Use when the user asks to "plan an influencer campaign", "build a campaign blueprint", "track or close a creator campaign", or "record a late campaign correction"; produces the plan and, when requested, a non-canonical evidence tracker with scoped identity, publication, reconciliation, close, and reopen receipts. Not for individual creator briefs — use brief-generator; not for overall product launches without creators — use launch-tier-planner; not for sending, publishing, amplifying, or paying — use the owning execution workflow. 达人营销策划/种草方案/活动追踪与关账
Install with Codex or Claude Copy this prompt, paste it into Codex, Claude, or another assistant, and let it review the skill page and install it for you.
A direct command skips the review prompt. Inspect the source before running it.
The command stays on one line. Scroll horizontally to inspect it before copying.
Prefer a local copy? Download the files currently available to SkillsMP.
File Explorer
3 files
Showing SKILL.md
SKILL.md
Source instructions · Read-only preview
name
campaign-planner
slug
aaron-campaign-planner
displayName
Campaign Planner · 活动规划
summary
红人活动整体规划:目标、阶段、创作者组合、时间线与风险预案
description
Use when the user asks to "plan an influencer campaign", "build a campaign blueprint", "track or close a creator campaign", or "record a late campaign correction"; produces the plan and, when requested, a non-canonical evidence tracker with scoped identity, publication, reconciliation, close, and reopen receipts. Not for individual creator briefs — use brief-generator; not for overall product launches without creators — use launch-tier-planner; not for sending, publishing, amplifying, or paying — use the owning execution workflow. 达人营销策划/种草方案/活动追踪与关账
Use when planning a new influencer campaign, launching a product with influencer support, building seasonal or always-on creator programs, or maintaining an existing creator-campaign tracker: recording verified publication checkpoints, deriving exception queues, reconciling payment/measurement evidence, closing creators or the campaign, and handling late evidence or manual reopen. Planning does not write briefs or execute outreach; tracking does not publish, amplify, pay, or mutate external systems.
argument-hint
<brand or product> [budget] [platform] [timeframe]
Designs an influencer campaign from strategy to execution plan — an actionable blueprint that ties business objectives to creative execution.
Scope edge — product launches: this skill owns the creator lane of a launch. The launch itself — tier/type decision, launch calendar, press motion, community launch day, readiness gate — belongs to the launch discipline (launch-tier-planner and siblings), which hands this skill the creator-channel sub-plan aligned to the launch-registry date and stage. "Launch a product with creators" starts here; "launch a product" starts there.
Quick Start
Create an influencer campaign plan for [product launch]
Plan an influencer campaign for [brand] with [budget] targeting [audience] during [timeframe]
Skill Contract
Reads: requested mode (plan-only | tracker-only | both); brand, product, audience, campaign type, budget, timeline, and constraints for plan authoring; and, for execution-ready tracking, an existing campaign_id, exact versioned campaign-plan reference/hash, locked §8 measurement contract, and locked non-empty creator scope. If memory-management is active, prior audience profiles and past-campaign benchmarks load from the hot cache.
Writes: the campaign plan inline by default. With exact save authorization, write it to memory/influencer/campaign-planner/YYYY-MM-DD-<topic>.md. For execution tracking, emit separate JSON artifacts that validate against the shared five-kind control schema (evidence observations, the locked measurement contract, action intents/receipts only for an actual executor action, and the Cycle Retro), then let the controller bind their exact refs/hashes to the selected run ancestry. The Markdown/YAML tracker in references/templates.md §10 is a deterministic read-only Influencer compatibility view, marked authoritative: false; its domain blocks are not themselves schema-valid control artifacts, and editing them never changes runtime state. Standalone hosts without the validator may save a semantic-only compatibility snapshot after exact WARM authorization, but it must be marked NOT_VERIFIED and cannot claim single-head, receipt, persistence, or close enforcement. Reuse an explicit upstream/user ; in a plan-authoring mode, generate one random if none exists and preserve it through the lineage. never invents a missing ID, plan, contract, scope, or checkpoint. Every row preserves a stable opaque ; every saved is qualified-resolver-backed and opaque. Raw names, handles, URLs, shortcodes, provider IDs, and hidden locator maps remain transient.
