Research posts and creators, or turn your goals into a social calendar and drafts.
Install gooseworks: https://gooseworks.ai/install.md # then, in Claude Code, Cursor, or Codex: /gooseworks use the run-organic-social skill
A post list or focused research request: after shared setup, use marketing-data-access (../marketing-data-access/SKILL.md) and its focused-reads.md reference for requested post rows, dates, engagement and bounded rankings. For interpretation or voices, use the existing platform-research step, restricted to that question. Deliver the requested list/analysis; skip the full-plan opening, calendar, account-strategy intake and eight-example-per-platform gate. A top-voices list states its ranking evidence and coverage, rather than claiming an exhaustive directory.
Say once, at the start, in these words or close: "I'll look at how each platform works right now and what's getting watched in your space, then give you a plan and two weeks of posts. I can't post for you yet; you'll post by hand."
A report the marketer opens with one click (delivered through the write-report skill (../write-report/SKILL.md)):
Every post is checked for claims, voice and disclosure. Posts that speak for a named person are flagged for that person's yes.
For brand voice, read an existing company guide at brand/voices/blog-<brand-slug>.md alongside the account’s card and brand rules. Reuse transferable tone and bans; keep account-specific social patterns and never turn blog article length into a post limit. Preserve a guide’s provisional status.
marketing-guide skill (../marketing-guide/SKILL.md) §1–2 (connect, brand, brain read) before anything else; the job line is "I plan and write your social through it." Then, for this playbook:
get_started shows a card, write one line and carry on with this ask; don't call card_wait (the card has no organic-social option). Even when it named the brand, call gooseworks:brand_read with no brand_id for its coworker_agent_id.required_reads, then chapters 11 how we sound, 14 rules, 15 how we work together (who approves a person's posts), 17 accounts and tools, 19 this week and what's coming, 20 our plan, 23 what is working; chapter 12 (look and files) only for designed slots. gooseworks:brand_knowledge_search for gaps: disclosure, off-limits topics, buyer language. Then gooseworks:brand_read lean with only sections: ["accounts","sources","organic"] (the brain doesn't hold own posts or saved sources).gooseworks:brand_read lean with sections: ["summary","kit","learnings","accounts","sources","organic","products"], then product_ids for the products in play; gooseworks:knowledge_search for operating rules, disclosure and off-limits topics.detail: "full": on a settled brand it returns 100k+ characters no host can read; the lean read carries every rule and claim.run-organic-social-platform-research skill (../run-organic-social-platform-research/SKILL.md) §3b on that account only and deliver it as a report through write-report: why it works, what to adapt, and five tests for this brand. Then offer the full plan in one line.
One post, not a plan? If the ask is a single post or script, use the brand read and the voice card only. Write that one post to the platform's research card if a fresh one exists (social/research/, under 30 days). Otherwise do one quick web check of how that platform treats that format now. Run the run-organic-social-post-check skill (../run-organic-social-post-check/SKILL.md), then offer the full plan in one line.run-memory skill (../run-memory/SKILL.md) with the inputs in references/run-record.md. It finds the last social calendar's run record, checks last time's recommendations and format tests against what went out and how it did, and writes the "since last time" lines that open the report. No record means a first run: say so, invent nothing, and freeze the baselines from this run's audit. A record exists and a calendar is running → this is a refill: go to step 7. "Start from scratch" (or "ignore earlier work") makes this a first run: don't read or carry the last record, earlier calendars or notes, and say so in one line ("Starting fresh: I won't use earlier plans or notes."); the brand's rules and approved claims still apply, and what the marketer says in this conversation holds.1. Free pass (ask nothing). Keep it proportionate: research the platforms the marketer named, or the top 1–2 recommended. Roughly 10–15 web searches and 15–20 page reads per main platform is enough; stop a line of search when it repeats what you have. Write the report once, then revise it in place. Hard gate: no draft report without a dated "how it works now" card and at least 8 linked examples of what's working in the niche, for each platform the marketer named (or the top recommended one). Other recommended platforms get a lighter read: the card plus 3 examples, deepened once the marketer picks them. If the host has no web search or browser, say so at the top of the report and in chat, and plan nothing from memory.
Pick the platforms in scope first: the ones the marketer named. Otherwise recommend 2–3 from where the audience is and what the brand can make, and say why.
Platform research (the run-organic-social-platform-research skill, ../run-organic-social-platform-research/SKILL.md), for each platform in scope:
Saved files, all in the coworker scope:
social/ (the last calendar, social/accounts.md, social/research/). A saved research card under 30 days old is reused only if it still meets this step's bar (8+ examples with numbers, at least 3 from accounts under 10k followers, a platform or sized-study source). Otherwise top it up. Files from other people's tests or drafts are not history.brand/voices/content/repurposed/, content/channels.md, content/pov/reports/performance/<date>.mdMine the raw material that's already public:
Real shipped things, numbers and screenshots found here are what the first posts are built from.
