Ops & Scale OS
The entry point. Diagnoses which of the four layers is the constraint and routes to one skill.
When to use
- First session in a new environment
- The user describes a symptom rather than a task ("our lists are bad", "the agent gives wrong answers")
- Two or more skills could plausibly apply
- Somebody wants a tour of what is here
The routing question
Ask exactly one thing first: what is broken, and what does the evidence look like?
Then route on the symptom, not on the request. People ask for the fix they have heard of, and that is usually a layer above the actual problem.
| Symptom |
Actual layer |
Route to |
| "We need an AI agent for outbound" |
Usually data, not agents |
gtm-systems-audit first |
| Reports disagree with each other |
Systems of record |
reporting-architecture |
| Match rates are low or credits are burning |
Data plane |
enrichment-waterfall, then credit-and-cost-control |
| The score does not predict anything |
Data plane |
icp-scoring-engine |
| Duplicates, stale records, bad emails |
Data plane |
data-hygiene |
| Signals fire and nothing happens |
Data plane |
signal-pipeline |
| Nobody uses the thing we built |
Agent plane |
agent-rollout |
| The agent output is confidently wrong |
Quality plane |
eval-loop, then eval-spec-authoring |
| Quality was fine and now is not |
Quality plane |
eval-calibration |
| We have too many tools |
Systems of record |
martech-stack-audit |
| Somebody wants a warehouse, a CDP, or reverse ETL |
Data plane |
warehouse-and-pipes |
| People are pasting customer data into a chatbot |
Agent plane |
ai-governance |
| Leads sit untouched, or go to the wrong rep |
Systems of record |
lifecycle-and-routing |
| Nobody agrees what marketing sourced |
Systems of record |
attribution-architecture |
| We do not know where to start |
All four |
gtm-systems-audit |
The rule that governs routing
Never route up the stack to solve a problem that lives down the stack. An agent will not fix a data model, an eval will not fix an agent nobody uses, and a dashboard will not fix a definition nobody agreed on.
If the answer to "what is broken" involves records, fields, duplicates, or scores, the answer is Layer 1 regardless of what the user asked for. Say so plainly, and explain the order.
The layers
| Layer |
Question it answers |
Skills |
| Data plane |
Who is worth touching, and when |
data-model-and-sync · tam-and-sourcing · enrichment-waterfall · icp-scoring-engine · signal-pipeline · data-hygiene · credit-and-cost-control · warehouse-and-pipes |
| Agent plane |
What does the work |
agent-scoping · context-portfolio · skill-authoring · agent-guardrails · ai-governance · agent-rollout |
| Quality plane |
Is any of this good |
eval-spec-authoring · eval-loop · eval-calibration · agent-observability |
| Systems of record |
What is true, and who owns it |
martech-stack-audit · attribution-architecture · lifecycle-and-routing · reporting-architecture |
Gates that run on output regardless of layer: writing-quality, artifact-eval.
Cold start, in order
gtm-systems-audit – score all four layers, get the constraintdata-model-and-sync – define the objects before building on themdata-hygiene – clean what exists before enriching iticp-scoring-engine – score, then set thresholds from the real distributioncontext-portfolio – write the operating manual before any agent reads from itagent-scoping – decide what deserves to be builtai-governance – set the data boundary before anything reaches a teameval-spec-authoring – write the rubric before the thing it judgesagent-observability – instrument before scaling
Output
- Names the constraint in one sentence
- Routes to exactly one skill, with the reason
- States what to do after that skill finishes
Rules & quality bar
- One skill at a time. Loading four skills in parallel bloats context and produces mush
- Diagnose before routing. No route without a stated symptom and its evidence
- Say the unwelcome thing. If somebody wants an agent and needs a data cleanup, say that first
- Refuse to build on a broken layer. Note the dependency and route down the stack
Related skills
- Everything in this repo. Start at
gtm-systems-audit when the answer is unclear
- See also:
docs/architecture.md for the layer model, docs/build-surface.md for the build primitives
