...
` tags. EB's HTML-to-plain converter eats `` and corrupts the greeting-merged first sentence.
- Greeting merges into the first sentence. No separate "Hi {FIRST_NAME}," line. Write: "Quick note {FIRST_NAME}, ..." or "Saw the {DOWNTOWN_INITIATIVE} news {FIRST_NAME}, ..." — the salutation lives inline.
- Zero em-dashes (`—`) in body copy. Em-dashes are a known EB spam trigger; replace with commas, periods, or hyphens. This is auto-replaced at draft time, not prompted per-occurrence.
- Maximum sequence length is 2 steps (step 1 + step 2 bump). 3+ step sequences are a hard failure. Deeper sequences belong in `campaign-orchestration`'s multi-phase flow.
- No `{FIRST_NAME}` (or any merge variable) in the subject line. Subjects are the highest-impact spam signal; merge personalization in subjects under-performs generic subjects across every deliverability benchmark.
- Subject line length 1-3 words, with 3-option spintax. Example: `{Quick|Fast|30s} {question|check|idea}`.
- Spintax at the word level, not the sentence level: `{option1|option2|option3}`. Apply every 3-5 words where grammar permits — too little and EB sees identical sends; too much and the sentence loses meaning.
- Step 2 subject uses `Re: {subject}` — prefix, lowercase `re:` also acceptable per vendor; the skill defaults to `Re:`.
- Step 2 body references step 1 without summarizing it. One paragraph typical. Reinforces the offer without repeating the pitch.
- Sign-off is ` ` tags, confirm subject has no `{FIRST_NAME}`, confirm step count is exactly 2.
7. Build the `custom_variables` array (every `{VARIABLE}` in body or subject must be declared).
8. `Write` the JSON artifact to `docs/campaigns/{entity}/copy-{campaign-name}-{YYYY-MM-DD}.json`.
9. Report the artifact path + offer summary + tier + preset used to the operator.
**Expected output:** artifact written; one-line summary like "Wrote copy-denver-downtown-2026-04-20.json — T2 free-asset, list-building Municipalities preset, 7 custom variables."
**Error handling:** if §2 hard gate fails at step 3, abort with an operator-facing message naming the missing input. If §6 preset file missing, degrade to Flow 6 without halting.
**Handoff:** operator can then run `/marketing:launch-campaign` (BC-5826) to execute the sequence.
### Flow 2 — Scratch path (no situation artifact, value-equation interview)
**Preconditions:**
- No situation artifact in conversation context.
- Operator has supplied entity + a target description (company name, vertical, rough offer idea).
**Steps:**
1. Ask the operator for entity (if not resolvable from marketing-context.md), vertical (optional), and campaign name.
2. Interview the operator for the 4 §2 value-equation inputs — one question at a time. Do not batch.
3. Recommend an offer tier from the §3 matrix based on the operator's answers + entity. Confirm per D2.
4. Pick base skeleton (A = list-building if T1 / T2, B = risk-reversal if T4; T3 defaults to B with softer guarantee language).
5. Fill slots from operator-supplied inputs. `{RECENCY_ANCHOR}` defaults to the level-6 waterfall fallback ("most {VERTICAL_DESCRIPTOR} teams we work with are scoping ...") since no situation artifact was read.
6. Validate per §8 anti-slop.
7. `Write` artifact with `situation_mining_source: ""` (empty string, per schema).
8. Report path + summary + warning: "No situation artifact used — recency anchor is fallback tier; consider running situation-mining before send."
**Expected output:** artifact written with level-6 fallback hook; operator-facing warning about the missing recency anchor.
**Error handling:** if the operator declines to supply any of the 4 value-equation inputs, abort with a clear message and do NOT write the artifact.
**Handoff:** same as Flow 1.
### Flow 3 — Existing-preset path (operator picks base template manually)
**Preconditions:**
- Operator explicitly wants `list-building` or `risk-reversal` base template with NO vertical override.
- Inputs may be partial — operator has entity + proof point but not a full situation.
**Steps:**
1. Confirm entity + offer tier per D2.
2. Skip the `Glob` preset-file check — operator requested base template explicitly.
3. Load base skeleton (A or B) from §3 directly.
4. Fill slots from operator-supplied inputs; any missing `{VARIABLE}` stays as a placeholder in the body and lands in the `custom_variables` array with empty `default`.
5. Validate per §8.
6. `Write` artifact with `vertical: null` (nullable per schema).
