# Channel Advisor > Use this skill to recommend a channel mix and tactics for a specific marketing goal, audience, and budget context. Invoke when the user asks to "recommend a channel mix", "which channels should we use", or wants a channel strategy for a goal or campaign. Produces recommended primary and supporting channels, tactical recommendations, and flags prerequisites the user may be missing. - Skill: `clever-cc-plugins/channel-advisor` (Agent Skill) - Install (CLI): `npx skillmds@latest add clever-cc-plugins/channel-advisor` - Raw SKILL.md: https://api.skillmd.com/api/skills/clever-cc-plugins/channel-advisor/raw - Safety review: pending - Works with: Claude Code, Claude.ai, OpenAI Codex - Category: Marketing & Growth - Author: clever-cc-plugins (https://skillmd.com/u/clever-cc-plugins) - Updated: 2026-09-17 - Page: https://skillmd.com/skills/clever-cc-plugins/channel-advisor --- # Channel Advisor Skill This skill recommends a channel mix and tactics for a specific goal, audience, and budget context, sequenced against the RACE funnel model. This skill follows the shared step sequence in `../_shared/skill-contract.md`. Steps below add only channel-advisor-specific behavior. ## Step 0–1: Recall learnings and load context If `.claude/learnings.md` exists, read it silently. Apply relevant entries — in particular, channel preferences, constraints, or vetoed channels. Do not announce this step. If the file is absent, continue normally. Read the `## Context files` table in `CLAUDE.md`. Match the following needs against the **Summary** column semantically — never against a label or filename: - **Business goals/KPIs** — campaign objectives, business targets, success metrics, growth stage, revenue targets - **Target audience** — positioning, audience profile, customer segments, buyer personas, organization background (prefer a `positioning` output file if registered) Note silently if present (do not ask about): current marketing channels, budget, and a prior strategy performance review (`strategy-performance-review` output) if registered — factor validated/invalidated channel bets into the recommendation. ## Step 2: Gating — Three-State Model This skill gates on **exactly two** needs: **business goals** and **target audience**. For each, resolve one of three states: 1. **Covered** — present in the context table (Summary match) → use silently, no ask. 2. **Absent-but-supplied** — ask **once** (never twice in one invocation); user answers → use, non-degraded. 3. **Absent-and-unanswered** — asked once, no usable answer → that section is labelled **⚠ DEGRADED**, never fabricated, and the rest of the recommendation still generates. ### Check business goals Search the context table Summary column for semantic matches: business goals, KPIs, objectives, targets, growth stage, success metrics. If **covered**, use silently. If **absent**, ask once: > "I need to understand what this channel recommendation should achieve. Briefly: What are your > primary business goals or KPIs right now? (e.g. revenue growth, lead volume, retention, brand > awareness) Is there a specific target or metric this recommendation should support?" If the user provides usable answers, proceed **non-degraded**. If the user declines or the answer is too vague to use, label the section `## Business Goals ⚠ DEGRADED` in the output and continue. ### Check target audience Search the context table Summary column for semantic matches: target audience, positioning, organization background, customer segments, buyer profile. Prefer a `positioning` output file's Summary if one is registered; else fall back to organization background. If **covered**, use silently. If **absent**, ask once: > "I need to understand who this channel recommendation is for. Briefly: Who is the target > audience? (role, company type/size if B2B, or demographic if B2C) What stage are they > typically at when they encounter you? (just learning you exist, actively comparing options, > or already a customer)" If the user provides usable answers, proceed **non-degraded**. If the user declines or the answer is too vague to use, label the section `## Target Audience ⚠ DEGRADED` in the output and continue. ## Step 3: Calibrate Take the goal from `$ARGUMENTS` if given, else from loaded business goals or the Step 2 gating answer. No campaign-type inference is needed here — just goal, audience, and whatever current-channel/budget context is loaded. ## Step 4: Generate Apply RACE from `../_shared/campaign-frameworks.md`. **State the mapping being applied** before generating. Example: > "Using RACE to sequence channels by funnel stage." Produce a channel recommendation that includes: - **Primary channel(s)** — with rationale tied to the loaded goal and audience. - **Supporting channel(s)** — channels that reinforce the primary channel(s). - **Tactical recommendations** — at least one concrete tactic per recommended channel. - **Existing-channel calls** — for each existing channel loaded from context, a maintain/grow/reduce/exit call, based on whether the advisor is recommending continued/increased investment (maintain/grow), reduced investment (reduce), or discontinuing it (exit) — including channels the advisor is NOT recommending going forward. Omit this element entirely if no current-channels context is loaded. - **Prerequisites** — for any recommended channel where the loaded context suggests it may be out of reach (budget, capability, or skills the user may not have), a "prerequisites" note naming what's missing. ## Step 5–6: Quality check and delimited output Review the generated channel recommendation for: - **Completeness**: All five elements above are present (existing-channel calls only when current-channels context was loaded). - **Extractability**: Primary channels, supporting channels, and tactical recommendations are explicitly named so the user can act on them immediately. - **Framework alignment**: RACE stages are clearly sequenced in the recommendation. - **Consistency**: All sections reinforce the same goal and audience. Fix any gaps before proceeding. If Step 2 ended in a degraded state for one or both gated needs (business goals and/or target audience), prefix the output with: ``` ⚠ DEGRADED OUTPUT — generated without: ``` For example, if target audience is degraded but business goals are not, prefix with: `⚠ DEGRADED OUTPUT — generated without: target audience` If both are degraded, prefix with: `⚠ DEGRADED OUTPUT — generated without: business goals, target audience` If neither need is degraded, omit the prefix. Return the channel recommendation inside clear delimiters: ``` ───────────────────────────────────────────────────────────────── ## Channel Recommendation ───────────────────────────────────────────────────────────────── [Generated channel recommendation here] ───────────────────────────────────────────────────────────────── ``` ## Step 7: Feedback Store learnings tagged `[cc-concept:channel-advisor]` in `.claude/learnings.md`. Examples: ``` [cc-concept:channel-advisor] client vetoed paid social entirely, treat as excluded from all future recommendations — 2026-07-19 [cc-concept:channel-advisor] client's existing email list is small (<500) and shouldn't be weighted as a primary channel yet — 2026-07-19 ``` If the session revealed a channel preference, constraint, or exclusion that future runs should apply, record it now.