Decompose Nested Intent From Transcript

---
name: decompose-nested-intent-from-transcript
description: Take a specific transcript and produce an intent map of the client's system — restructured to lead with the critical opportunities and critical leverage, in the user's exact voice, with every relevant verbatim quote integrated and noise removed. The artifact is the highest-fidelity signal about the client's intent — their actual words on the most substantive opportunities, with all critical context that makes those opportunities lucid, clear, and born. Use when a transcript needs to become a customer-reviewable map that aligns all parties around the leverage points the user would have surfaced if writing it themselves.
triggers:
  - "/decompose-nested-intent-from-transcript"
  - "build intent map from transcript"
  - "nested intent map"
  - "transcript to intent map"
  - "map this conversation"
---

# Decompose Nested Intent From Transcript

## The Skill — The Canonical Objective (the user's exact words)

Your objective is to take a specific transcript and try to extend the exact way in which the user communicates — the exact way, the exact exact way. You're gonna be an extension of the user's voice. What you're gonna do is you're gonna go ahead and structure that in such a way that you're building a map of the intent of the client, so you're restructuring the conversation such that it leads with the critical opportunities and the critical leverage just like the user would've done in the same exact way — but you'll improve it in an additive fashion, and you'll remove things that are noise that are not relevant. So the result is a direct communication about the leverage and opportunity that effectively integrates the highest fidelity signal about the intent of the client, which is their actual words on the most substantive opportunities, and all of the critical context out of which those opportunities are made lucid, clear, and born.

### How to derive this from any transcript

Read the transcript fully and locate the moments where the user is reflecting back to the client what the user is hearing — those passages are the calibration target for the artifact's voice. Identify what the client is actually asking for help with by tracking which specific opportunities the user named as critical leverage — those become the load-bearing structure of the artifact. Then restructure: lead with those critical opportunities, weave the supporting context as integrated multi-perspectival paragraphs, embed every quote that grounds signal, strip every utterance that was noise to the actual intent. The artifact is what the user would have written themselves if the user had taken three hours to translate the conversation into a map for the client to review.

## The Voice Calibration Examples (verbatim demonstrations of how the user reflects)

The four examples below are the canonical demonstrations of the user's reflective communication style. They are pulled directly from the source transcript and are the voice the artifact must mirror.

### Example 1 — Reflecting back the asset thesis by holding environment + technology + market position simultaneously

> "So — if you dig up [the resource], it harms the environment. The value isn't in the dug-up [resource] necessarily — it's a store of value. So if you leave it in the ground but can prove it's there, then you don't have to dig it up. This is a new way of exchanging on the value of [the resource] that's actually superior for the planet. And there's new technology that has made this possible. Is that right?"

> "So if my parents sold a house and weren't ready to buy one and were going to lose [a large sum] in taxes — they can store their value in your asset class and it fulfills the requirement."

> "You're taking the [established trust structure] — already a proven, massively advantaged opportunity — and then solving the [multi-year] lock-up problem programmatically, so you become one of the only and best stores of value on the planet to a savvy investor. Stable, inflation-resistant, and liquid."

What this shows: you compress an entire model (environment + tax + structural innovation + market position + competitive advantage) into a few sentences. You don't sequence the perspectives — you weave them and check the synthesis with "is that right?"

### How to derive this synthesis pattern from a transcript

When a client describes a multi-component model across several minutes, the user's pattern is to restate it back compressed into one short utterance that holds all the components at once. To produce this in the artifact, scan the transcript for the client's explanations of structure, isolate the load-bearing components, and recompose them into a single sentence-or-two that names environment + mechanism + market + advantage in one breath. Confirm the synthesis lands by ending with the client's own corrective opening ("is that right?") even if implicit — that posture keeps the artifact honest about which parts are derived and which were said directly.

### Example 2 — The "you're in your groove" synthesis (the canonical example)

> "My read so far is that you're in your groove. The things you've shown me represent an unfathomable scope of opportunity almost at your fingertips. What you showed me with your investors is the most sophisticated investor management I've seen — further than anyone else I know. The network is extensive. [The flagship project] alone could be your only focus and would be worth it. The [head-of-state] connection. And the system is effectively working and has what it needs to do what it's going to do a month from now.
>
> If you hit that timeline — keeping it on the rails might be the highest leverage thing. Whatever the lawyers are doing, if one's being lazy — detect it, do the work for them, give it back so they can give their approval.
>
> My actual sense of the highest leverage opportunity is this: if we took your will, with you steering it as you already are, and made it more extensible — if I'm asking the right questions that surface what you'd do if you had twice as many hours in the day, and then spotting the 40% of those that could be effectively automated, and just building those automations. Your will unfolding at an accelerated pace."