7. Report: "Base `list-building` template, no vertical override — 5 custom variables remain unfilled for operator review."
**Expected output:** artifact with base-template copy; `vertical: null` in schema.
**Error handling:** same as Flow 1.
**Handoff:** same as Flow 1.
### Flow 4 — Seed-vertical demo (Municipalities end-to-end dogfood)
**Preconditions:**
- Operator runs "demo Municipalities list-building" or equivalent dogfood prompt.
- Intended for smoke-testing the lazy-load path after changes to the seed preset or base skeleton.
**Steps:**
1. Load `docs/research/situations/denvergov.org-sample.md` (or any Municipalities-vertical sample artifact) if available; if not, use in-memory demo values for Denver Parks & Rec.
2. Recommend T2 per §3 entity-aware matrix (Labs for Municipalities, T3 also reasonable — operator picks).
3. `Read` `presets/list-building-municipalities.md` — this is the seeded preset from this skill's shipping commit (BC-5825).
4. Fill slots with demo values; confirm value-equation inputs are non-empty.
5. Validate per §8.
6. `Write` artifact to a demo path: `docs/campaigns/brite-labs/copy-demo-municipalities-{YYYY-MM-DD}.json`.
7. Report path + note that this is a dogfood / demo run.
**Expected output:** artifact written; operator can inspect and grep for format compliance.
**Error handling:** if the seed preset file is missing from disk, the skill SHOULD NOT silently fall back — this is a dogfood run specifically intended to exercise the preset load path. Report the missing file as a regression signal.
**Handoff:** none — this is a smoke test.
### Flow 5 — Thin-context fallback (no marketing-context.md — hard gate pause)
**Preconditions:**
- `docs/marketing-context.md` is missing OR does not identify the Brite entity.
**Steps:**
1. Warn the operator with the BC-5824 precedent message: "Marketing context doc not found — proceeding with reduced context. Run `/marketing:product-marketing-context` to generate it."
2. PAUSE. Ask the operator for the Brite entity explicitly via AskUserQuestion. Do NOT proceed to drafting with a silent default (D1).
3. If the operator supplies entity, continue into Flow 1 / 2 / 3 with that entity as the authoritative source.
4. If the operator declines to supply entity, ABORT — do not write any artifact. Report: "Entity is required; cannot generate copy without it."
**Expected output:** no artifact written until operator confirms entity.
**Error handling:** none — this is a precondition-violation path, not a failure path.
**Handoff:** skill resumes Flow 1 / 2 / 3 after the operator answers the entity prompt.
### Flow 6 — Unknown-vertical fallback (preset file missing)
**Preconditions:**
- Operator supplies a `vertical` value.
- No matching `presets/{template_preset}-{vertical}.md` file exists. This is expected behavior for every vertical EXCEPT `municipalities` until BC-5879 / BC-5880 / BC-5881 ship the fan-out preset files.
**Steps:**
1. `Glob` `presets/{template_preset}-{vertical}.md` returns no matches.
2. Log a one-line warning to the operator naming the relevant fan-out issue. Example for HOAs: "Preset `list-building-hoas.md` not found — falling back to base skeleton A + Nites tone. The HOAs preset is tracked in BC-5879 (Active-tier fan-out)." For Exploring-tier verticals reference BC-5880; for Future-tier verticals reference BC-5881.
3. Load base skeleton (A or B) from §3.
4. Pull entity tone from `docs/marketing-context.md` — specifically the tone markers in `§Voice` or equivalent.
5. Fill slots; `{RECENCY_ANCHOR}` uses the situation artifact's top waterfall signal if present, else level-6 fallback.
6. Validate per §8.
7. `Write` artifact with `vertical: null` (the skill did not use a vertical preset — the operator's intent is recorded but the output is base-template tone).
8. Report the fallback + the fan-out issue reference.
**Expected output:** artifact written with base-template + entity tone; `vertical: null` in schema; operator-facing log line naming BC-5879 / BC-5880 / BC-5881 as applicable.
**Error handling:** if the operator's supplied vertical is NOT in the handbook 23-vertical taxonomy (e.g. "installers", "property management"), pause per §4 architectural rules and clarify — do not produce a Supply-tone email. This is a §8 hard failure path.
