Social Media Manager

Marketplace listing update: added the launch explainer film (`media.demo_video_url`). No code, schema, copy or guidance changes; hero and screenshots unchanged from 0.2.1.

**Propagation patch. Card + copy only — no change to any reference, recipe, learning, registry
item or script.** The bundle's behaviour at 0.2.1 is identical to 0.2.0.

### Changed — marketplace card media (SPEC-3)

- `media.hero_image_url` + all three `screenshots` replaced with a v0.2.x asset kit that actually
  shows what this version does. The previous set was rendered for 0.1.5 and illustrated the loop,
  not the channel decision. New frames: the channel pick, the three axes, the acquisition-vs-home
  surface split, and the compounding-vs-decaying contrast. Dark-board chalk, reference-anchored
  (Mode B) against the house style, every frame viewed before shipping.

### Fixed — two inaccurate claims on the live card

- **"Composes 11 shipped Di-Atomic skills"** was wrong. There are 11 peer dependencies, but only
  six are `@di-atomic/*`; the other five are the `@opvs-ai/*` coordination layer. The feature that
  carried this claim is replaced by one describing channel selection, and `long_description` now
  states the composition accurately.
- **"12 ready playbooks"** was still live on the marketplace card despite being corrected in the
  0.2.0 draft. It never reached a published manifest because 0.2.0 shipped before the fix was
  staged. This release carries it: 13.

### Note

`^0.2` picks this up; `^0.1` does not, and never did — see the caret-range trap. Consumers pinned
to a `0.1.x` caret still need a deliberate re-pin.

**Wave-4 enrichment from the book corpus. `change_class: additive` — no removals, no restructure,
no rewrites.** The skill was strong operationally and thin strategically: 11 references of which
one was strategic, exactly one platform playbook, no channel-selection guidance at all, and no way
to check where any claim came from. This wave adds the strategy layer.

### Added — references (11 → 18)

- `channel-strategy-selection.md` — which platforms this brand should be on, and why. Three axes
  (searcher vs scroller · compounding vs decaying asset · native to the creator), the depth-vs-width
  decision with its gate, and warm-before-cold.
- `platform-framework.md` — how to build a playbook for **any** platform, including ones the corpus
  never covered. Goal/strategy/tactics, the six steps, the acquisition-vs-home meta-pattern, and the
  land-grab timing rule.
- `dream-100-targeting.md` — finding where the audience already gathers: the congregation inventory,
  the submarket rule, honest yield numbers, the outreach sequence, and the social-proof cascade.
- `platform-playbook-instagram.md`, `platform-playbook-facebook.md`, `platform-playbook-youtube.md`
  — at parity with the existing X playbook. Strategic layer only; perishable tactics are quarantined
  in a dated section.
- `sources-and-further-reading.md` — citation notation, **evidence classes**, named further reading,
  and an explicit statement of what the corpus does *not* cover.

### Added — recipes (12 → 13)

- `channel-selection.md` — the channel decision as a runnable workflow. Runs **before** a calendar
  exists; gates fan-out on an established primary channel.

### Added — learnings (21 → 25)

- `evidence-class-practitioner-claims` — why nothing extracted from a practitioner book may enter
  the benchmark class, with the corpus's own self-contradiction as the worked example.
- `platform-reward-decay-and-the-land-grab` — new surfaces are over-rewarded during adoption, then
  throttled. Test immediately; never assume today's free reach is permanent.
- `fanout-distributes-it-does-not-grow` — fan-out is a distribution strategy, not a growth strategy.
- `owned-list-is-the-ignition` — owned ignition → organic lift → paid amplification, in that order.

### Added — scripts (5 → 6)

- `verify-citations.mjs` — every `cite:` resolves to a real page span whose distinctive terms
  actually appear there. Runs format+range validation standalone, and full claim/span verification
  against a corpus. **159/159 citations pass.**

### Appended to existing references (instruction-deltas — nothing rewritten)

- `content-review-gate.md` — the boost test, the ABCDQ brand-fit check, and the mechanism that
  justifies the gate: engagement is account-level, so a weak post taxes everything else you publish.
- `content-pillars-framework.md` — the JK5 rotation's two placement rules (no two same-category
  posts adjacent; people follow for one pillar and buy later).
- `posting-cadence-by-platform.md` — the give-to-ask ratio with its anchors, the reason 3.5:1–4:1
  is a monetisation *floor* rather than a growth target, integrated vs intermittent asks, and a
  stop condition.

### Changed

- `SKILL.md` — routes to the new references and the new recipe (without them the additions are
  unreachable), two new hard rules (channel strategy precedes calendar work; nothing from the
  corpus is a benchmark), and refreshed layer counts. **No existing rule or branch altered.**

### Unchanged

- The 61 registry exemplars, the 12 pre-existing recipes, the 5 pre-existing scripts, and every
  existing branch of the SKILL.md decision tree. **All 5 pre-existing scripts pass unchanged.**

### Notes

- Sourced from *Traffic Secrets* (Brunson, 2020) and *$100M Leads* (Hormozi, 2023). No source text
  ships in the bundle — restatement plus citation only.
- ⚠ Per [[feedback-caret-0x-range-trap]], a caret pin never picks up a 0.x minor: consumers pinning
  `^0.1` must be re-pinned deliberately to reach 0.2.x.
- LinkedIn and TikTok playbooks are **deliberately not included** — the corpus has no material on
  either and says so explicitly. Use `platform-framework.md`; properly researched playbooks are
  separate work.

