/architect-perms — Design or audit Permission Sets + Permission Set Groups
Design or audit Permission Sets and Permission Set Groups for Salesforce orgs.
Wraps agents/permission-set-architect/AGENT.md. Two modes: design (from a persona) or audit (against the live org).
Step 1 — Collect inputs
Ask the user:
1. Mode: design or audit?
2. Target org alias (required — live-org probe in both modes)?
3. If mode = design:
Persona description (job title + objects touched + features used + sensitivity)?
Example: "SDR in NA, works Leads + Opportunities, cannot export data, light reporting"
4. If mode = audit:
Scope? org / psg:<PSG_Name> / ps:<PS_Name> / user:<username>
5. If mode = design:
Extending an existing PSG? (Optional — pass the PSG name)
If mode is design and persona is under 10 words, STOP and ask.
Step 2 — Load the agent
Read agents/permission-set-architect/AGENT.md + mandatory reads. In particular, read templates/admin/permission-set-patterns.md — it's the taxonomy the audit + design both conform to.
Step 3 — Execute the plan
Design mode:
- Decompose persona into 6 PS categories
- Probe for reusable Feature PSes (
list_permission_sets+describe_permission_set) - Propose the PSG composition
- Identify muting opportunities
- Emit metadata stubs
- Emit deployment order
Audit mode:
- Scope the probe
- Classify each PS against the taxonomy
- Detect anti-patterns (Modify All Data on persona, single-user PSes, muting PS with no parent, etc.)
- Score the org
Step 4 — Deliver the output
- Summary + confidence
- Findings table (audit) / Composition table (design)
- Metadata stubs (design only, fenced XML)
- Deployment order (design only)
- Recommended refactors (audit only)
- Process Observations
- Citations
Step 5 — Recommend follow-ups
/sharing-auditif the audit surfaced broad record-level access concerns (Wave B agent)/catalog-integrationsif integration-user permissions surfaced as findings (Wave B agent)/analyze-field-impactif FLS changes are implied/design-objectif missing objects underlie the design gap
What this command does NOT do
- Does not assign Permission Sets to users.
- Does not deploy metadata.
- Does not modify an existing PS in place (refactors are proposed as new PSes + migration).
- Does not audit Sharing Rules / OWD (that's the Wave B
sharing-audit-agent).