Audience Classifier — Who Is This For?
A problem statement that doesn't fit its audience has failed regardless of its other qualities. Statement-grader's audience-fit axis depends on the audience being correctly identified upstream — that's this skill's job. The audience tag drives the entire compression subdomain's routing.
For the canonical mapping of audience → compression skill, see [[minto-scqa]] and the compression skills directly.
The Five Audience Tags
| Tag |
Who |
Channel typical |
Time budget |
Routes to |
| LLM |
A language model that will execute the resulting prompt |
API call, in-context |
Model-consumed (~10K tokens fine) |
cursed-speech |
| Exec |
Decision-maker, time-pressured, trust intact |
Slack, email, brief, deck |
30 sec – 5 min |
bluf-shaper (short) or executive-distiller (memo) |
| Peer |
Collaborator with parallel context |
Doc, email, meeting |
2–15 min |
scqa-formatter |
| Self |
The user thinking through their own problem |
Note, scratch buffer |
Whatever's needed |
scqa-formatter (skip Q step) or Polya restatement |
| Public |
Mixed-audience reader who may be skeptical |
Essay, post, talk, public memo |
5 min – 30 min |
executive-distiller (with Concern-priming) or scqa-formatter (Concern-priming variant) |
Detection Signals
Read the statement plus any context the user provided (channel, recipient, purpose). Apply these:
| Signal |
Likely audience |
| Output is described as "a prompt", "ask Claude", "send to GPT/Claude" |
LLM |
| Recipient is named as a title role (CEO, CMO, board, VP) |
Exec |
| Phrase like "memo for the team", "share with engineering", "draft for the working group" |
Peer |
| Phrase like "I'm thinking through", "trying to figure out", "for my own clarity" (no specific recipient) |
Self |
| Phrase like "publish", "post", "essay", "talk", "newsletter", "public statement" |
Public |
| Recipient is a colleague at peer level with similar context |
Peer |
| Reader is described as time-pressured or "needs the answer fast" |
Exec |
| Reader is described as skeptical, opposed, or requiring persuasion |
Public (Concern-priming variant) |
| Output is described as "for me", "for myself", "to remember" |
Self |
If multiple audiences are plausible (e.g., "I want to think this through AND share with the team"), default to the more constrained audience — sharing imposes more structure than self-thinking, so route Peer.
Sub-Context Tags
Beyond the primary audience tag, surface sub-context that downstream compression skills can use:
| Sub-context |
Values |
Used by |
| Time budget |
≤30s / 30s–5min / 5–30min / open |
bluf-shaper vs executive-distiller choice |
| Skepticism level |
trusting / neutral / skeptical |
SCQA variant selection (standard vs Concern-priming) |
| Channel |
slack / email / doc / deck / verbal / API |
Length and formatting bounds |
| Familiarity |
high context / partial context / cold |
Background-section depth |
These don't change the primary tag; they tune the compression skill's behavior.
Process
- Scan for explicit signals in the statement and any context (channel, recipient, purpose).
- Apply the detection signal table — count matches per audience tag.
- Resolve conflicts. If multiple tags get matches, apply the "more constrained wins" rule. If genuinely tied, ask the user (this is the right time for L4 user-assist).
- Tag sub-context — time budget, skepticism, channel, familiarity.
- Return primary tag + sub-context.
Output Format
AUDIENCE — [first 60 chars of statement...]