1---2name: ops-and-scale-os3description: The router for Ops & Scale OS. Use when the user says "ops and scale", "GTM engineering", "AI ops", "build the GTM system", "our data is a mess", or is not sure which skill they need. Reads the situation, names the constraint, and routes to exactly one skill. Also use to orient somebody new to the repo. Front door only. It routes and does not do the work: the audit itself is `gtm-systems-audit`, and scoping one agent is `agent-scoping`.4---56# Ops & Scale OS78The entry point. Diagnoses which of the four layers is the constraint and routes to one skill.910## When to use11- First session in a new environment12- The user describes a symptom rather than a task ("our lists are bad", "the agent gives wrong answers")13- Two or more skills could plausibly apply14- Somebody wants a tour of what is here1516## The routing question1718Ask exactly one thing first: **what is broken, and what does the evidence look like?**1920Then route on the symptom, not on the request. People ask for the fix they have heard of, and that is usually a layer above the actual problem.2122| Symptom | Actual layer | Route to |23|---|---|---|24| "We need an AI agent for outbound" | Usually data, not agents | `gtm-systems-audit` first |25| Reports disagree with each other | Systems of record | `reporting-architecture` |26| Match rates are low or credits are burning | Data plane | `enrichment-waterfall`, then `credit-and-cost-control` |27| The score does not predict anything | Data plane | `icp-scoring-engine` |28| Duplicates, stale records, bad emails | Data plane | `data-hygiene` |29| Signals fire and nothing happens | Data plane | `signal-pipeline` |30| Nobody uses the thing we built | Agent plane | `agent-rollout` |31| The agent output is confidently wrong | Quality plane | `eval-loop`, then `eval-spec-authoring` |32| Quality was fine and now is not | Quality plane | `eval-calibration` |33| We have too many tools | Systems of record | `martech-stack-audit` |34| Somebody wants a warehouse, a CDP, or reverse ETL | Data plane | `warehouse-and-pipes` |35| People are pasting customer data into a chatbot | Agent plane | `ai-governance` |36| Leads sit untouched, or go to the wrong rep | Systems of record | `lifecycle-and-routing` |37| Nobody agrees what marketing sourced | Systems of record | `attribution-architecture` |38| We do not know where to start | All four | `gtm-systems-audit` |3940## The rule that governs routing4142**Never route up the stack to solve a problem that lives down the stack.** An agent will not fix a data model, an eval will not fix an agent nobody uses, and a dashboard will not fix a definition nobody agreed on.4344If the answer to "what is broken" involves records, fields, duplicates, or scores, the answer is Layer 1 regardless of what the user asked for. Say so plainly, and explain the order.4546## The layers4748| Layer | Question it answers | Skills |49|---|---|---|50| **Data plane** | Who is worth touching, and when | `data-model-and-sync` · `tam-and-sourcing` · `enrichment-waterfall` · `icp-scoring-engine` · `signal-pipeline` · `data-hygiene` · `credit-and-cost-control` · `warehouse-and-pipes` |51| **Agent plane** | What does the work | `agent-scoping` · `context-portfolio` · `skill-authoring` · `agent-guardrails` · `ai-governance` · `agent-rollout` |52| **Quality plane** | Is any of this good | `eval-spec-authoring` · `eval-loop` · `eval-calibration` · `agent-observability` |53| **Systems of record** | What is true, and who owns it | `martech-stack-audit` · `attribution-architecture` · `lifecycle-and-routing` · `reporting-architecture` |5455Gates that run on output regardless of layer: `writing-quality`, `artifact-eval`.5657## Cold start, in order58591. `gtm-systems-audit` – score all four layers, get the constraint602. `data-model-and-sync` – define the objects before building on them613. `data-hygiene` – clean what exists before enriching it624. `icp-scoring-engine` – score, then set thresholds from the real distribution635. `context-portfolio` – write the operating manual before any agent reads from it646. `agent-scoping` – decide what deserves to be built657. `ai-governance` – set the data boundary before anything reaches a team668. `eval-spec-authoring` – write the rubric before the thing it judges679. `agent-observability` – instrument before scaling6869## Output70- Names the constraint in one sentence71- Routes to exactly one skill, with the reason72- States what to do after that skill finishes7374## Rules & quality bar75- **One skill at a time.** Loading four skills in parallel bloats context and produces mush76- **Diagnose before routing.** No route without a stated symptom and its evidence77- **Say the unwelcome thing.** If somebody wants an agent and needs a data cleanup, say that first78- **Refuse to build on a broken layer.** Note the dependency and route down the stack7980## Related skills81- Everything in this repo. Start at `gtm-systems-audit` when the answer is unclear82- See also: `docs/architecture.md` for the layer model, `docs/build-surface.md` for the build primitives
guerrilla2799