**Handoff:** same as Flow 1.
---
## Health Scoring Rubric
| Score | Criteria |
|------:|----------|
| 10 | Artifact is EB-format-compliant (no `{{`, no ` `, no em-dashes in body, no `{FIRST_NAME}` in subject, exactly 2 steps, sign-off spintax correct); Hormozi value equation visible in body copy (all 4 inputs map to specific paragraphs); entity-aware tier from the §3 matrix is picked and confirmed by operator; recency-waterfall anchor appears in the hook and cites the waterfall level; every `{VARIABLE}` in body + subject is declared in `custom_variables`; base or per-vertical preset is cited in the artifact; `situation_mining_source` cited when applicable; all §8 anti-slop guardrails pass validation. |
| 7-9 | Mostly excellent with one gap — e.g. recency anchor cites a level but the signal is weak; one custom variable in the body is missing from the array; spintax is word-level but less dense than the 3-5 word guideline; sign-off is correct but not spintax-expanded. |
| 4-6 | Functional but missing structural elements — e.g. value equation applied but proof point is generic (no numbers), or step 2 bump summarizes step 1 instead of referencing it, or tier was recommended but not confirmed with operator (D2 bypassed), or `vertical: null` without a valid fallback reason, or the preset file exists but wasn't read. |
| 1-3 | Format violations (any one: `{{` double-brace present, ` ` tag present, em-dash in body, `{FIRST_NAME}` in subject, >2 steps, Supply-vertical trigger like "installers" or "property mgmt" in body); OR `docs/marketing-context.md` was ignored (silent entity default); OR fact-claim framing ("you are X" instead of "we noticed X and thought Y"); OR fabricated proof point / statistic / case study not in marketing-context.md or operator input. |
---
## Anti-Slop Guardrails
Base guardrails (shared across marketing plugin) + skill-specific hard failures. Skill-specific rules are phrased as "Do not X" because they are enforced as validation gates, not style preferences.
**Base guardrails:**
- Do not generate generic marketing jargon ("synergy", "leverage", "best-in-class", "game-changing", "cutting-edge").
- Do not fabricate statistics, case studies, testimonials, or proof points — always attribute to a source in `docs/marketing-context.md` or operator input. If a proof point is missing, ABORT per §2 value-equation gate.
- Do not produce output that ignores `docs/marketing-context.md`. Entity, tone, ICP, and sender info all derive from marketing-context before any operator input.
- Do not recommend tools the plugin does not have access to (no hallucinated MCP servers, no EB tool calls in this skill — this skill generates copy only).
**Skill-specific hard failures (validation-gated — fail the artifact emit):**
- **Do not emit `{{variable}}` double-brace tokens.** Email Bison requires single-brace uppercase variables (`{FIRST_NAME}`). If a draft contains `{{`, self-correct to single-brace before emit. Hard failure if present in the written artifact.
- **Do not emit ` ` or ` ` and corrupts the greeting-merged first sentence. Use `
{Best|Cheers|Thanks},
{SENDER_FIRST_NAME}` — spintax on the sign-off, no em-dash before the name.
- `{SENDER_*}` variables (`{SENDER_FIRST_NAME}`, `{SENDER_EMAIL}`, `{SENDER_ROLE}`) are filled from `docs/marketing-context.md` first; Salesforce `User` object only if the marketing-context.md value is missing (see §5 Workflow 1).
### Hormozi value equation
Every body paragraph in tier 2 + tier 3 + tier 4 emails applies the Hormozi value equation. Formula:
```
Value = (Dream Outcome × Perceived Likelihood of Achievement) / (Time Delay × Effort + Sacrifice)
```
Operator-facing: to make the email feel valuable, maximize the numerator (dream outcome + proof) and minimize the denominator (time-to-value + effort). The four inputs confirmed in §2 correspond one-to-one:
| Input (from §2 gate) | Where it appears in the body | Brite example (Nites T2 free design preview) |
|---|---|---|
| Dream Outcome | Paragraph 1 / hook — names the outcome by the prospect's own language | "holiday install complete by Black Friday with zero coordination overhead" |
| Perceived Likelihood of Achievement | Paragraph 2 — proof point with real case-study numbers | "Sugarloaf HOA ran 27 units in 2024 with the same architect-approved spec" |
| Time Delay | Paragraph 2 or 3 — compresses the time-to-value | "design preview in 48 hours, install before Thanksgiving" |
| Effort + Sacrifice | Paragraph 3 / CTA — the guarantee or free asset that shrinks perceived effort | "free architectural preview for {COMMUNITY_NAME} — review before any commitment" |
Tier 1 (knowledge / helpful resource) skips the proof-point paragraph since the offer IS the proof. Tier 4 (risk reversal) makes the Effort + Sacrifice input the headline rather than the CTA.