Own-account audit (the run-organic-social-account-audit skill, ../run-organic-social-account-audit/SKILL.md) and voice cards (the voice-card skill (../voice-card/SKILL.md), social additions in references/voice-card.md).
2. The draft report and one message.
run-organic-social-post-check skill (../run-organic-social-post-check/SKILL.md) (step 5). A post that needs the person shows the finished post plus one "needs from you" line, never a skeleton of brackets.spend-gate skill (../spend-gate/SKILL.md) branch B), with what it adds;claims-gate skill, ../claims-gate/SKILL.md). Name each line exactly. "Check the brand kit" is not a yes to a line;3. Deepen (only what the marketer said yes to).
4. Write the posts (references/slot-anatomy.md; plan with the run-organic-social-plan-calendar skill (../run-organic-social-plan-calendar/SKILL.md) and references/calendar-craft.md).
5. Check and remake (the run-organic-social-post-check skill, ../run-organic-social-post-check/SKILL.md): voice match, the "would a person post this?" test, AI-tells, claims through the claims gate, disclosure and rights, the person gate, trend expiry. A weak draft is remade or the slot is dropped. Never pad.
gooseworks:social_post_create render: false).render: true. Never social_post_approve. Never organic_creative_queue. Posting stays parked. A hand-built carousel draft from already-rendered slides is allowed on a managed account only after a yes (references/graphics.md §6).6. Deliver the final report through the write-report skill (../write-report/SKILL.md), shaped by references/report-shape.md:
save-to-brain skill (../save-to-brain/SKILL.md), with a gooseworks:file_url link.Also save:
social/calendar/<start-date>.md: the grid, the posts, the posting sheet and the tests;social/accounts.md;social/research/;file_url download link (same scope, expires_in_seconds: 3600).Before you deliver, run write-report §3's checks, and also scan for [ brackets. Name posts by day ("Monday's post"); the one allowed fill-in reads "fill in: <what>".
In chat, at most 3 lines (on a refill, the one "since last time" result that matters most; the first move; the link) plus the decisions left, at most 5 short items.
Save the run record (full plans and refills): the End steps of the run-memory skill (../run-memory/SKILL.md), shaped by references/run-record.md. It holds the per-post and per-format scorecard, each format test with its verdict or carried forward, and every recommendation with a review-by date. Add "Saved this run's notes to your brand, so the next refill starts from here." to the link line, not as a line of its own. In a scheduled run, say nothing about it unless the save failed.
Revisions ("rewrite them in my voice"): edit the report file and the workspace copy in place. Then reply in at most 3 lines: what changed, the link, and anything still open. Show a post in chat only when the marketer asks to see that post. Never re-paste the set.
7. Refill (every two weeks, or on request). A refill is not a first plan: skip step 1's full free pass and step 2's draft. In order:
run-organic-social-platform-research skill (../run-organic-social-platform-research/SKILL.md) §3c). Convergence, spikes, early topics and controversies feed the flex slots.8. Offer the repeat (judge first). At the end of a full plan, judge whether a repeat would help this brand. It usually does when the marketer means to keep posting.
spend-gate skill (../spend-gate/SKILL.md) branch C). Switch it on only on the marketer's yes, and name any per-run data ceiling in that yes; without one, scheduled runs use free sources only.social_strategy_upsert.social/plans/<YYYY-MM-DD>-social-plan.md; and save-to-brain (versioned: keep-history), titled "Social plan — <Brand> — <YYYY-MM-DD>".social/research/<date>-<platform>.mdsocial/calendar/<start-date>.md (+ -sheet.csv)social/accounts.mdsocial/audit/<date>.mdcontent/pov/ (person answers)research/credits-log/<YYYY-MM>.md (spend-gate)runs/run-organic-social/social-calendar/<YYYY-MM-DD>.md: the run record, through run-memory (also saved to the brand, keep-history)save-learning skill (../save-learning/SKILL.md), with the marketer's own words, prefix [social YYYY-MM-DD].social_strategy_upsert after a yes; the posted status as a record via social_post_patch patch.status.The final report is delivered and saved with a link, and:
Nothing was posted, and nothing was invented.
Explain what worked, what is uncertain and what to try next.
Audit email flows and draft messages for subscribers. You handle sending.
Audit search visibility, inspect available SEO metrics, find gaps and plan topic clusters or programmatic pages.
Test one thing at a time, start small, and keep a steady flow of new creative. A practical plan for a month of testing, with the Goose settings that support it.
One install prompt gives you a Grok bot that researches your brand, makes ads and product photos, writes social posts and finds influencers.