campaign_id
campaign-<UUIDv4>
tracker-only
creator_ref
live_post_ref
Promotes: only with separate exact authorization, promote approved campaign name, objective, budget, go-live date, and KPI targets to memory/hot-cache.md; never promote tracker stages or payment state. After a creator row is closed, another exact authorization is required for each registry proposal containing evidence-backed actual rate, signed rights window/expiry, or measured performance baseline; only creator-registry can make those facts canonical. Forecast targets, stage, next_action, due_at, and payment_status remain WARM working state.
Done when:
The selected mode is explicit: plan-only completes §§1–9, tracker-only completes only §10 from its required existing inputs, and both completes §§1–9 before §10.
In a plan-authoring mode, objectives, strategy, influencer mix, deliverables, timeline, budget allocation, contingency, and KPIs are concrete only where the user supplied them or a compatible source-dated planning anchor supports them. Every unsupported required choice stays NEEDS_INPUT; the skill never fills it from a repository default. The §8 measurement design is execution-locked only when its campaign/plan binding, immutable plan hash/version, authorization, non-empty creator scope, and unique per-creator/deliverable checkpoints are present; otherwise report the exact lock inputs as NEEDS_INPUT/DONE_WITH_CONCERNS and do not create a close-eligible tracker.
If tracking was requested, each creator has validated source artifacts and one deterministic non-authoritative projection block with stage, next action, due date, rights expiry, evidence references, and external-payment handoff state. The projection carries its source artifact refs/digests and current head; it is never hand-edited as the state source.
Each tracker creator uses one stable opaque creator_ref; every identity resolution, publication, terminal-checkpoint resolution, creator close, campaign close, migration, and late event has an immutable ref plus exact campaign/creator scope and a single non-forked current head where applicable.
Every saved live_post_ref is opaque and qualified-resolver-backed. External actions require current exact authority and matching intent/receipt artifacts; a prior save, plan, gate, path, capability, or projection never grants authority. Raw post locators and reusable blanket approvals never enter the artifact set or projection.
A tracked live post has a receipt per required checkpoint, and each checkpoint's latest explicitly unsuperseded receipt controls the close gate; a mismatch, unknown disclosure, or changed/removed post is evidence for review, never silently treated as approved.
A closed creator row points to its unique current close-receipt head. A closed campaign points to the unique current campaign-close head, proves exact equality with the locked non-empty creator scope, and passes the strict gates in template §10; neither closed value is treated as a synonym for success.
A material late event is appended without rewriting prior receipts. A reference correction appends fresh close receipts when corrected gates still pass; a failed gate cannot receive a passing receipt, and manual reopen is reserved for new campaign-owned work under template §10's stage/action baseline.
A plan-authoring result names brief generation and open approvals. A tracker-only result instead names the evidence/action owner implied by the current stage, or stops as chain-complete when nothing remains; it never routes backward to brief generation by default.
Primary next skill: in plan-only/both, brief-generator; in tracker-only, use the stage-specific handoff in Next Best Skill or stop when complete.
This family is Tier 1: every skill works with no live integrations. Supplied brand, audience, budget, and timeline support a plan skeleton, but do not determine messaging, platform/tier mix, content format, promo mechanics, contingency, rate, or KPI targets. Those choices require user direction or compatible source-dated planning evidence; otherwise retain NEEDS_INPUT.
Optional connectors that strengthen the plan when available:
~~influencer database — size the influencer mix and validate tier follower ranges.
~~social platform analytics — set platform-specific reach and engagement benchmarks.
~~CRM — align conversion targets and attribution with existing pipeline data.
~~analytics — pull past-campaign actuals for realistic KPI and budget-efficiency targets.