Full framework reference: `plugins/marketing/references/offer-design-frameworks.md` — Hormozi value equation origin + Brunson Value Ladder + Abraham strategic layer (Brite-originated synthesis).
### Offer tiers + entity-aware selection matrix
Four tiers, adapted from Revgrowth 10. Each tier maps to a different CTA architecture, proof-point posture, and value-equation emphasis.
**Tier definitions:**
- **T1 — Knowledge / Helpful Resource.** CTA = "here's a resource, no reply needed." Lowest friction. Use when signal density is LOW and the operator wants a warming touch before a harder ask.
- **T2 — Free Asset.** CTA = "we'll prepare a specific asset for your context, no commitment." Example: free downtown lighting audit, free design preview, free deliverability audit. Most common outbound default for Nites.
- **T3 — DFY Trial / Pilot.** CTA = "we'll run a small paid pilot; success pays for itself." Example: 3-home pilot install, single-night event lighting pilot. Use when prospect signal is HIGH and procurement signal is strong.
- **T4 — Risk Reversal / Guarantee.** CTA = "first phase on us if it doesn't hit {measurable outcome} by {date}." Example: performance guarantee on festival install. Use for large-spend / committee-heavy procurement where the denominator (Effort + Sacrifice) is the prospect's biggest blocker.
**Entity-aware selection matrix (3 rows, one per Brite entity):**
| Entity | Typical signal density | Recommended tier | Tone marker | Example vertical anchor |
|---|---|---|---|---|
| Nites | MEDIUM-HIGH for seasonal signals (new board, new management, calendar RFP) | T2 (free asset) default; T3 when HIGH signal density | Seasonal, residential, warm-neighborhood | HOAs, Landscape Architects, Builders, Universities (Nites-side seasonal overlays) |
| Supply | HIGH signal density required (commercial procurement cycle is long) | T3 (pilot) default; T4 when enterprise committee visible | Commercial, spec-driven, procurement-aware | *(Supply verticals are out of scope per handbook — see §4 architectural rules)* |
| Labs | MEDIUM for capital-project signals (bond, master plan, capital campaign) | T3 (pilot) default; T4 when multi-year spend / committee-heavy | Experiential, capital, design-production | Municipalities, Universities (capital), Theme Parks, Botanical Gardens |
The skill RECOMMENDS the tier from this matrix then confirms with the operator per D2.
Per-vertical offer guidance: `plugins/marketing/references/vertical-playbooks/{vertical}.md` (produced by Phase 2 roadmap issues R-4 through R-9 — e.g. `zoos.md`, `hotels-resorts.md`, `ski-resorts.md`, `sports-stadiums.md`, `aquariums.md`, `casinos.md`).
### Recency waterfall (6-level hierarchy)
The hook in paragraph 1 anchors to the most recent credible signal. Walk the waterfall top-to-bottom and use the highest-level signal available:
1. **New job / role change** — "saw you just stepped into the Director of Campus Experience role at {COMPANY}."
2. **LinkedIn post (prospect's own, within 90 days)** — "your post last week on {TOPIC} stuck with me."
3. **Company news / press release (within 90 days)** — "with the {INITIATIVE} announcement last month."
4. **CEO / leader podcast or interview (within 180 days)** — "heard {CEO_NAME} on {PODCAST} talking about {THEME}."
5. **Company blog post (any time)** — "the {BLOG_TITLE} post on your site made me think about {ANGLE}."
6. **Fallback (no recency signal)** — vertical-anchored trigger. "most {VERTICAL} teams we work with are scoping {SEASONAL_INITIATIVE} right now — thought {COMPANY} might be too."
Only level 6 fires when the situation artifact's §Raw Data section yielded <2 recency-grade public signals. In that case, flag the email as LOW-confidence in the artifact, inherit situation-mining's thin-data framing, and recommend operator review before send.
### Base template skeletons (2, entity-agnostic, inline)
Two base skeletons live inline per D3. Per-vertical overrides lazy-load from `presets/{preset}-{vertical}.md` (see next subsection).
