The moves you reach for once the Six-Layer Stack is second nature.
The Session 5 hour was deliberately 101: name the six layers, install the toolkit, customize one skill end-to-end. If you walked out of that room and the framework already feels obvious, this page is for you. Five moves that the live hour didn't have time for — each one is what separates a team that uses skills from a team that compounds them.
The 101 move: one trigger phrase fires one skill. The 201 move: one trigger phrase fires a sequence of skills, where each skill's output is the next skill's input.
Concrete example from the toolkit: a single trigger — "ramp the Acme account" — chains four skills that all ship in GTMify/aigtm: prospect-research → abm → meeting-prep → deal-strategy. The first skill's company profile feeds the second skill's account plan and buying-committee map. The second skill's stakeholder list becomes the third skill's "who do I prep for" input. The third skill's briefing becomes the fourth skill's strategy starting point. Forty-five minutes of work for one prompt.
| Pattern | What it does | When to reach for it |
|---|---|---|
| Sequential chain | Skill A's output is Skill B's input | Multi-step research, account warm-up, end-to-end deal review |
| Fan-out / fan-in | One input → N skills run in parallel → one synthesizer skill merges results | Portfolio-wide audits, multi-account briefings, batch enrichment |
| Conditional dispatch | A router skill picks which skill to run next based on its own output | "Triage this lead" workflows where the next move depends on tier |
In the live hour we edited a single skill's Layer 1 (Context) with your company, ICP, and competitors. That's the universal pattern. There are four customization patterns underneath it that compound differently:
Put your company, ICP, top 3 competitors, value prop, and typical deal size into ~/.claude/CLAUDE.md (or the platform equivalent). Every skill reads it automatically. Highest leverage. Lowest maintenance.
For named accounts, drop an accounts/<account-slug>.md file with the account's own ICP variant, internal stakeholders, deal history, and competitive dynamics. Skills that touch named accounts reach into it first. Especially powerful for the meeting-prep and deal-strategy families.
For roles that come up repeatedly in your motion (CRO, VP RevOps, FP&A lead), keep a personas/<role>.md file with the role's typical priorities, vocabulary, and decision-making patterns. Skills that draft for these roles inherit the voice automatically.
Layer 1 isn't write-once. Put a recurring 15-minute calendar block on the first Monday of every quarter to refresh: new competitor, repositioned value prop, revised ICP. Skills inherit the new world automatically — no per-skill edits required.
This is the question the live hour didn't have time for. Three cases where reaching for a skill is the wrong move:
Layer 4 (Constraints) in the 101 hour was a single sentence: "No more than 250 words. No 'it depends.' If you don't have enough info, ask one clarifying question instead of guessing." Good. Now scale it across a team and a portfolio.
| Constraint type | Example | What it prevents |
|---|---|---|
| Source-bound | "Only use information from the pasted context. If a fact is not present, say 'unknown' rather than inferring." | Confident-sounding fabrications about specific accounts, deals, or people |
| Quantitative bounds | "Do not include any percentage or dollar amount unless it appears verbatim in the source." | Plausible-but-invented stats that survive into customer-facing artifacts |
| Tone bounds | "Hedge any claim about competitor pricing. Never say 'they charge X' — say 'public list price suggests X'." | Legal exposure on competitive collateral |
| Refusal bounds | "If the input does not contain a stage, value, or close date, refuse the task and request the missing fields." | Skills running on bad input and producing confident garbage |
The 101 move: one in-line example in Layer 5 to anchor the output. The 201 move: treat examples as an asset. When a teammate produces a great output, add it to the skill's example library. The skill gets sharper every week without anyone editing the prompt itself.
Mechanics:
examples/ subdirectory.What this unlocks: a team writing about a skill's output, not fighting it. The library is the artifact that turns a one-person prompt into a team-owned asset.
If you're going to chain skills (Section 1), the Output Spec stops being about readability and starts being about contract. Two skills compose cleanly when the first emits a structured object the second can parse.
Concrete pattern: a research skill that emits plain prose looks like:
Acme is a $50M ARR healthcare-payments company...
Their CFO recently mentioned compliance as a top priority...
The CTO has a strong preference for open-source tooling...The same skill with a 201 Output Spec:
{
"company": "Acme Health",
"arr_band": "$25M-$75M",
"industry": "healthcare-payments",
"stated_priorities": ["compliance", "cost-takeout"],
"stakeholders": [
{"name": "CFO", "stated_concern": "compliance"},
{"name": "CTO", "stated_preference": "open-source"}
],
"confidence": "medium",
"missing_fields": ["renewal_date", "current_stack"]
}The first version is readable. The second is composable. A downstream deal-strategy skill can consume it without re-parsing prose. Every chained pipeline I run depends on this discipline.
~/.claude/CLAUDE.md. Every skill instantly inherits.examples/. Reference them in Layer 5. Watch quality compound."The 101 move is to write a better prompt. The 201 move is to write a prompt you never have to write again — and that your teammates can extend without asking you first."