See CONNECTORS.md for the free/keyless data recipe per category. Without a connector, ask the user for the missing inputs and proceed only with fields supported by user-provided evidence; return a useful skeleton plus NEEDS_INPUT for the rest.
Instructions
Choose and state one mode before doing work. plan-only runs §§1–9 and does not instantiate §10. tracker-only runs only §10 and requires an existing campaign_id, exact versioned plan_ref plus plan hash, the current locked §8 measurement contract, and its locked non-empty creator scope; if any binding is missing, mismatched, forked, or unverified, return NEEDS_INPUT instead of rebuilding a plan or creating a placeholder tracker. both runs §§1–9 before §10. If creator selection is not yet locked, return the plan plus the exact missing scope/contract inputs and, only when explicitly requested, an inline non-close-eligible partial tracker header; do not invent creator identities or call it execution-ready. In plan-only or both, reuse an explicit upstream/user campaign_id; for a new campaign with none, generate random campaign-<UUIDv4> once and preserve it unchanged. In tracker-only, a missing ID is blocking; never generate or deterministically derive it from a campaign name, date, creator handle, or mutable plan content.
For plan-authoring modes, work the nine steps in order. Each has a fill-in template in references/templates.md — copy the matching block and assemble it in step 9. Replace brackets only with supplied or source-backed values; never invent a number, identity, benchmark, or fact to eliminate a placeholder. Keep an unresolved optional field Unknown, and return NEEDS_INPUT when a required field is absent.
Gather campaign requirements — capture campaign_id, brand, value prop, audience, campaign type, timeline, budget, and constraints (template §1).
Define objectives — one SMART primary objective plus secondary objectives, with explicit success and failure definitions (template §2).
Develop strategy — big idea, strategy statement, audience, key messages, campaign pillars, platform split, and differentiation only from user-approved canon or compatible source-dated evidence; otherwise preserve the decision fields as NEEDS_INPUT rather than defaulting to UGC, promo-code, or another tactic (template §3).
Define influencer criteria — tier mix, must-have and preferred selection criteria, exclusions, ideal profile, and relationship types. Use only a user-declared or compatible source-dated follower taxonomy; references/influencer-tiers.md supplies the recording contract, not universal ranges or performance claims (template §4).
Plan content requirements — deliverables by platform/format, required elements, creative direction, themes, and the approval chain from supplied decisions and current platform/rights evidence; do not invent a default UGC format, promo mechanic, or content mix (template §5).
Create the timeline — key dates, a four-phase week-by-week plan, and a Gantt view (template §6).
Allocate budget — break down the supplied total by category, declared follower band, and platform only from user-approved rules or compatible source-dated cost evidence. Add contingency, CPM/CPE, or cost-per-content targets only when their rule/anchor is supplied; otherwise use NEEDS_INPUT and do not invent a percentage or rate (template §7).
Establish success metrics — primary KPIs vs source-dated comparators, secondary and conversion metrics, reporting cadence, and a non-canonical pre-execution measurement contract covering the baseline, outcome unit, readback window, attribution basis, decision rule, decision owner, exact campaign/plan version/hash, lock authorization, non-empty creator scope, and structured unique per-creator/deliverable publication checkpoints (template §8). Every external comparator requires an opaque source ref, observation date, and comparable window/cohort; otherwise keep it Unknown/NEEDS_INPUT rather than presenting an “industry average.” A scope or contract change is a new immutable version plus explicit §10 migration; never edit the locked block in place.
Compile the plan document — executive summary, the full sections above, and an appendix with risk mitigation (template §9). Return it inline by default; save only after authorization for the exact WARM path.