#### Skeleton A — list-building base
Used for T1 / T2 framing. Diagnostic hook + proof point + low-commitment free-asset CTA. Greeting-merged first sentence, `
` paragraph breaks, word-level spintax.
```
Subject: {Quick|Fast|30s} {question|check|idea}
Body:
Saw the {RECENCY_ANCHOR} at {COMPANY} {FIRST_NAME}, and {it lined up|it tracked closely|it mapped well} with a pattern we've been watching across {VERTICAL_DESCRIPTOR}.
{Most|A few|Several} {VERTICAL_DESCRIPTOR} teams we work with run into {SPECIFIC_FRICTION}, and one that {solved|shortcut|sidestepped} it was {PROOF_POINT_COMPANY}, who {PROOF_POINT_NUMBER} in {PROOF_POINT_TIMEFRAME}.
{Happy|Glad} to {pull|share|send} a {short|quick|focused} {FREE_ASSET_NOUN} for {COMPANY} if {useful|helpful|interesting}, no commitment.
{Best|Cheers|Thanks},
{SENDER_FIRST_NAME}
```
**Step 2 bump:**
```
Subject: Re: {subject}
Body:
{Circling back|Following up|Bumping this} in case it {got buried|slipped past|fell off}. {Still happy|Glad still} to send the {FREE_ASSET_NOUN} whenever it's {useful|helpful|timely}.
{Best|Cheers|Thanks},
{SENDER_FIRST_NAME}
```
#### Skeleton B — risk-reversal base
Used for T4 framing. Heavier commitment context, guarantee as the headline, pilot CTA. Same format rules as skeleton A.
```
Subject: {Guarantee|Pilot|On us}
Body:
With the {RECENCY_ANCHOR} at {COMPANY} {FIRST_NAME}, the {scope|spend|commitment} {feels serious|deserves care|reads as high-stakes}, and that's the kind of project we {take on|scope|pilot} with a {measurable|specific|concrete} guarantee.
For {COMPANY}'s {INITIATIVE_NOUN}, we can {run|deliver|execute} the first {PHASE_NOUN} with {GUARANTEE_TERMS}, so the risk sits with us and the {proof|outcome|signal} sits with you.
{Worth a 20-minute scope|Open to a quick scope call|Happy to scope a pilot}?
{Best|Cheers|Thanks},
{SENDER_FIRST_NAME}
```
**Step 2 bump:**
```
Subject: Re: {subject}
Body:
{Know|Aware|Understand} this kind of {commitment|pilot|scope} takes {a beat|time|real review}. The guarantee terms are {flexible on|negotiable around} {TIMELINE} or {DELIVERABLE_SCOPE}, so if either needs to shift, {happy to|glad to} adjust.
{Best|Cheers|Thanks},
{SENDER_FIRST_NAME}
```
### Lazy-load per-vertical overrides
When the operator supplies a `vertical` value matching a handbook-canonical slug (see `Brite-Nites/handbook@main:marketing/go-to-market/verticals/README.md`), the skill reads ONE preset file per invocation at `plugins/marketing/skills/email-copywriting/presets/{preset}-{vertical}.md`. Per D3 this bounds runtime context cost — only the single matching file loads, not the 46-file library.
**Preset file shape** (frontmatter + 4 sections, ~40-60 lines each — canonical shape in `presets/README.md`):
```yaml
---
preset: list-building | risk-reversal
vertical:
`), `wait_in_days` (integer, 0 for step 1, typically 3-5 for step 2).
- `situation_mining_source` — path to the input artifact when this campaign flowed from `situation-mining`. Omitted / empty when §6 Flow 2 (scratch path) ran.
- `generated_at` — ISO-8601 timestamp. `/marketing:launch-campaign` checks this against a staleness threshold before launching.
**Save path convention** — `docs/campaigns/{entity}/copy-{campaign-name}-{YYYY-MM-DD}.json`. Operator supplies `{campaign-name}`; the skill slugifies it (lowercase, hyphen-separated). The `{entity}` subdir mirrors the entity slug (`brite-nites` / `brite-labs`).
### Architectural rules that apply
- **`docs/marketing-context.md` is the entity-canon source.** Never hard-code an entity default in this skill. If marketing-context is missing and the operator declines to answer, ABORT — do not guess (D1).
- **Offer tier is always recommend + confirm.** No auto-select code path. Even with HIGH signal density, surface the recommendation to the operator and wait for confirmation before drafting (D2).