What this shows: three paragraphs holding eleven distinct things at once — assessment of her current state, scope recognition, sophistication observation on investor management, the network, the flagship project as standalone-worthy, the head-of-state connection, system readiness, timeline, leverage observation about lawyer friction-removal, your actual proposal of the will-extension model, the automation pattern. None of those things are in their own paragraph. They're integrated. Each sentence is multi-perspectival.

### How to derive this density from a transcript

When the user is articulating the user's read of where the client is, the user holds many distinct observations in a single passage — assessment, recognition, opportunity-naming, leverage-observation, proposal — without sequencing them into separate paragraphs. To produce this in the artifact, gather all the observations the user made about a given holon and compose them as one or two paragraphs where each sentence carries multiple lenses. Resist the impulse to give each observation its own sentence or its own paragraph — the integration is the value. The reader should feel the weight of multiple things being held simultaneously, not a sequential list of items.

### Example 3 — Strategy synthesis

> "You have a proven offer. You have a funnel that gets people on a call with you, which results in a purchase. One thing I could do is download what's working and scale it, which would increase available capital immediately.
>
> The strategy: build a private community, grow it through LinkedIn virality (I've cracked the algorithm — was getting 50,000 views per post average), use story artifacts to drive top-of-funnel, then invite the community to the private offer. The private community is not 'the public' so the securities law restriction doesn't apply.
>
> There's also a crowdfunding platform exception — you can use a crowdfunding platform and advertise a security to the public. That's been legal for about four to five years."

What this shows: you name what she already has (proven offer + funnel), name the immediate-leverage move (download + scale), then construct an integrated strategy in one sentence (community → virality → story artifacts → top-of-funnel → private offer) with the legal nuance woven in (private community exception), then add an alternative path (crowdfunding exception). Five components, one paragraph, signal-dense.

### How to derive a strategy synthesis from a transcript

When the client has multiple disconnected things going (a proven offer, a funnel, dormant socials, a regulatory constraint), the user's pattern is to weave them into one strategic flow that names what's working, names the immediate-leverage move, constructs the integrated strategy in one sentence, weaves the constraints in mid-flow, and adds the alternative paths at the end. To produce this in the artifact, identify the client's existing assets and constraints, then draft a single paragraph that flows from "what you have" through "what we do with it" with the constraints integrated as natural conditions rather than as separate caveats. Constraints are part of the strategy, not separate from it.

### Example 4 — Top-of-funnel + scale problem + solution all at once

> "Top of funnel: independent advisors, private investor groups, property sales networks. They're all on LinkedIn. And you'd hit a problem of your calendar being too full if this works — so we record a VSL, put all the common investor questions in it, use it to filter people before calls."

> "I'd probably buy the data of all of them, start with that chunk, refine the funnel until it's perfect."

What this shows: audience identification + channel + risk-of-success + solution-to-that-risk + execution path — all in two short sentences. You don't separate "here's the audience" from "here's the risk" from "here's the solution." You name the audience and immediately load the next two perspectives onto it.

### How to derive this density from a transcript

When the user names a tactical play, the user loads the success-path and the failure-mode into the same sentence rather than treating them as separate considerations. To produce this in the artifact, when you describe a recommended initiative, pre-load the constraint or risk that would emerge if it succeeded, and the solution to that constraint, into the same passage. The reader sees the play and the protection-against-its-own-success in one breath, which is how the user actually thinks about plays.

## What I See In The User's Voice From These Passages (the calibration summary)

You hold multiple perspectives simultaneously in the same sentence and let the integration carry the meaning. You don't sequence — current state, then target state, then leverage. You bring all the lenses to bear at once and trust the listener to see them woven. You compress models into single utterances. When you reflect back what you heard, you're not summarizing — you're synthesizing into a new statement that holds more than what was said because it integrates what was said with what you saw and what you now sense.

The artifact should sound like this. Each holon's body is paragraphs that synthesize current state + target + gap + leverage + relationship to parent + relationship to siblings + quotes — all woven, the way you wove "you're in your groove" — not separated into per-perspective slots.

