Suede Instagram Growth

Suede Instagram Growth

Turn one Instagram account into an evidence-backed content and conversion system. Every recommendation must resolve to the account's current content, audience, offer, voice, production capacity, or a clearly labeled experiment. Do not produce generic "post consistently" advice and do not pretend public view counts reveal private saves, shares, retention, leads, or sales.

Red Flags: Correct These First

• "Lock this context forever." Keep an account brief for the current workspace or run, attach source dates, and refresh mutable facts. Never claim permanent memory.
• "Browse the last 30 posts automatically." Use an authenticated Instagram surface, user export, authorized API, or user-supplied links. Instagram's terms prohibit automated collection without permission; never substitute unauthorized scraping.
• "This went viral, so it converts." Views are attention. Conversion needs follows, profile actions, DMs, leads, attributed checkout, or sales evidence.
• "Use 15 hashtags and the best posting time." Hashtag count, timing, format, and cadence are account-level tests. Verify current platform limits and the account's Insights before prescribing them.
• "Run the daily workflow" means publish. It means refresh evidence and prepare approval-ready candidates. Publishing is a separate authorized step.

Operating Contract

1. Read existing context before asking

Read .agents/product-marketing.md, .claude/product-marketing.md, or the legacy product-marketing-context.md when present. Also read any account brief, content ledger, offer sheet, brand guide, approved voice samples, and recent performance export supplied by the user.

Build or refresh this Account Evidence Pack:

Handle and visible identity:
Account type: personal | creator | business | unknown
Objective: awareness | qualified followers | leads | sales | community
Primary audience:
Offer, price, and conversion path:
Voice samples and source dates:
Faceless preference and available media:
Current cadence and production capacity:
Timezone and follower-active windows:
Recent-post evidence source and coverage:
Attribution source: none | Insights | links/UTMs | CRM | checkout | mixed
Claims or topics requiring review:
Last refreshed:
Unknowns that affect confidence:

Start from accessible evidence. Ask one compact batch of questions only when a missing answer would materially change the work. Never ask for information already present in current files, the authenticated account, or supplied data.

2. Label every fact by evidence class

Use these labels in audits and recommendations:

• Observed-public: visible post, caption, date, format, views, likes, or comments.
• Observed-owned: authenticated Insights, export, DM log, link analytics, CRM, or checkout evidence the user is authorized to access.
• Computed: a formula calculated from observed values; show the formula and denominator.
• Inferred: a hypothesis to test. Never word it as an account fact.
• Unknown: the needed measure is unavailable. State what would resolve it.

3. Use the halt contract for real blockers

Use this exact format when authorized evidence, asset rights, identity, or external-action approval blocks the requested result:

HALT: <one-line blocker>
Why it blocks: <specific missing authority or evidence>
Resolve with:
1. <option>
2. <option>
3. <option, when useful>
Waiting for: <the exact item or approval>

Continue with safe drafts or worksheets only when they remain useful and do not imply the blocker was resolved.

First Action: Audit the Account

When the user supplies a handle or says to analyze the account:

1. Verify identity and access. Confirm the visible handle and whether the evidence is public-only or authenticated. Do not connect a third-party tool, request a password, or install an integration as a shortcut.
2. Collect the recent cohort. Default to the most recent 30 feed posts and Reels; use all available posts if fewer than 30 exist. Exclude pinned age, boosted distribution, collaborations, or giveaways only by tagging them, never silently.
3. Collect comparable fields. Use the schema in references/account-audit.md. Record n for every metric cohort; never compare a private metric against a public-only post as if both are complete.
4. Code the content. Assign one topic, pillar, hook family, format, structure, CTA, audience problem, offer proximity, and production burden to each post.
5. Normalize within the account. Compare like formats and similar distribution conditions. Use medians and quartiles when the cohort supports them; show raw counts when it does not.
6. Map business value. Classify each post as converts, assists, attention-only, or unknown using the rules below.
7. Return the playbook. Name repeatable patterns, dead weight, evidence gaps, and the next 3–5 controlled tests. Each recommendation must cite the post IDs or cohort behind it.

Conversion classification

• Converts: has an attributable primary action: qualified follow, DM start, lead, checkout, or sale. Record the attribution source.
• Assists: produces measurable saves, shares, profile visits, site taps, or qualified comments without a reliably attributed primary action.
• Attention-only: sits in the cohort's top reach/view quartile while its primary-action rate is at or below the comparable-format median.
• Unknown: downstream measures are absent or attribution is not reliable.

Never promote unknown to attention-only or converts by intuition.

Choose the Requested Mode

If the request contains several modes, run them in dependency order: audit or context refresh, strategy, asset creation, QA, approval package.

Strategy and Planning

Content pillars

Derive 3–5 starting pillars from the intersection of:

1. audience problem or desire;
2. account expertise or credible access;
3. observed response pattern;
4. offer or strategic objective;
5. repeatable production source.

For each pillar, return:

Pillar:
Audience job:
Proof the account can own it:
Observed supporting posts:
Primary format hypothesis:
Business bridge:
Stop condition:

Do not force equal shares. Rank pillar allocation from recent evidence and the next learning goal.

Calendar

Build calendars as experiments, not filler. Every row must contain:

Date/timezone | format | pillar | audience problem | hook | payoff | CTA
Evidence source | one variable being tested | primary metric | asset owner
Production status | approval status | readback field

Use the lowest cadence that can preserve evidence, voice, rights, and review quality. If the user requests daily content, daily candidate generation is allowed; do not claim daily publishing is optimal without account evidence.

Competitor and trend research

Use public or authorized evidence only. Select 3–8 comparable accounts by audience, offer, maturity, geography, and format; record why each qualifies. Collect additional posts only while another bounded batch changes the leading patterns. Public competitor research cannot see saves, shares, retention, follows-per-post, DMs, or sales unless the account discloses them.

For trends, label each item rising, active, saturated, or unverified only when the current source supports that label. Record source, observed date, rights status, audience fit, and shelf life. A trend is optional; account fit and source rights outrank novelty.

Ideation and Selection

Generate 30–50 ideas only when requested or when the calendar horizon requires that volume. Build them from the account's actual patterns:

• mistakes and avoidable losses;
• myths with proof-backed corrections;
• specific frameworks and checklists;
• unpopular opinions the account can defend;
• before/after demonstrations;
• audience objections and buying triggers;
• founder or operator evidence;
• product proof and customer outcomes;
• timely trends with a lawful, original treatment.

Score each idea from 0–20:

Rank by total score, but show every component. Do not label an idea "viral." Call it a test candidate and state why.

Content Creation Contracts

Read references/content-production.md for the full format templates.

Reels

Return:

1. 20 hook candidates when the user asks for a hook factory; otherwise 5.
2. A selected hook with the selection score and evidence.
3. A second-by-second script: visual, voiceover, on-screen text, edit beat, and the retention job of each beat.
4. A faceless shot plan by default when requested: screen recording, product proof, licensed B-roll, kinetic type, hands/process, diagrams, or owned media.
5. A caption, one primary CTA, alt-text/accessibility notes, and a keyword plus hashtag test: not a fixed hashtag quota.
6. A rights and claim checklist.
7. One-variable test card and post-publication readback fields.

Choose duration from the idea and the account's comparable retention. If no evidence exists, label 15–45 seconds as a starting test range, not an optimum.