For tracker-only or both, read references/templates.md §10 for the Influencer domain fields and compatibility view. On Governed hosts, validate the source control artifacts first and generate the tracker as a read-only projection; never accept a tracker edit as an append, migration, stage/evidence/pointer change, reopen, or close. Creator rows must equal the locked scope and keep the active block at exactly eight fields; identity and close pointers stay adjacent. Identity and state heads must be resolvable, same-scope, unique, and non-forked. Any scope/contract/identity/checkpoint/verification/receipt/close/late-event gap shows PARTIAL CHECKPOINT COVERAGE — NOT CLOSE-ELIGIBLE; elapsed time never advances state. On semantic-only hosts, return the same view inline or save a NOT_VERIFIED compatibility snapshot after exact path-scoped authorization. payment_status records external readiness/evidence only and never sends money.
Each publication checkpoint gets an immutable, same-campaign/creator/checkpoint domain publication_receipt under fresh host authorization for the append-publication-receipt view update and a single-head supersession chain. This YAML block maps to shared evidence-observation; it is not an action-receipt and does not prove that this skill published anything. A real publish needs a separate exact action-intent before the executor and matching action-receipt after it. live_post_ref is only a qualified-resolver-backed opaque ref; raw URLs/slugs/provider IDs stay transient, and unresolved input remains unknown. verified disclosure/version match requires the exact observation, frozen approved asset/auditor, and evidence refs. Missing, mismatched, cross-scope, or forked evidence blocks close and routes the asset to creator-content-auditor; never infer approval or mutation.
A creator closes only when each applicable checkpoint has one controlling verified publication receipt or evidence-backed terminal non-applicability resolution and every §10 gate passes. A fresh atomic close-creator authorization must name its receipt append, row/evidence changes, and pointer update; campaign close needs a separate equivalently scoped close-campaign authorization and an exact creator→current-close mapping. Any fork blocks both branches. Corrections preserve history, re-evaluate gates, and append newly authorized close heads only when warranted; a reference-only correction does not manually reopen work.
Late rights/post/attribution/payment/data evidence uses the campaign-bound §10 late_event_note under a fresh append-late-event authorization. Any accompanying stage/action/evidence/pointer mutation must be named in that atomic authorization or separately approved. supersede-artifact binds the exact old and replacement refs of the same scope and meaning. manual-reopen is only for new campaign-owned work; otherwise append the correction and, if gates pass, fresh close heads. Preserve history and never auto-reopen or invent a reopened stage.
When the user asks what needs attention, generate the template §10 exception queue from the validated source artifacts (or a clearly labeled semantic-only compatibility snapshot) using an explicit as_of time and user-selected rights horizon. It is a read-only projection—no cron, polling, automatic stage change, projection write-back, or external mutation.
Example
User: "Create a campaign plan for a new sustainable sneaker launch targeting Gen Z on TikTok and Instagram with a $50K budget"
Output: A plan skeleton preserving the supplied audience, platforms, and $50K total. Sustainability claims/message canon, creator mix, content format, promo/attribution mechanic, rates, contingency, KPI targets, and exact dates remain NEEDS_INPUT until the user supplies them or compatible source-dated anchors. No micro-heavy, UGC, promo-code, or percentage default is inferred. (Fuller walkthrough in references/templates.md.)
Reference Materials
references/templates.md — fill-in templates for all nine planning steps, the optional lightweight WARM tracker, the worked example, and success tips.
references/influencer-tiers.md — declaration contract for user/source-dated partner models and follower bands; it supplies no universal range or performance claim.
Tracker-only, publication/approval blocker: creator-content-auditor — verify the cited live post against the frozen approved asset; this does not authorize an edit or amplification.
Tracker-only, readback/reconciliation due: performance-analyzer — produce the dated §8 readback artifact; final reporting may package that artifact only after the evidence exists.
Tracker-only, open operational item: hand off only to the named owner/skill required by the current next_action; do not route back to briefing for a late-stage row. If every close gate passes and no action remains, stop and report chain-complete.
Termination note: keep a visited-set of skills invoked this session. If the applicable next skill has already run this session, stop and report the chain complete rather than re-invoking. Do not chain deeper than 3 hops from the originating request.