### How to operationalize this voice in the artifact

Before writing any paragraph in the artifact, name to yourself the multiple perspectives that paragraph needs to hold. Then compose a single sentence that opens with one perspective and folds the others into the same sentence as relative clauses, parenthetical observations, or coordinated phrases. If you find yourself starting a new sentence with a transition word like "additionally" or "separately," that's the failure signal — restructure to weave the next perspective into the existing sentence rather than starting a new one. The integration is mechanical at the sentence level, not just at the paragraph level. When you finish a paragraph, count the distinct lenses it holds — if it holds fewer than three, you have not yet synthesized; if it holds three or more, the integration is happening.

---

# What Wrong Looks Like (failure modes observed during the teaching session)

These are the failure modes that produced correction cycles when the skill was being taught. Each one is the kind of mistake that breaks the integration the canonical objective requires. Future agents loading this skill should recognize these patterns in their own drafts and restructure before producing.

## Failure mode — Lifecycle-grouping instead of holon-grouping

The agent produced two top-level sections — one called "Where they are" containing the company's cash position, team composition, asset readiness, pipeline status, and dormant outbound channels all jumbled in shared paragraphs; and another called "Where they're going" containing the post-launch posture, treasury operational state, fund deployment, network propagation, and impact-as-marketing endpoint all jumbled together. A reader could not navigate to "the cash position" and see its full lifecycle, because the lifecycle was fragmented across the two top-level buckets. This is the structural inversion failure: the agent treated perspectives as the parent and holons as the children, when the holon must be the parent and the perspectives integrate within it.

What right looks like instead: each holon (the cash position, the team configuration, the asset readiness, the pipeline, the dormant outbound) becomes its own section. Inside each section, the concept's current state, its evolution, its target state, and the intent of the traversal all live together — woven into integrated multi-perspectival paragraphs as the canonical voice examples demonstrate.

## Failure mode — Block-quoted callouts followed by commentary

The agent produced sections where each paragraph was structured as "interpretation in normal text" followed by a `>` block-quoted callout, followed by more interpretation, followed by another block quote. The reader's eye toggled between two registers — interpretive prose and quoted callout — instead of reading a single integrated paragraph. The quote and the interpretation never integrated because they were formatted as separate things.

What right looks like instead: an interpretive sentence opens, a quote is integrated mid-sentence with light framing, additional synthesis observations follow that use the source's facts (without always direct-quoting every one), and a closing interpretive sentence lands the thought. The quote is a fragment of the same paragraph, not a callout. The four canonical voice examples above demonstrate this pattern — every quote sits inside the prose, never set apart from it.

## Failure mode — Conversational meta inside the artifact

The agent wrote "What's deepest under this branch" and "the third thing surfacing here" and "let me articulate this" inside the artifact. A reader has no shared context for "branch" or "the third thing" — those are conversation-with-the-principal terms, and they make the artifact unreadable to anyone except the principal. Worse, if the source-conversation participant reads the artifact and finds themselves narrated about ("she said directly," "{Name} was emphatic"), the experience is jarring and the artifact loses its ability to serve its purpose of being customer-reviewable.

What right looks like instead: every reference inside the artifact is to a named concept the reader can locate by reading the document. No numbered branches. No "first thing / second thing." No process meta. No "she said directly" framing. Quotes drop in unframed. The agent's reasoning happens before writing; the writing shows only the result.

## Failure mode — Promoting candidate ideas to confirmed scopes

The agent wrote "The AI Wizard Scope" as a section heading inside the artifact, treating an idea floating in the conversation as if it were a confirmed bounded thing. The principal had explicitly said the AI wizard role was one candidate way of relating to the client, not a confirmed scope. Naming it as "the scope" presented an idea as shared reality, which is hallucination embedded in the artifact.

What right looks like instead: candidate ideas are articulated as candidates. Language reflects their actual status — "one candidate way of relating to X" or "an idea surfaced in the conversation about Y" — not "the X scope" or "the Y initiative." The artifact reflects what was committed to vs. what is open.

## Failure mode — Private context bleeding into client-reviewable material

The agent included context about a team member's [private family matter] inside the artifact, casually, in the same register as operational content. The principal had shared that context only so the agent would understand why certain work-modules were open, not so it would appear in client-reviewable material. Including it was inappropriate for the client's review.

What right looks like instead: maintain an explicit separation between "context for understanding" (which shapes the agent's interpretation but doesn't appear in the main intent-map artifact) and "content for the artifact" (which the client could read). When private context exists about a named individual, the right move is not to delete it — that loses signal future agents will need — but to write it to a separate file at `partnerships/{partner-slug}/people/{name}.md` and reference that file from the main artifact only with a flag like `*[private context for {Name} captured separately at people/{name}.md]*`. The main artifact stays customer-reviewable; the context is preserved where future agents can find it but the client never sees it. When in doubt about whether a piece of context belongs in the main artifact, route it to `people/` or `context/` rather than deleting it or inlining it.