All notable changes to this skill are documented here. Format: [Keep a Changelog](https://keepachangelog.com/). Semver.

Deeper AgentBoard symbiosis — the skill now uses the **full live 34-field `content_calendar` schema** and provisions its own board. Additive; no breaking change.

### Added
- **`recipes/provision-content-calendar.md`** (recipe #12) — the first-run workflow: resolve the brand's `content_calendar` board, and if none exists, **create one** (`agentboard.createBoard(board_type:"content_calendar")` auto-seeds the 6 columns + the 34-field Blotato schema — never hand-rolled), resolve the Blotato `social_account` map via `vibe-publish.listAccounts`, then hand to weekly planning. Added to the SKILL.md router (🏗 "do this FIRST for a new brand").
- **Board-resolution step in Auto-start** — every run now resolves-or-provisions the board before routing. The board IS the plan; there is no planning without it.

### Changed
- **SKILL.md now carries the full 34-field schema**, not the stale 6. Hard Rule 1 + the `/#agentboard` coordination block now name **`social_account`** (required — a brand has several Blotato accounts) and **`media_urls`** (a persisted `media.spideriq.ai` URL) plus the hidden per-platform publish group (`media_type · yt_* · pinterest_* · tiktok_* · fb_* · ig_* · li_page_id · threads_reply_control`). `createTask` ignores custom_fields → set via `updateTask(custom_fields_merge:true)`.
- **`weekly-calendar-planning`** now resolves `social_account` per card and defers `media_urls` to Create (persisted URL only).

### Fixed
- **BRIEF peer-chain drift** — listed the old `@spideriq/publish-skills`; the social publisher is `@opvs-ai/vibe-publish` (already correct in the manifest). BRIEF also gained the v2.5 `coordination_plan` (state/long_form/learning/handoff) + narrowed the `opvs-protocol` cross-ref (peer skills are composed same-agent, not messaged).

Marketplace-card metadata only — no change to the skill's guidance, recipes, or behavior.

### Added
- **Marketplace asset kit (SPEC-3).** Top-level `media:` block with a hero image + 3 screenshots (dark-board chalk: #131313 board / white chalk / gold #EEBF01), produced via media-generator → SpiderGate `kie/nano-banana-2` and SpiderMedia-persisted. Concepts: the closed social-ops loop, the specialist-skill pipeline with its human checkpoint, and a scattered-tools vs one-conductor before/after.
- **`marketing:` block** — first-person `long_description` + 5 verb-led feature bullets (the loop · human review gate · sharper every cycle · 11 skills one conductor · 11 playbooks).

### Note
- Shipped as part of the di-atomic.com propagation pass (blog + changelog + listing page + this card).

Wave-3 enrichment — additive, no scope change. Grounded from a deployed agent's own social workflow (Lola @ VayaPin, `SOCIAL_MEDIA_WORKFLOW.md`, 2026-07-14).

### Added
- **`recipes/social-ops-loop.md`** (recipe #11) — the canonical end-to-end orchestration order every other recipe is a slice of: `social-media-manager → content_calendar card(s) → voice-builder → copy-engine → hook-generator → post-writer → media-generator → Review gate → @opvs-ai/vibe-publish → agentmemory learnings`. Added to the SKILL.md router (🧭) and README recipe count (10 → 11).

### Changed
- **SKILL.md — Publish invariant block appended** (after the 8 hard rules, decision tree untouched): (1) nothing publishes from Review without a **human** approval — `verify-review-gate-presence.mjs` only checks the gate *exists*; (2) **never put an expiring raw provider/temp URL (`kie.*`, `tempfile.*`) on a Review or publish card** — the SpiderGate `generate-media` path already returns a **stored** `media.spideriq.ai` URL (`stored=true`), so use that directly; only persist when all you hold is a raw provider URL; (3) publish only via `@opvs-ai/vibe-publish` with a `mediaUrls` array (reinforces Rule 2). Persist nuance grounded in the 2026-07-14 Lola harvest G2.

### Note
- **Ref2 (content_calendar card anatomy)** and the **v2.2 Coordination cross-refs** from the Wave-3 directive were already fully present in 0.1.2 (`references/agentboard-content-calendar-anatomy.md` + the SKILL.md Coordination block) — verified, no duplicate authored.
- Deferred (out of this pass, per directive): Lola BC1/BC2 → VayaPin workspace; cross-vendor G4/G5/D1/D2 → separate @spideriq / @opvs-ai asks.

