadsheets. You approve every substantive choice.
Already unzipped? You have the kit: this page sits next to START-HERE.md.
One instruction to begin
Open your agent in the kit folder (or give it the files), and say:
START-HERE.md and begin Phase 0.The agent does the rest one phase at a time. Each phase ends at a gate, where it shows you what it produced and waits for your decision.
How it works
Fig. 1 · End state
- Your spreadsheetsread-only snapshots, never edited
- Your agent teamextract · compute · verify, inside budget limits
- Gateyou approve
- Outputscapacity · cost · revenue · projections, with a run manifest
↻ Maintenance loop: drift checks · periodic back-test · versioned changes
Fig. 2 · Install
- 0Environment probe (gate)
- 1Exemplar hunt (gate)
- 2Discovery interview (gate)
- 3Source inventory (gate)
- 4Metric dictionary (gate)
- 5Plan tailoring (gate)
- 6Build
- 7Back-test (gate)
- 8Dry run + release (gate)
- 9Maintenance (gate)
You decide
Metric definitions, sources, assumptions, methods, interpretation and release are yours. When the agent is unsure whether something is your call, it asks.
Everything is recorded
Every run writes a manifest: which files (with dates), which definitions and versions, which checks, and what you changed.
Checks, not confidence
The workflow must reproduce reports you already trust before it is released. A separate agent verifies what the builder built.
Install checklist
Tick phases as their gates are passed. Ticks are saved in this browser only.
Agents and budgets
Each phase has one role. Where your tool supports helper agents (sub-agents), the budget is the most that may run at once. The agent asks before exceeding it. You can change these defaults in Phase 5.
| Phase | Agent role | Budget | Key output | Gate |
|---|---|---|---|---|
| 0 Environment probe | Installer | 0 | environment.md | tool picture |
| 1 Exemplar hunt | Scout | ≤ 3 | exemplars.md | reference exemplar |
| 2 Discovery interview | Interviewer | 0 | requirements.md | requirements |
| 3 Source inventory | Cartographer | ≤ 1 per source family | sources.md | canonical sources |
| 4 Metric dictionary | Analyst | ≤ 2 | metrics.md | each definition |
| 5 Plan tailoring | Planner | 0 | workflow-plan.md | plan |
| 6 Build | Builder | ≤ 3 | recipes, calculations, templates | via Phase 7 |
| 7 Back-test | Verifier (not the Builder) | ≤ 2 | backtest.md | accept / send back |
| 8 Dry run + release | Orchestrator | ≤ 1 | manifest, runbook.md | release |
| 9 Maintenance | Maintainer | ≤ 2 | check reports | any change |
Whether to also cap time or usage per phase is left open for you to decide in Phase 5.
When the agent stops and asks
Before it:
- writes to, or changes, any shared system or source file;
- sends data anywhere outside the environment you cleared;
- handles personal or confidential data you have not cleared;
- changes a metric definition, assumption or tolerance;
- exceeds a phase budget;
- releases an output;
- installs software, adds a connector or changes permissions.
After the first release (optional)
Eight add-ons in follow-up/ make any agent workflow more durable. None is needed for the core to work.
F1 Instructions file
Rules loaded every session.
F2 Skill writing
Turn repeated tasks into versioned skills.
F3 Audit trail
"Where did this number come from?"
F4 Monthly review
Stale files, drift, definitions vs practice.
F5 Risk check
A structured pause before sensitive actions.
F6 Roles
Analyst, reviewer and presenter kept separate.
F7 Portfolio ledger
Only if you run several workflows.
F8 Start/close routine
Every session opens and closes the same way.