## Failure mode — Undefined jargon left in quotes

The agent quoted source material that referenced specific named tools ([the SaaS product], [the in-house tool]) without ever defining what those tools were or why they mattered. The quote landed as fancy-sounding jargon — the reader could not tell whether to care or what those terms even named.

What right looks like instead: every named entity in a quote has a brief grounding nearby. "[The in-house tool] — the in-house data-room tool the team built — replaced [the SaaS product], the standard SaaS data-room product." Now the quote is meaningful.

## Failure mode — Stripping interpretation in over-correction

The agent, after being corrected for hallucinated assumptions, over-corrected by removing all interpretive synthesis and producing only quote-grounded factual descriptions. The artifact lost its ability to articulate nested intent because the synthesis layer that connects quotes into a coherent meaning was gone.

What right looks like instead: interpret nested intent honestly. Every interpretive claim sits next to a quote that grounds it. Where interpretation goes beyond what was said, label that as interpretation rather than presenting it as quote-derived fact. The interpretation is the work; stripping it to be "honest" defeats the purpose.

## Failure mode — Label headings instead of essence headings

The agent used headings like "Fund," "Cash," "Team," "Pipeline" — labels rather than essences. A reader scanning headings sees a list of nouns, not a map of nested intent.

What right looks like instead: headings carry the essence of the concept, not the label. A heading like "The Core Asset Going Live In Roughly A Month Is The Legitimacy Event That Converts The Vision Into A Capital Pipeline Advisors Can Underwrite Against" is the right kind of heading. Each heading carries enough nuanced signal that scanning only the headings would tell the full story.

## Failure mode — One perspective per paragraph

The agent proposed a structure where each holon would have separate H3 paragraphs for "where this initiative currently sits," "where this initiative is heading," "the gap and leverage," and "how it serves the parent." This is the same fragmentation pattern moved one level deeper — perspectives separated into their own paragraphs instead of integrated into shared paragraphs. The four canonical voice examples above demonstrate the opposite — perspectives weaving inside the same paragraph.

What right looks like instead: each paragraph holds multiple perspectives at once, woven together as the canonical examples demonstrate. The heading above the paragraph also synthesizes — it does not name a single lens.

## Failure mode — Walls of text with no headings

The agent produced sections with single paragraphs running 200-400 words covering multiple concepts. The reader had to mentally re-segment the paragraph to find the part relevant to the concept they cared about.

What right looks like instead: when a paragraph is covering more than one concept, split it into sub-paragraphs each governed by its own heading. Headings carry the nesting. Frequent headings at varying levels guide the reader through the structure visually.

## Failure mode — Holons fractured across the artifact

In one prior run, the agent created a section called "Where the company is going" that talked about a key asset going live, then a separate section called "Where the company is now" that talked about that same asset sitting at the legal threshold. The asset's holon was fractured across two top-level sections, forcing the reader to read multiple pages to assemble the picture for that one thing. This is the holon-boundary violation: perspectives became the parent, and the holon got split across them.

What right looks like instead: the parent intent (the company's spanning purpose, derived if not stated) is at the top. Distinct initiatives in service of that parent are children. Each initiative is its own holon — its own section — and within that section all perspectives integrate. The Fund section contains the Fund's current state, target, gap, leverage, parent connection, and quotes — all in one place, woven into integrated paragraphs.

## Failure mode — Artifact scatter instead of partner-folder-first

In one prior run, the agent wrote partner-specific artifacts to multiple parent folders simultaneously — placing VALIDATION inside a tooling-initiative folder, PROPOSAL-DRAFT inside an unrelated venture folder, and OPEN-QUESTIONS fragmented across both. None went to `partnerships/{partner-slug}/`. A reader looking for "everything about this partner" had to traverse three directory trees to assemble the picture, and a cleanup subagent had to be dispatched after-the-fact to consolidate.

What right looks like instead: before any artifact is written, the agent ensures `partnerships/{partner-slug}/` exists (creates it if not), and writes every partner-specific artifact under that root from the first write. Reusable infrastructure (live transcript pipes, shared coaching loops, tooling that supports multiple partners) lives in `1.5_initiatives/{tool-name}/` and the partner folder cross-links to it. The partner folder never absorbs reusable tooling; the tooling never absorbs partner artifacts. See "Where The Output Lives" below for the canonical structure.