### Added
- **`@opvs-ai/vibe-publish` is now the named social publisher** (`createPost` → Blotato; brand key auto-resolved from OPVS Integrations — no raw Blotato key). Added as a `peer_dependency` (>=2.0.1). Replaces the "Vibe/Blotato" hand-wave across SKILL.md, DEPLOYMENT.md, and the anatomy/scheduling references.
- **Card → `createPost` field map** in the anatomy reference + DEPLOYMENT. Full Blotato coverage: threads (`additionalPosts[]`), carousels (2–10 `mediaUrls`), reels/stories (`target.mediaType`), per-platform target extras.

### Changed
- **Documented the Blotato-ready content_calendar board schema** OPVS shipped: `social_account` (Blotato accountId — required to pick 1 of a brand's several accounts), `media_urls` (public CDN URLs), `bluesky` added to the `platform` enum, and the hidden per-platform "publish" group (`media_type`, `yt_*`, `pinterest_*`, `tiktok_*`, `fb_*`, `ig_*`, `li_page_id`, `threads_reply_control`).
- Hard Rule 2 reframed: publish + schedule via `vibe-publish` `createPost` (`scheduledTime`/`useNextFreeSlot`); SpiderFlow is now optional.

### Fixed
- **`content.mediaUrls` must always be sent** (`[]` for text-only) — omitting it returns an opaque Blotato 400. Now a hard rule; root-caused live 2026-07-12 (agent omission, not the runtime). vibe-publish 2.0.1 also injects the default.
- Documented that `createTask` ignores `custom_fields` — set them in a second `updateTask`/`bulkUpdateTasks` call.

### Note
- Marketplace 0.1.1 (published 2026-06-22) predates the schema-fix work in [0.1.1] below; **0.1.2 is the first release to carry both the content_calendar schema fix and the vibe-publish integration.**

### Fixed
- **content_calendar schema alignment.** v0.1.0 referenced `custom_fields` + column names that don't exist in the OPVS `content_calendar` board schema (`board_type_schemas.py`) and would be silently stripped at runtime (Pydantic). Corrected across SKILL.md, references/agentboard-content-calendar-anatomy.md, references/spiderflow-scheduling-catalog.md, instructions/DEPLOYMENT.md, and recipes (weekly-calendar-planning, cross-platform-fanout, crisis-response-pause, content-review-gate):
  - `status` → `publish_status` (draft|scheduled|published|failed|cancelled).
  - Dropped `pillar`, `language`, `variant_of`, `asset_url` from custom_fields → pillar/language now ride as **labels**, the repurpose source as a `spawned` task link, asset URLs as attachments.
  - Added the real schema fields `post_type`, `hashtags`, and `source_vibe_post_id` (the hidden link from a card to its published Vibe/Blotato post).
  - Column lifecycle corrected to the real **Ideas → Draft → Review → Scheduled → Published → Archived**; the **Review** column is the content-review gate (was a mislabeled "review stage").
  - Publish callback flips the card to **Published** (`publish_status: published`), not a non-existent `posted`.
  - crisis-response-pause: no `paused` status exists — pull cards Scheduled → Draft or set `publish_status: cancelled`.
- Corrected the `content-calendar-custom-fields-silently-stripped` learning, which had itself named the wrong persisting fields.

### Added
- Initial release. Guidance-only social-media orchestrator ("the social-media department in a skill").
- **SKILL.md** router (≤500 lines, 8 hard rules, 4 v2.2 coordination cross-references, 10-recipe intent table).
- **references/** (11): content-calendar anatomy · spiderflow scheduling · scrapecreators trend/reply/competition matrix · content-pillars framework · posting-cadence-by-platform · reply-mining playbook · social-health-metrics · content-review-gate · compliance-social-vocabulary · multilingual-cadence-shifts · platform-playbook-x (+ reddit/bluesky/linkedin/instagram).
- **recipes/** (10): weekly-calendar-planning (flagship) · trend-driven-post · reply-mining-engagement · competition-monitor · cross-platform-fanout · social-health-review · content-review-gate · compliance-social-cadence · multilingual-rollout · crisis-response-pause.
- **registry/** (55): 40 view-ranked SMM teardowns (live, metric-bearing) + 5 platform playbooks + 6 content-pillar frameworks + 5 cadence/benchmark tables + 4 SMM-process blueprints.
- **learnings/** (20): 15 patterns/gotchas/anti-patterns + 5 per-script gold-nuggets research entries.
- **scripts/** (5): verify-cadence-vs-platform · lint-vanity-metrics · verify-review-gate-presence · verify-cross-platform-fanout · verify-coordination.
- **instructions/**: AGENTS.md (non-Claude runtime rules) + DEPLOYMENT.md (handoff payload shapes + downstream consumers).
- Deep research backing: 60 curated registry seeds, 142 scrapecreators exemplars captured, 4 MIT upstreams consumed, plus a dedicated X-focus "posts that convert" deep dive.

### Scope
- **EN-only** in v0.1.0. DE/RU/FR/RO/HE/ES planned for v0.2.0.
- Compliance-aware (REACH/CLP/biocides) via post-writer's compliance gate.
- Distinct from `@di-atomic/community` (members/partnerships/champions).

### Peer chain
Composes 4 OPVS coordination skills + @spideriq/publish-skills + 6 @di-atomic skills (post-writer, hook-generator, voice-builder, distribution-launcher, media-generator, copy-engine).