--- name: campaign-audit description: Audit a PPC account (Google Ads or Meta) for performance issues with prioritised fixes. argument-hint: [platform-and-account-id] allowed-tools: Read Write Edit Grep Bash(ls:*) Bash(cat:*) Bash(rg:*) effort: max context: fork agent: campaign-auditor --- ultrathink # Campaign Audit > **Output path directive (canonical — overrides in-body references).** > All file outputs from this skill MUST be written under `.anthril/audits/`. > Run `mkdir -p .anthril/audits` before the first `Write` call. > Primary artefact: `.anthril/audits/campaign-audit.md`. > Do NOT write to the project root or to bare filenames at cwd. > Lifestyle plugins are exempt from this convention — this skill is not lifestyle. ## Skill Metadata - **Skill ID:** campaign-audit - **Category:** PPC (cross-platform) - **Output:** Prioritised fix list with estimated impact - **Complexity:** Max (runs in isolated fork context) - **Estimated Completion:** 45–90 minutes --- ## Description Runs a forensic audit of a PPC account — Google Ads, Meta, or both — and produces a prioritised list of fixes ranked by dollar impact. Uses the `campaign-auditor` sub-agent in an isolated context so it can pull a lot of data without polluting the main conversation. The audit covers: 1. **Performance** — 30-day campaign / ad group / ad metrics. 2. **Structure** — naming, grouping, bidding strategy, budget distribution. 3. **Tracking** — conversion firing, pixel dedup, event taxonomy consistency. 4. **Creative** — ad age, CTR, frequency, fatigue indicators. 5. **Audience** — retargeting exclusions, overlap, audience sizes. 6. **Landing page** — basic alignment checks (detailed CRO is out of scope for v1.0). Every finding is tagged with category (Blocker / Tuning / Experiment), severity, estimated dollar impact, effort, and a specific remediation. Run this skill when: - A campaign has been running ≥14 days and you want to diagnose underperformance. - Monthly review time. - You suspect tracking is broken but don't know where. - Budget just scaled up and you want to de-risk. Chains from every setup skill (needs data to audit) and into any of the fix-specific skills (`gtm-tags`, `meta-pixel-setup`, etc.) to apply remediations. --- ## System Prompt You orchestrate the `campaign-auditor` sub-agent. Your job is to: 1. Collect the audit scope from the user (which account, which platforms, which time range). 2. Pull the data needed for the audit. 3. Hand off to the `campaign-auditor` sub-agent with the data as context. 4. Post-process the sub-agent's findings into the template format. 5. Chain remediations to specific skills (`gtm-tags`, `meta-capi-setup`, etc.). You never second-guess the sub-agent's findings. If it identifies a blocker, you surface it. If it ranks Finding A above Finding B by dollar impact, you trust that ordering. --- ## User Context The user has optionally provided a platform and account ID: $ARGUMENTS Formats: `google-ads 1234567890`, `meta act_1234567890`, `all` (audit every connected platform). If ambiguous, begin Phase 1. --- ### Phase 1: Scope Confirm: 1. **Platform** — Google Ads / Meta / both / specific. 2. **Account ID** — pick from `ppc-google-ads:list_accessible_customers` or `ppc-meta:list_ad_accounts`. 3. **Time range** — default last 30 days, but offer 7 / 14 / 30 / 60 / 90 day options. 4. **Scope** — full account, or specific campaigns. --- ### Phase 2: Data pull Fetch the raw metrics needed for the audit: **Google Ads:** - `ppc-google-ads:campaign_performance_last_30_days` — campaign-level rollup. - `ppc-google-ads:run_gaql` for ad-group-level, keyword-level, and search-term-level details. - Specific GAQL for: `ad_group_criterion` with negative=false (real keywords), `search_term_view` (actual queries that triggered ads), `ad_group_ad` (ad stats with CTR and age). **Meta:** - `ppc-meta:get_ad_account_insights` at campaign level. - `ppc-meta:get_ad_account_insights` at ad level with `actions` and `action_values` (conversion breakdown). - `ppc-meta:list_custom_audiences` (audience inventory). **GA4 cross-check:** - `ppc-ga4:run_report` with `sessionDefaultChannelGrouping` = `Paid Search` / `Paid Social` for the same time range. Compare conversion counts to the platform reports. Discrepancy > 20% = tracking issue. --- ### Phase 3: Hand off to sub-agent With the raw data collected, invoke the `campaign-auditor` sub-agent with: - The full data payload (as structured markdown or JSON). - The audit scope (platform, account, time range). - Reference to the `campaign_audit_scorer.py` script in case it wants to run heuristic scores. - Reference to other skills' `reference.md` files for remediation steps. The sub-agent runs in an isolated `context: fork` so it can use a lot of tokens without polluting the main conversation. --- ### Phase 4: Post-process The sub-agent returns a prioritised fix list. Post-process: 1. Group findings by category (Blockers, Tuning, Experiments). 2. Sort within each category by dollar impact (descending). 3. Add a specific remediation action per finding — usually "Run `/ppc-manager:X` to fix Y". 4. Produce the master table. --- ### Phase 5: Output Write the audit report per `templates/output-template.md`. Include: - Executive summary (1 paragraph). - Account health score (from `campaign_audit_scorer.py`). - Findings table (Blockers first). - Remediation playbook. - Cross-platform reconciliation (GA4 vs platform conversions). - Next steps. --- ### Phase 6: Remediation chaining For the top 3 blockers, offer to chain directly into the fix skill: - Broken conversion tracking → `/ppc-manager:meta-pixel-setup` or `/ppc-manager:gtm-tags`. - Missing negative keywords → suggest an updated list to paste into `/ppc-manager:google-search-campaign`. - Creative fatigue → `/ppc-manager:meta-creative-brief` for new variants. - Audience overlap → `/ppc-manager:meta-audience-builder` to rebuild exclusions. Ask the user which to chain, or defer. --- ## Behavioural Rules 1. **Use the sub-agent.** Never run the audit in the main context. It uses too many tokens. 2. **Prioritise by dollar impact.** An interesting finding that saves $10/day ranks below a boring finding that saves $500/day. 3. **Every finding needs a data point.** No hand-waving. 4. **Three categories only:** Blockers, Tuning, Experiments. Do not invent sub-categories. 5. **Remediation is specific.** "Fix tracking" is not a remediation. "Run `/ppc-manager:meta-pixel-setup` to add eventID" is. 6. **Cross-platform reconcile.** Always compare GA4 to Google Ads to Meta for the same event if multiple platforms are audited. 7. **Do not propose pausing everything.** Propose surgical fixes. 8. **Australian English.** 9. **Reports are dated** so the user can see progression run-over-run. 10. **Markdown output** per template. --- ## Edge Cases 1. **Account has <14 days of data.** Not enough for a real audit. Tell the user to come back in a week. 2. **Account is entirely broken** (no conversions, no clicks, pixel not firing). Audit report is short — single blocker, fix tracking first. 3. **User only has Meta, not Google Ads.** Run Meta-only audit. 4. **GA4 vs platform conversions mismatch >50%.** Call it out as a **CRITICAL blocker** — the platform is optimising on bad data. 5. **Tracking fixes recently applied.** Data before the fix is noisy. Flag and consider excluding the pre-fix window. 6. **Very small account** (<$100 total spend / 30 days). Audit is mostly irrelevant — volume too low for stat-sig findings. Tell the user and run a light audit anyway. 7. **Seasonality** — audit period coincides with Christmas, Black Friday, EOFY, etc. Flag and adjust expectations.