## Failure mode — Conflating the enablement tooling with the partner initiative scope

When the principal demonstrates a tool during a call (a live transcript pipe, an auto-wake coaching loop, a real-time intent-mapper), in one prior run the agent treated that tool as a child initiative inside the client's holon tree — listing the principal's own tooling as if it were one of the partner's strategic initiatives. This is mis-parenting: the tool is infrastructure the principal brought to the call, not a thing the client is doing. It does not belong inside the client's nested intent map.

What right looks like instead: the tool gets its own initiative folder at `1.5_initiatives/{tool-name}/` if one does not already exist. The partner folder cross-links to the tool with a note like "See `1.5_initiatives/{tool-name}/` for the infrastructure used during this session." If the tool is genuinely co-developed with the partner (the partner is funding it, building it, or owns part of the IP), it gets a dedicated holon inside the partner's tree with the cross-link to the shared tooling folder explicit. Default posture: tooling is the principal's, the partner folder references it.

# The Holon Structure (what right looks like at the structural level)

The artifact is a tree of nested intents. The parent intent — the spanning purpose the system serves, which may need to be derived if the client did not state it explicitly — sits at the top. Beneath the parent are distinct initiatives, each one a child holon serving the parent. Beneath each child are sub-children where the substance warrants further unpacking. The boundary of any node is defined by it being a distinct initiative derivable in service to its parent.

A holon only has children when those children are substantive enough to warrant their own section. If a perspective on a holon is fully expressible in one paragraph, it stays as a paragraph within that holon — it does not become a separate child node. Only when the substance of a perspective opens into its own gap and unpacking does it earn its own holon status.

Within each holon, the gap between current state and target state is what matters. Naming "current state is X" and "target state is Y" is meaningful only because together they define the gap of Y minus X — and it's the gap that holds the leverage. Unpacking the gap (the why-chain that explains what's holding the closing back) is what generates the children of a holon.

Cross-references between holons are where the integration becomes visible. One holon's content is often the cause of another holon's gap — citrulline malate sitting inside the supplements holon being what blocks the pickleball holon's gap is the canonical example. The artifact names these cross-links explicitly, because the relationship between holons is where the value lives, not just inside any single holon.

## The pickleball example (the user's worked demonstration of holon structure)

> "I have the aspiration of being incredibly helpful, healthy. And so in order to do that one thing I'm doing is taking on pickleball and another thing that has synergy with that is that I've got this DIY Gatorade drink that I'm making and within that there's 20 different supplements and within that there's a ton of intent around those supplements. So I'm reviewing my life map, then I might be looking at those supplements and I might be curious about the gap between what I envision for those supplements for like full vitality and where I'm at right now, but I'm also gonna wanna know why I'm doing it. I'm gonna want a reminder of how that connects to its parent intent which is to be incredibly healthy and the sibling intent of like playing pickleball for six hours a day."

> "each distinct pile of initiatives to fulfill a parent initiative — that's the boundary of the holon — and then within each thing that is that you're using the perspectives of the gap from where we're at to where we're going and where there's leverage within that and the direct quote all of that context gets packaged in many paragraph sections with H with headings and so on within like the boundaries of the lines or the distinct intentions that are aspiring to serve the parent intent, not things like each perspective we take on the thing the perspectives are integrated. This is an integrally informed system where the relationship between the different perspectives is where all the value lies so you cannot separate by perspective any initiative."

## The unpacking example (the user's worked demonstration of how children of a holon emerge)

The pickleball holon: target is six hours a day, current is three hours a day, gap is the missing three hours. The gap can be unpacked: what's holding back closing the gap? The back can't handle it. Why can't the back handle it? Not enough oxygenation in the blood today. Why? Skipped citrulline malate, which lives in the sibling supplements holon. Each "why" in the chain becomes a deeper child of the pickleball holon, and the cross-link to the supplements holon (where citrulline malate lives) is named explicitly. Without the chain, "the gap is three hours" is meaningless; with the chain, the leverage point is visible (take the citrulline malate).

# Where The Output Lives (canonical filesystem layout for every run)

This is the scaffolding shared across every run of this skill, regardless of which partner is being mapped. The partner folder is created BEFORE any artifact is written, not after-the-fact. Scattering artifacts across `ventures/`, `1.5_initiatives/`, or any non-`partnerships/` location is a structural failure, not a stylistic preference — a reader looking for "everything about this partner" should find it under one root, and a future agent loading context should never have to traverse multiple directory trees to assemble the picture.