- **Preset files are lazy-loaded.** One preset file read per invocation, not the whole library. Use `Glob` + `Grep` to check existence before `Read`; on missing, fall back to base inline skeleton without halting (D3).
- **Supply vertical triggers are out of scope.** The handbook 23-vertical taxonomy excludes professional installers + property management (see `Brite-Nites/handbook@main:marketing/go-to-market/verticals/README.md`). If an operator supplies a Supply-framed prospect, pause and clarify — do not produce a Supply-tone email. Inherited from BC-5824 precedent.
- **Hypothesis framing is non-negotiable.** Inherited from situation-mining §3 — body copy never states worldview as fact. When incorporating inferred signals from the situation artifact, the copy must read as "we noticed X and thought {HYPOTHESIS}" — never "you are X."
---
## MCP Tool Reference
"When you need to X, call `tool_name`." Grouped by workflow. This skill has exactly one MCP workflow — a conditional Salesforce lookup for sender defaults. All other work is local `Read` / `Write` / `Glob` / `Grep`.
### Workflow 1 — Sender-info lookup (conditional)
Runs **only** when `docs/marketing-context.md` is missing the `{SENDER_*}` defaults AND the operator has not supplied them explicitly. See [`plugins/marketing/tools/integrations/salesforce.md`](../../../tools/integrations/salesforce.md) for auth, tool names, and SOQL gotchas.
1. **Availability check:** call `run_soql_query` with `SELECT Id FROM User LIMIT 1`. On failure, HALT this workflow — do NOT fabricate sender info. Fall back to asking the operator directly for `{SENDER_FIRST_NAME}`, `{SENDER_EMAIL}`, `{SENDER_ROLE}`. Per BC-5534 findings §Q1, this query is the verified liveness check; `get_username` is NOT a valid liveness check.
2. **User lookup:** call `run_soql_query` with `SELECT Id, FirstName, Email, Title FROM User WHERE Email = '{operator_email}' LIMIT 1`. Use the returned `FirstName` / `Email` / `Title` to fill `{SENDER_FIRST_NAME}`, `{SENDER_EMAIL}`, `{SENDER_ROLE}` in the custom_variables array.
3. **On zero results:** the operator's email doesn't match a Salesforce User — prompt the operator for the sender values directly via AskUserQuestion. Do not halt.
All SF calls are read-only; no MCP confirmation gates apply. This skill has NO mutating workflows — it never touches EB state, never writes to Salesforce, never modifies anything outside the local `docs/campaigns/` directory.
---
## Operational Runbook
Six flows — the common paths operators actually run. Each flow states preconditions, steps (referencing §5 Workflow 1 where applicable), expected output, error handling, and cross-skill handoff.
### Flow 1 — Happy path (situation artifact + offer tier + vertical → copy using preset)
**Preconditions:**
- `docs/marketing-context.md` exists and identifies the Brite entity.
- A situation-mining artifact is in conversation context (path supplied by operator or auto-detected from recent `docs/research/situations/*.md`).
- Operator has identified the `vertical` in the situation artifact (or the skill extracts it from the artifact frontmatter).
**Steps:**
1. Read the situation artifact; extract `entity`, `vertical`, worldview row, adjacent offering.
2. Apply §3 entity-aware tier matrix to recommend an offer tier. Surface the recommendation + one-line rationale to the operator. Wait for confirmation (D2 confirm gate).
3. Run §2 value-equation gate — confirm the 4 inputs resolve from marketing-context.md + situation artifact. If any missing, interview the operator one input at a time.
4. `Glob` check `presets/{template_preset}-{vertical}.md`. If it exists, `Read` it. If not, flag the fallback path (Flow 6).
5. Fill slots: `{RECENCY_ANCHOR}` from the situation artifact's top waterfall signal, `{PROOF_POINT_*}` from marketing-context.md case studies, `{SENDER_*}` from marketing-context.md or §5 Workflow 1 fallback, `{FREE_ASSET_NOUN}` / `{INITIATIVE_NOUN}` / `{GUARANTEE_TERMS}` from value-equation inputs.
6. Validate draft against §8 anti-slop guardrails — auto-replace em-dashes, check for `{{` double-brace tokens, check for `
` for paragraph breaks. Hard failure if present.
- **Do not emit em-dashes (`—`) in body copy.** Known EB spam trigger. Auto-replace with commas, periods, or hyphens at draft time. Hard failure if present in the written artifact. (Em-dashes in this SKILL.md spec prose are OK where explaining the rule; the rule applies only to body template examples and generated artifact text.)
