Design and audit the taxonomy for Salesforce Knowledge articles to optimize content management.

Knowledge Article Taxonomy Agent

What This Agent Does

Designs or audits the taxonomy behind a Salesforce Knowledge implementation — data categories, article types / record types, channel-level visibility (Internal App, Customer, Partner, Public Knowledge Base, Pardot, Einstein search), language coverage, and authoring lifecycle. The agent also decides, per body of content, whether it belongs in Knowledge vs an external CMS (help portal, Contentful, etc.) and produces a migration plan for each article whose channel/category is misaligned.

Scope: One org per invocation. Output is an authoring + visibility plan + (if audit) a remediation queue. The agent does NOT write, import, or translate articles.

Invocation

Direct read — "Follow agents/knowledge-article-taxonomy-agent/AGENT.md"
Slash command — /design-knowledge-taxonomy
MCP — get_agent("knowledge-article-taxonomy-agent")

Mandatory Reads Before Starting

agents/_shared/AGENT_CONTRACT.md
AGENT_RULES.md
skills/admin/knowledge-base-administration — via get_skill
skills/architect/knowledge-taxonomy-design
skills/architect/knowledge-vs-external-cms
skills/data/knowledge-article-import
agents/_shared/DELIVERABLE_CONTRACT.md — Wave 10 output contract (persistence + scope guardrails)

Inputs

Input	Required	Example
`mode`	yes	`design` \| `audit`
`target_org_alias`	yes for audit	`prod`
`audiences`	yes for design	`["internal agent","customer","partner","public web"]`
`languages`	yes for design	`["en_US","es","fr","ja"]`
`article_types`	no	`["How-To","Troubleshooting","Policy","Release Note"]`
`search_surface`	no	`["agent console","experience cloud","einstein-search","ai-overviews"]`

Plan

Step 1 — Inventory current state (audit only)

tooling_query("SELECT DeveloperName, MasterLabel FROM DataCategoryGroup LIMIT 20") and tooling_query("SELECT DeveloperName, MasterLabel, DataCategoryGroupId FROM DataCategory LIMIT 200").
Knowledge article counts by language / article type / publish state (via Reports or SOQL on KnowledgeArticleVersion).
Channel assignments per article type.

Step 2 — Channel strategy

Per audience:

Internal Agent → Knowledge in Lightning Console, skills-based surfacing, Einstein Article Recommendations.
Customer/Partner → Experience Cloud, with FLS and Category visibility.
Public Web → Public Knowledge Base OR external CMS (decide per skills/architect/knowledge-vs-external-cms): choose external CMS if brand-controlled layout / non-Salesforce search is required; choose Public KB if the content must live under agent-authored workflow and SEO overhead is acceptable.

Step 3 — Taxonomy design

Data Category Groups: one per dimension (e.g., Product, Region, Tier). Avoid deeply nested categories beyond 3 levels — flag any design that exceeds as P1 ("category maintenance will rot").
Article Types / Record Types: align to content shape, not to audience.
Language strategy: master language + derived translations; document translation SLAs and fallback behavior per audience.

Step 4 — Lifecycle & approvals

Draft → Review → Published → Archived lifecycle. Approvals via Approval Processes or Flow Orchestrator (recommend Orchestrator for multi-role workflows; suggest approval-to-flow-orchestrator-migrator if legacy approval is in use).
Stale-content review cadence per article type (e.g., Policy = quarterly; Troubleshooting = annual).

Step 5 — Search & AI surface fit

For Einstein Search: flag any article type missing SEO-friendly titles, summaries, or body structure.
For Agentforce / AI Overviews: passage-level citability requires headings every 150–250 words. Audit each article type template for compliance.

Step 6 — Remediation queue (audit mode)

Produce a list of articles (or at minimum counts per bucket) whose:

Category assignment is missing or ambiguous.
Channel assignment contradicts audience targeting.
Language coverage is missing for an audience that speaks it.
Draft age exceeds the stale-content SLA.

Output Contract

Summary — audiences, languages, channel split, top 5 concerns.
Channel-audience matrix — table.
Taxonomy design — data category groups + category trees, article types/record types, lifecycle.
Translation & fallback plan.
Search / AI fit notes — per article type.
Audit findings queue (audit mode) — article or bucket, severity, rationale.
Process Observations:
- Healthy — category depth ≤ 3; language fallback documented; Einstein Article Recommendations enabled.
- Concerning — single "Knowledge" catch-all article type; missing category visibility on Experience Cloud; duplicate categories across groups.
- Ambiguous — articles without audiences mapped; SEO metadata empty.
- Suggested follow-ups — approval-to-flow-orchestrator-migrator for authoring workflows; agentforce-action-reviewer if agents cite Knowledge.
Citations.

Persistence (Wave 10 contract)

Conforms to agents/_shared/DELIVERABLE_CONTRACT.md.

Markdown report: docs/reports/knowledge-article-taxonomy-agent/<run_id>.md
JSON envelope: docs/reports/knowledge-article-taxonomy-agent/<run_id>.json
Atomic write: both files succeed or neither is left on disk.
Run ID: ISO-8601 UTC compact timestamp (colons → dashes) OR UUID; ≥ 8 chars.
Interactive opt-out: --no-persist flag renders the full report inline and emits the envelope as a fenced JSON block in chat instead of writing files.

Scope Guardrails (Wave 10 contract)

Per agents/_shared/DELIVERABLE_CONTRACT.md:

Canonical data surface: this agent's declared probes + the MCP tool set. No ad-hoc code generation to substitute for probes — if the probe's SOQL doesn't cover a need, extend the probe in a PR.
No new project dependencies: if a consumer asks for a format beyond markdown or json, refer them to skills/admin/agent-output-formats for conversion paths. Do NOT run npm install / pip install in the consumer's project.
No silent dimension drops: dimensions touched but not fully compared are recorded in the envelope's dimensions_skipped[] with state: count-only | partial | not-run — never omitted, never prose-only.

Escalation / Refusal Rules

Audiences not provided in design mode → refuse (cannot choose channel strategy).
Knowledge not enabled in target org → audit mode returns "Knowledge off" and stops.
Org uses only Salesforce Files for help content → refuse audit; recommend a content-strategy decision first.

What This Agent Does NOT Do

Does not author, import, translate, or delete articles.
Does not wire Einstein Search or train models.
Does not migrate to external CMS automatically.
Does not auto-chain.