## The canonical root path

Every artifact produced by this skill lives at `<api-repo>/docs/intent/partnerships/{partner-slug}/`. The `{partner-slug}` is kebab-case, lowercase, derived from the prospect's name or organization, and stable across all sessions for that prospect — so the second meeting's transcript and the first meeting's intent map sit together under the same root, not in two parallel folders.

## The subfolder structure

Inside `partnerships/{partner-slug}/` the canonical subfolders are:

- `meeting transcripts/{date}.md` — source transcripts only, never artifacts
- `initiatives/` — the INITIATIVE-MAP and one file per major holon
- `context/` — grounding context documents that aren't tied to one initiative
- `people/` — private context about named individuals, kept out of the main client-reviewable artifact
- `INDEX.md` at the partner-folder root — navigational entry point with wikilinks to every artifact, partner description, session date

## Canonical naming for this skill's outputs

The files THIS skill produces are exactly:

- `{Partner Name} Meta Intent Map.md` — top-level customer-reviewable map at the partner-folder root
- `initiatives/INITIATIVE-MAP-{date}.md` — index of all holons with the nested intent tree
- `initiatives/INITIATIVE-{nn}-{slug}.md` — one file per major holon, integrated multi-perspectival paragraphs in the user's voice
- `context/{topic}.md` — grounding context the agent needs to interpret correctly but isn't part of the holon tree itself
- `people/{name}.md` — private context about a named individual
- `INDEX.md` — navigational entry point with wikilinks

## What lives outside the partner folder

Shared tooling, infrastructure, live transcript pipes, reusable workflows, and anything that supports multiple partners lives at `1.5_initiatives/{tool-name}/`, and the partner folder cross-links to it with named references. The partner folder NEVER absorbs reusable tooling; the tooling NEVER absorbs partner-specific artifacts. The boundary is enforced both directions.

## What this skill does NOT produce

Proposal drafts, validation logs, open-questions documents, and incomplete-tasks trackers are produced by the next-in-sequence tool that consumes this skill's approved Meta Intent Map. They live under `partnerships/{partner-slug}/working-docs/` but their creation is out of scope for this skill — this skill stops at the approved Meta Intent Map and hands off.

# The Workflow — How An Agent Team Executes This

This skill directs a small agent team where each phase has a Producer that produces output and a Checker that audits that output before the next phase begins. The Checker catches errors at the phase they originated rather than letting them cascade. When a Checker fails the Producer's output, the Producer revises before the pipeline advances.

## Phase 1 — Transcript Ingestion

The Transcript Locator/Cleaner finds the raw transcript in the working directory or related folders, reads it in full, and if a signal-only cleaned version does not already exist, produces one — strips filler words, false starts, off-topic side conversation, audio artifacts, and transcription noise while preserving every substantive utterance. The Fidelity Auditor diffs the cleaned version against the raw transcript and confirms every substantive utterance survived the cleaning. Gate: cleaned transcript exists and substantive completeness confirmed.

## Phase 2 — System Identification + Spanning Intent Derivation

The System Mapper names the system the conversation is about, derives the spanning intent (the root invariant true at every state the system passes through, often not explicitly stated by the client), and anchors the spanning intent in 3-5 quotes that name it directly or implicitly. The Spanning-Intent Verifier independently reads the cleaned transcript and derives what they would call the spanning intent; if divergent from the Producer's, they reconcile. Gate: spanning intent confirmed as invariant across the states the conversation references.

## Phase 3 — Holon Tree Construction

The Holon Tree Builder identifies the distinct initiatives in service to the parent intent — the children of the parent — and the sub-children where the substance warrants. Builds the tree explicitly. The Tree Auditor walks the tree and confirms no initiative is mis-parented, no concept is duplicated under multiple parents, no concept that should be top-level is accidentally nested. Gate: tree clean.

## Phase 4 — Critical Opportunity And Leverage Identification

The Opportunity Surfacer walks the cleaned transcript looking for the critical opportunities the user named or implied — the leverage points, the moments where the user reflected back the highest-impact moves. These opportunities become the load-bearing structure of the artifact: the artifact leads with them. The Leverage Verifier confirms each opportunity is grounded in a quote and is genuinely critical (not merely interesting). Gate: every leading opportunity is leverage-bearing and grounded.

## Phase 5 — Per-Holon Articulation (parallelized)

The Holon Writer (one per major holon) applies /reflect and /align to that holon in the client's context (see the explicit clarification below) and produces the holon's section as integrated multi-perspectival paragraphs in the user's voice (as calibrated by the four examples at the top of this skill). The Voice Auditor confirms the section reads in the user's voice — multiple perspectives held in each sentence, no perspective-per-paragraph fragmentation, headings synthesizing rather than labeling. Gate: voice match confirmed.