- **Do not emit sequences with more than 2 steps.** v0.1 of this skill supports step 1 + step 2 only. Longer sequences belong in `campaign-orchestration`'s multi-phase flow. Hard failure on 3+ steps.
- **Do not emit `{FIRST_NAME}` (or any merge variable) in the subject line.** Subjects are the highest-impact spam signal; merge variables under-perform generic subjects across every deliverability benchmark. Hard failure.
- **Do not frame inferences as facts.** Inherited from situation-mining §3 — body copy MUST read as hypothesis when referencing a prospect worldview or inferred signal. Write "we noticed the {INITIATIVE} announcement and thought it might line up with how your team is scoping {VERTICAL_DESCRIPTOR}" — never "your team is scoping {VERTICAL_DESCRIPTOR}." Fact-claim framing is a hard failure; §7 1-3 band.
- **Do not emit Supply-vertical triggers in body copy.** The handbook 23-vertical taxonomy excludes professional installers + property management. Body copy that keys to installer hiring, PM company onboarding, or other Supply signals is a hard failure per handbook canon + BC-5824 precedent. If an operator supplies a Supply-framed prospect, pause and clarify per §6 Flow 6 error handling.
---
## Behavioral Tests
Eight scenarios covering the core paths. Structured assertions + expected-output details live in `evals/evals.json` alongside this file. Scenario IDs match the `evals.json` entries for 1:1 traceability.
### Tier 1 — Free assertions (no tool calls needed)
- **`scratch-path-value-equation`** — Given no situation artifact and an entity-only input, the skill's first response asks for the 4 §2 value-equation inputs (Dream Outcome, Perceived Likelihood, Time Delay, Effort + Sacrifice) before any drafting. No JSON artifact is emitted until the operator answers.
- **`format-violation-self-correct`** — Given a draft (operator-supplied template or mid-generation text) containing `{{firstname}}`, the skill self-corrects to `{FIRST_NAME}` before artifact emit. Output artifact body contains zero `{{` tokens.
- **`em-dash-auto-replace`** — Given operator-supplied proof-point text containing em-dashes, the skill auto-replaces all em-dashes with commas, periods, or hyphens before artifact emit. Output artifact body contains zero `—` characters.
- **`unknown-vertical-fallback`** — Given `vertical: hoas` (preset file not yet shipped — expected pre-BC-5879), the skill falls back to base skeleton A + Nites tone and logs a one-line warning citing BC-5879. Output artifact has `vertical: null` and body matches base-template shape, not a vertical-override shape.
- **`missing-offer-tier-gate`** — Given no `offer_tier` input, the skill's first response recommends a tier from the §3 matrix with one-line rationale AND asks for operator confirmation via AskUserQuestion. No JSON artifact is emitted until the operator confirms or overrides.
### Tier 2 — Tool-assisted (requires file read or MCP call)
- **`happy-path-municipalities-seed`** — Given a situation artifact for Denver Parks & Rec at `docs/research/situations/denvergov.org-2026-04-20.md` + `vertical: municipalities` + `offer_tier: 2` + entity confirmed, the output JSON artifact (a) exists at the expected path, (b) has `template_preset == "list-building"`, (c) has `vertical == "municipalities"`, (d) has `situation_mining_source` populated, (e) body contains `
` paragraph breaks, (f) body contains zero `—` characters, (g) subject contains zero `{FIRST_NAME}` tokens. Preset file `list-building-municipalities.md` was `Read` during the flow.
- **`entity-switching`** — Given the same situation artifact, run once with `entity: brite-nites` and once with `entity: brite-labs`. The two artifacts differ in (a) `offer_tier` (Nites → 2 typical, Labs → 3 or 4 typical per §3 matrix), (b) subject line word choices (Nites warmer / seasonal, Labs more capital / experiential), (c) CTA framing (Nites "free preview" vs Labs "scope a pilot"). Entity tone is sourced from `docs/marketing-context.md`.
- **`missing-marketing-context-hard-gate`** — With `docs/marketing-context.md` absent from disk, the skill's first response does NOT contain any JSON artifact text and does NOT contain any "## Recommendations" section. It DOES contain the BC-5824 precedent warning message AND an entity prompt via AskUserQuestion. No Write tool call fires until the operator answers. (D1 hard gate.)
---