─────────────────────────────────────────────
Primary tag: [LLM | exec | peer | self | public]
Confidence: [high | medium | low]
Sub-context:
Time budget: [≤30s | 30s–5min | 5–30min | open]
Skepticism: [trusting | neutral | skeptical]
Channel: [slack | email | doc | deck | verbal | API]
Familiarity: [high | partial | cold]
Routes to: [recommended compression skill, given the primary tag + sub-context]
Rationale: [which signals matched]
Edge Cases
| Pattern |
Handling |
| Statement specifies two audiences explicitly (e.g., "for the CEO and the LLM that will analyze it") |
Run compression twice. Tag both audiences; orchestrator's deep mode will produce both shapes. |
| Statement has no explicit audience signals |
Default to self (the user is thinking, no recipient specified) with low confidence. Recommend the user specify if a different audience is intended. |
| Audience is a class of person, not an individual ("for engineering managers") |
Treat as peer (collective collaborator). Skepticism level depends on the topic. |
| Audience is an LLM but the user wrote it as if for a human |
Tag LLM but flag the prose-shaped framing — cursed-speech will need to restructure for the canonical order. |
| Audience is "everyone" or "the world" |
Tag public; flag low confidence; recommend narrowing scope. Public-as-mass-audience usually conceals a more specific intended reader. |
Output Contract for six-eyes
Called from Phase 2 (Diagnose), in parallel with problem-typology and stakes-assessor. Returns:
- Primary tag
- Sub-context tags
- Recommended compression skill (for Phase 5 routing)
six-eyes Phase 5 uses the primary tag to pick the compression skill; the sub-context tunes its behavior. If primary tag is LLM, also signal that cursed-speech will need the audited statement plus the downstream task description.
Connections
bluf-shaper, scqa-formatter, executive-distiller, cursed-speech (binding-vow) — downstream consumers of the audience tagproblem-typology (binding-vow) — runs in parallel; some typologies (adaptive, wicked) suggest specific audience patternsstakes-assessor (binding-vow) — runs in parallel; high stakes often imply exec audience or publicminto-scqa (binding-vow) — canonical audience-fit decision rules for SCQA variant selectionstatement-grader (binding-vow) — uses the audience tag to score the audience-fit axis
Sources
- Minto, B. (1987, 2003). The Pyramid Principle. (Audience-driven structure decisions.)
- See [[minto-scqa]] for the canonical audience → structure mapping.
1---2name: audience-classifier3description: Classify the intended audience of a problem statement as LLM / exec / peer / self / public. Use during binding-vow's Phase 2 diagnosis. The audience tag drives the entire compression subdomain routing — bluf-shaper for exec, scqa-formatter for peer, cursed-speech for LLM, executive-distiller for public. Returns the audience tag plus inferred sub-context such as time budget and skepticism level.4---56# Audience Classifier — Who Is This For?78A problem statement that doesn't fit its audience has failed regardless of its other qualities. Statement-grader's audience-fit axis depends on the audience being correctly identified upstream — that's this skill's job. The audience tag drives the entire compression subdomain's routing.910For the canonical mapping of audience → compression skill, see [[minto-scqa]] and the compression skills directly.1112## The Five Audience Tags1314| Tag | Who | Channel typical | Time budget | Routes to |15|---|---|---|---|---|16| **LLM** | A language model that will execute the resulting prompt | API call, in-context | Model-consumed (~10K tokens fine) | `cursed-speech` |17| **Exec** | Decision-maker, time-pressured, trust intact | Slack, email, brief, deck | 30 sec – 5 min | `bluf-shaper` (short) or `executive-distiller` (memo) |18| **Peer** | Collaborator with parallel context | Doc, email, meeting | 2–15 min | `scqa-formatter` |19| **Self** | The user thinking through their own problem | Note, scratch buffer | Whatever's needed | `scqa-formatter` (skip Q step) or Polya restatement |20| **Public** | Mixed-audience reader who may be skeptical | Essay, post, talk, public memo | 5 min – 30 min | `executive-distiller` (with Concern-priming) or `scqa-formatter` (Concern-priming variant) |2122## Detection Signals2324Read the statement plus any context the user provided (channel, recipient, purpose). Apply these:2526| Signal | Likely audience |27|---|---|28| Output is described as "a prompt", "ask Claude", "send to GPT/Claude" | LLM |29| Recipient is named as a title role (CEO, CMO, board, VP) | Exec |30| Phrase like "memo for the team", "share with