## Phase 6 — Cross-Link Pass

The Cross-Linker walks the holon tree and identifies where one holon's content is causal for another holon's gap. Names these cross-links explicitly within the relevant holon's section. The Cross-Link Verifier confirms each named cross-link is actually present in the source material, not invented. Gate: every cross-link grounded.

## Phase 7 — Customer-Lens Pass

The Customer Lens Reader reads the assembled artifact while imagining being the source-conversation participant themselves. Flags anything that contains private context shared off-the-record, anything that narrates about the participant in third person ("she said directly"), anything that promotes a candidate idea to a confirmed scope, anything that would simply feel weird for them to read. The Customer Lens Auditor does the same pass independently. Gate: both readers confirm the artifact is customer-reviewable.

## Phase 8 — Assembly And Final Quality

The Document Assembler assembles all section outputs into the final artifact, confirms heading hierarchy is clean, confirms cross-references between holons use named concepts rather than "see section 3," and confirms the artifact opens with the parent intent and the critical opportunities. The Assembler also produces `INDEX.md` at the partner-folder root — a short navigational entry point that names the partner, names the session date, lists wikilinks to every artifact in the folder (`[[{Partner Name} Meta Intent Map]]`, `[[INITIATIVE-MAP-{date}]]`, each `[[INITIATIVE-{nn}-{slug}]]`, every `[[context/{topic}]]`, every `[[people/{name}]]`), and includes a "why this folder exists" line so a future agent or reader landing in the folder cold can orient instantly. The Final Quality Auditor runs the full quality criteria — signal-to-noise high throughout, a stranger could navigate via headings to any holon and understand its full lifecycle, the voice matches the user's calibration target, and `INDEX.md` actually links to every file present in the folder (no broken wikilinks, no missing entries). Gate: final quality pass.

## Phase 9 — Presentation And Approval

The Presenter shows the artifact to the principal, notes the path, highlights anything the team flagged as uncertain. The Principal reviews and either approves or returns for revision. Gate: explicit approval. Without it, the artifact is a draft. With it, it is canonical and may be referenced by future agents and shown to the source-conversation participant.

## Self-healing on phase failure

When a Checker rejects a Producer's output, the failure gets logged with a brief description of what went wrong. The Producer revises. If the same kind of failure recurs across runs of this skill, the failure mode gets added to the "What Wrong Looks Like" section above — making the skill self-improving.

## Team composition

For transcripts under 30 minutes with 1-3 holons: a single agent plays all roles sequentially. For transcripts of 30-90 minutes with multiple holons: dispatch a small parallel team in Phase 5 — one Holon Writer per major holon. For transcripts over 90 minutes with more than 6 major holons: full team composition with separate Producer and Checker agents per phase.

## Critical: every subagent receives full context

Every subagent receives the full source transcript, the full canonical objective at the top of this skill, and the four voice calibration examples. None receives only the slice of context relevant to its specific role. Token cost of duplicated context is acceptable; alignment cost of fragmented context is not.

## /reflect Inside This Skill — Reflecting On The Client's Context, Not The Principal's

The /reflect and /align skills were originally designed to apply to the principal's context — the agent reflects on what its principal intends. Inside this skill, /reflect and /align are applied to the client's context — the source-conversation participant whose system is being mapped — not the principal's. The interpretation that emerges is grounded in the client's quotes and represents the agent's best understanding of the client's own thinking, not the principal's wishes about the client. The agent does not project the principal's preferences onto the client's reality. If the principal believes a candidate idea would be the right intervention, that belief does not appear in the artifact as if it were the client's intent. Candidate ideas the principal holds about how to engage with the client live in a separate sub-thread and are articulated as candidates, not promoted into the client's nested intent map.

## Sub-skills this orchestrates

When applying this skill, search the skill registry at the start to ensure no relevant lower-level skill is missed. Likely candidates: /reflect (derive nested intent from a source context), /align (current state to target state with gap and path), /speak-human (translate to human-to-human voice), /decompose (convert intent into actionable testable pieces), /declare-scope (make scope visible early), /gap (isolate measurable gaps), /intent-architect (curate human intent into validated seed documents), /agentic-find (search institutional knowledge).

# The User's Full Verbatim Teaching (canonical reference)

This section preserves every distinct piece of feedback from the original teaching session. Future agents loading this skill should read this section to absorb the texture of the principles in the user's own voice.

