# Cx Macro Audit > Use to audit canned responses, macros, templates and saved replies for staleness, contradictions, overuse and gaps. Trigger for "audit our macros", "are our canned responses up to date", "which templates do agents actually use", macro sprawl, template cleanup, conflicting saved replies, or before migrating macros to a new helpdesk. - Skill: `rulebase-co/cx-macro-audit` (Agent Skill) - Install (CLI): `npx skillmds@latest add rulebase-co/cx-macro-audit` - Raw SKILL.md: https://api.skillmd.com/api/skills/rulebase-co/cx-macro-audit/raw - Safety review: PASS (external: skill-scanner PASS, skillspector CAUTION) - Works with: Claude Code, Claude.ai, OpenAI Codex - Category: Security - Author: rulebase-co (https://skillmd.com/u/rulebase-co) - Updated: 2026-08-19 - Page: https://skillmd.com/skills/rulebase-co/cx-macro-audit --- # Auditing macros and templates Macros are the highest-leverage content in a support operation and the least maintained. One stale macro sends a wrong answer hundreds of times before anyone notices — and because it is consistent, it does not look like an error. It looks like policy. That is the reason this audit matters more than its unglamorous framing suggests: **a wrong macro is the most efficient way an organisation has of being uniformly wrong**, and it is invisible to per-agent QA, which sees consistent behaviour and passes it. ## Start from usage, not from the list Most macro libraries have hundreds of entries and a small number that do nearly all the work. Pull usage counts first and the audit prioritises itself. Expect roughly: - **A short head** — a handful of macros covering most sends. **Audit these first and most carefully.** An error here has already reached thousands of customers. - **A long tail of near-zero usage** — candidates for deletion, and worth checking for duplicates of the head. - **A dead middle** — created for a campaign or a situation that has passed. If your helpdesk does not record macro usage, detect it by matching distinctive phrases from macro bodies against sent messages. Report the match rate so the reader knows how complete the usage picture is. ## What to check on the head **1. Factual currency.** Every claim — fee, timeline, eligibility, process step, link — verified against current policy. Check the last-edited date against the last policy change on that topic; a macro older than the policy it describes is the finding. **2. Contradictions between macros.** Two macros answering the same question differently is worse than either being wrong alone, because which answer a customer gets depends on which agent they reach. Group macros by topic and compare. This is the check that most often finds something serious and is almost never run. **3. Dead links and stale references.** Links to retired pages, removed features, deprecated forms, or an old brand name. **4. Placeholders that leak.** Unfilled variables reaching customers as `{{first_name}}` or, worse, a placeholder that silently renders as blank mid-sentence. Search sent messages for template syntax and for the tell-tale double spaces and orphaned commas a blank substitution leaves. **5. Tone and reading level** against the team's documented standard, and whether the macro still sounds like the current brand. Macros calcify a voice from years ago. **6. Compliance content.** Required disclosures present and current, and no macro making a commitment or giving regulated advice it should not. **7. Localisation currency.** Where macros are translated, check the translations were updated when the source changed. A source macro edited and its translations left behind is extremely common and means non-English markets are running the old policy. ## Usage pattern findings The way macros are used is as informative as their content: - **A macro used far more than the volume of its topic** suggests it is being applied to situations it does not fit — agents reaching for the nearest thing. - **Heavy editing before send**, where you can detect it, means the macro is close but wrong. That is a rewrite candidate with the correction already written by the agents. - **A high-volume contact driver with no macro** is the clearest gap. Cross-reference the contact-driver ranking against macro coverage; the uncovered high-volume drivers are the highest-value additions. - **Macros used almost exclusively by one agent** are personal shortcuts. Either promote them or accept that the library is not the real source of answers. - **Macros sent as a first response and followed by a repeat contact** are candidates for being unhelpful. Link macro usage to repeat-contact rate where you can; a macro with a materially higher repeat rate than the topic average is not resolving anything. ## Deletion is the point An oversized library is itself the problem: agents cannot find the right macro, so they improvise or reuse the nearest, and consistency collapses. A library that has doubled in size has usually got worse. Recommend deletion aggressively for the tail, but: - **Check for seasonal use** before deleting something unused for months. - **Deletion is often irreversible** in helpdesk configuration, and it may break automations, triggers or workflows that reference the macro. Check references first, export the current definitions before changing anything, and treat bulk deletion as a reviewed change with a plan rather than a cleanup task. - **Prefer deactivating to deleting** where the platform supports it. ## Traps - **Macro text is not what the customer received.** Agents edit before sending, and variables substitute. Audit the macro *and* a sample of sent messages that used it. - **Personal or team-scoped macros** may not appear in an admin list. Ask. - **Automation-sent templates** — triggers, autoresponders, satisfaction emails — are usually a separate object type and are audited even less often. Include them; they send at higher volume than any human-used macro. - **Do not judge a macro purely on tone.** A blunt, correct, well-used macro beats a warm one that misstates a fee. ## Present results to the user 1. **Usage distribution** — how concentrated the library is, and the coverage of your usage data. 2. **Findings on the high-usage head**, worst first, with an estimate of how many customers each error has already reached. That estimate is what gets these fixed. 3. **Contradictions**, grouped by topic. 4. **Gaps** — high-volume drivers with no macro. 5. **Localisation drift** — translations older than their source. 6. **Deletion candidates**, with reference checks and a note on reversibility. 7. **Rewrite candidates**, prioritised by usage × severity, with the agent edits as evidence where available.