engineering", "draft for the working group" | Peer |31| Phrase like "I'm thinking through", "trying to figure out", "for my own clarity" (no specific recipient) | Self |32| Phrase like "publish", "post", "essay", "talk", "newsletter", "public statement" | Public |33| Recipient is a colleague at peer level with similar context | Peer |34| Reader is described as time-pressured or "needs the answer fast" | Exec |35| Reader is described as skeptical, opposed, or requiring persuasion | Public (Concern-priming variant) |36| Output is described as "for me", "for myself", "to remember" | Self |3738If multiple audiences are plausible (e.g., "I want to think this through AND share with the team"), default to the *more constrained* audience — sharing imposes more structure than self-thinking, so route Peer.3940## Sub-Context Tags4142Beyond the primary audience tag, surface sub-context that downstream compression skills can use:4344| Sub-context | Values | Used by |45|---|---|---|46| **Time budget** | ≤30s / 30s–5min / 5–30min / open | `bluf-shaper` vs `executive-distiller` choice |47| **Skepticism level** | trusting / neutral / skeptical | SCQA variant selection (standard vs Concern-priming) |48| **Channel** | slack / email / doc / deck / verbal / API | Length and formatting bounds |49| **Familiarity** | high context / partial context / cold | Background-section depth |5051These don't change the primary tag; they tune the compression skill's behavior.5253## Process54551. **Scan for explicit signals** in the statement and any context (channel, recipient, purpose).562. **Apply the detection signal table** — count matches per audience tag.573. **Resolve conflicts.** If multiple tags get matches, apply the "more constrained wins" rule. If genuinely tied, ask the user (this is the right time for L4 user-assist).584. **Tag sub-context** — time budget, skepticism, channel, familiarity.595. **Return** primary tag + sub-context.6061## Output Format6263```64AUDIENCE — [first 60 chars of statement...]65─────────────────────────────────────────────66Primary tag: [LLM | exec | peer | self | public]67Confidence: [high | medium | low]68Sub-context:69 Time budget: [≤30s | 30s–5min | 5–30min | open]70 Skepticism: [trusting | neutral | skeptical]71 Channel: [slack | email | doc | deck | verbal | API]72 Familiarity: [high | partial | cold]73Routes to: [recommended compression skill, given the primary tag + sub-context]74Rationale: [which signals matched]75```7677## Edge Cases7879| Pattern | Handling |80|---|---|81| Statement specifies two audiences explicitly (e.g., "for the CEO and the LLM that will analyze it") | Run compression twice. Tag both audiences; orchestrator's deep mode will produce both shapes. |82| Statement has no explicit audience signals | Default to `self` (the user is thinking, no recipient specified) with low confidence. Recommend the user specify if a different audience is intended. |83| Audience is a class of person, not an individual ("for engineering managers") | Treat as `peer` (collective collaborator). Skepticism level depends on the topic. |84| Audience is an LLM but the user wrote it as if for a human | Tag `LLM` but flag the prose-shaped framing — `cursed-speech` will need to restructure for the canonical order. |85| Audience is "everyone" or "the world" | Tag `public`; flag low confidence; recommend narrowing scope. Public-as-mass-audience usually conceals a more specific intended reader. |8687## Output Contract for `six-eyes`8889Called from Phase 2 (Diagnose), in parallel with `problem-typology` and `stakes-assessor`. Returns:90- Primary tag91- Sub-context tags92- Recommended compression skill (for Phase 5 routing)9394`six-eyes` Phase 5 uses the primary tag to pick the compression skill; the sub-context tunes its behavior. If primary tag is `LLM`, also signal that `cursed-speech` will need the audited statement plus the downstream task description.9596## Connections9798- `bluf-shaper`, `scqa-formatter`, `executive-distiller`, `cursed-speech` (binding-vow) — downstream consumers of the audience tag99- `problem-typology` (binding-vow) — runs in parallel; some typologies (adaptive, wicked) suggest specific audience patterns100- `stakes-assessor` (binding-vow) — runs in parallel; high stakes often imply exec audience or public101- `minto-scqa` (binding-vow) — canonical audience-fit decision rules for SCQA variant selection102- `statement-grader` (binding-vow) — uses the audience tag to score the audience-fit axis103104## Sources105106- Minto, B. (1987, 2003). *The Pyramid Principle*. (Audience-driven structure decisions.)107- See [[minto-scqa]] for the canonical audience → structure mapping.
npbuilds