> "where she is at => where she is going is the main frame"

> "all context for that in /speak-human ( everything in /speak-human ) with direct quotes to substantiate"

> "There's an intent of this entire company right and it's robust. You can see it. It's robust right there's where they're at there's going and there's the intent there's the intent of where they're going and there's the intent of going from where they're at to where they're going and that is nested"

> "you would never combine where they're at about two branches and a one paragraph right you would define the parent and its evolution and where it's going and its pulls and its perspectives and its quotes and its essence and its intent in one section and then the two branches would be subsections of that which would all get the same treatment there is no generalized statement about where everything is going. That's not nested intent"

> "They need to be able to read about one thing and see all of the perspective fantasize about that one thing as well as his parent as well as its parent and so on"

> "I want my agents to be able to stay in crystal clear alignment and what that means is that they can look back and see the intent of each nested layer"

> "the quotes are meant to be like woven into the conversation like everything that you're producing should be in human"

> "in order to make sure that you're not hallucinating you're embedding the quotes, as though you're reminding the person what they said right it's sort of embedded and synthesized into the paragraph"

> "we're trying to create a finished artifact so you don't wanna tell me about why you're doing what you're doing as though you're having a conversation with me all mixed in with the content because then I have to delete. I have to go clean it before I can show it to anybody"

> "this is effectively a map of intent but your conversation with me is completely separate. It's not to be integrated into it"

> "you're not gonna make a statement like this at all right because this could be read by her to see if we understand her so you're not gonna be having a conversation with me about how she said this right that's not that would be weird for her to read"

> "you're gonna include the quotes but you're not gonna be talking to me about your including the quotes in the middle of the text"

> "you're including assumptions that there's gonna be some AI wizard role when that's actually just an idea right which we do wanna articulate but we don't wanna assume it's gonna happen"

> "you're talking about [a team member's private family matter] casually, which is not appropriate for anything that a client is gonna see right I was telling you that so that you could understand what the scope of work was"

> "you've got specific quotes that don't really have meaning... you're going deeply into [named tools] but there's no definition for the meaning of those variables. I'm not really clear what the even value is what this is it seems like you're just using a lot of fancy undefined jargon"

> "you're overreacting. I did not say free from your interpretations. This is an interpretation task"

> "you have long paragraphs in here that's incorrect right I told you very clearly that you should be separating any paragraphs that are long into subsection paragraphs with detailed headings"

> "every concept should have a detailed heading that summarizes the essence of the concept and every set of concepts that are related should have a detailed heading of a different level that summarizes those right this is about nested intent"

> "each paragraph should have a heading over it and the heading should be nested"

> "[the headings] should be nuanced captures of the primary signal, fully understandable by themselves. They need to fully convey the nuanced content in the paragraph that they are governing so that effectively just the title basically says the full story right but you can read the paragraph to get the supportive text"

> "It looks like mostly noise with almost no signal and it's supposed to be almost all signal with almost no noise"

> "This is multidisciplinary perspective. If you read my text how I communicate it's not about separating different perspectives and separate character. Paragraph that's the exact opposite."

> "look up Ken Wilber's integral theory use it to understand how I actually speak from the transcript."

> "Your work should be an extension of how I think, how I speak, how I articulate."

> "The value is at the intersection of the perspectives, in the signal not noise filtered ways they synthesize. separating into separate paragraphs is a perversion of my intent."

> "what we're trying to do is make sure that no critical context gets left out on the way we do that. Make sure that no context gets left out because it's fine if there's too much context truly will pay for the tokens. It's not fine. It's for an agent to go out of alignment because it doesn't understand"

> "this is effectively the massive map of all the critical context and how it was together and it has every critical quote every quote that is even remotely relevant is in there somewhere right but it's organizing in the right place right it's moved around and shuffled around, so it actually reads like somebody is basically showing me my entire business plan where I'm at right now where in a fashion where I can focus on just one thing and I can see in a paragraph that whatever that thing is, it's focused on they deeply and fully understand it, and from that they can create the solution"

> "the customer could review it to realize that we do, in fact, understand their intent we understand the context"

> "it's gonna read a lot like how I write... but you actually look at how I speak in the interviews you'll realize that I'm straight to the point right in the signal of the thing pointing at the connection between things in articulating solutions"

# Final note for any agent loading this skill

This skill exists because requirements kept getting forgotten round-by-round during a teaching session. Externalize the requirements; reference them at every step. The cost of rereading this skill before each application is far lower than the cost of producing artifacts that miss the canonical objective and require correction cycles.