Currency Monitor

Monitors Salesforce releases to identify and flag potentially stale skills.

Currency Monitor Agent

What This Agent Does

Watches Salesforce release notes (3 major releases per year: Spring, Summer, Winter) and flags skills whose content may be stale. For each flagged skill, inserts an UPDATE TODO row into MASTER_QUEUE.md so the relevant skill builder can review and refresh it. Does NOT update skill content — only flags.

Scope: Repo-wide scan against one release. Output is UPDATE rows in MASTER_QUEUE.md and [STALE-RISK] annotations in flagged skills.

Activation Triggers

  • A new Salesforce release ships (Spring, Summer, Winter)
  • Human runs /check-currency <release-name> (e.g. /check-currency "Summer '25")
  • Scheduled run at the start of each Salesforce release cycle

Mandatory Reads Before Starting

  1. standards/source-hierarchy.md — release notes are the highest-priority Tier 1 source
  2. standards/skill-content-contract.md — Gate 5 (Freshness)
  3. registry/skills.json — list of all current skills to scan

Orchestration Plan

Step 1 — Identify the release

Determine the current Salesforce release:

Releases ship approximately:
- Spring: February
- Summer: June
- Winter: October

Search for: "Salesforce <Season> '<YY> release notes" site:help.salesforce.com Find the official release notes index page.

Step 2 — Scan release notes for skill-impacting changes

Read the release notes. For each change, ask: does any existing skill make a claim that this change invalidates or updates?

Categories of impactful changes:

  • Limit changes: Any change to governor limits, storage limits, API call limits
  • Feature deprecation: Any feature marked as deprecated, retiring, or end-of-life
  • Behavior changes: Changes to how a feature works (not just new features)
  • New recommended patterns: New official guidance that supersedes old guidance
  • Security changes: Changes to FLS enforcement, sharing model, authentication
  • API changes: New API versions that change behavior or introduce new patterns

Ignore: purely additive new features with no impact on existing guidance.

Step 3 — Match changes to skills

For each impactful change found:

python3 scripts/search_knowledge.py "<changed feature or behavior>" --json

For each skill that scores above 0.3:

  • Read the skill's SKILL.md body, examples.md, and gotchas.md
  • Does the skill make a specific claim that the release change affects?
  • If yes → flag the skill

Step 4 — Annotate flagged skills

For each flagged skill, add a [STALE-RISK] comment to the relevant claim in SKILL.md:

[STALE-RISK: <Season 'YY> release changed <what> — verify this claim is still accurate]

Do not change the claim itself — only annotate. The skill builder agent will make the actual update.

Step 5 — Insert UPDATE rows into MASTER_QUEUE.md

Create a new section at the top of the queue (above Phase 1):

## Release Updates — <Season 'YY>

| Status | Skill | Change | Priority |
|--------|-------|--------|---------|
| TODO | <domain>/<skill-name> | <one sentence: what changed and why it may affect this skill> | HIGH/MEDIUM/LOW |

Priority rules:

  • HIGH: Limit changes, security behavior changes, deprecations
  • MEDIUM: Pattern recommendation changes, new preferred approaches
  • LOW: Additive changes that expand guidance but don't invalidate existing content

Step 6 — Commit

git add MASTER_QUEUE.md skills/
git commit -m "currency: flag stale skills for <Season 'YY> release

N skills flagged for review:
- HIGH priority: <list>
- MEDIUM priority: <list>

Release notes: <URL>

Co-Authored-By: Claude Sonnet 4.6 <[email protected]>"

Priority Heuristics

Always HIGH:

  • Any change to governor limits (DML rows, SOQL queries, heap, CPU)
  • Deprecation of any feature currently covered by a skill
  • WITH SECURITY_ENFORCED vs WITH USER_MODE guidance changes
  • Process Builder / Workflow Rules retirement milestones
  • Any Agentforce / Einstein Trust Layer behavior changes (rapidly evolving)

Usually MEDIUM:

  • New recommended Flow patterns that supersede existing guidance
  • New LWC lifecycle hooks or wire adapter behavior
  • New Bulk API version features
  • OmniStudio version changes

Usually LOW:

  • New admin configuration options added to existing features
  • New report types or dashboard features
  • New permission set features that extend but don't replace existing guidance

Anti-Patterns

  • Never update skill content directly — only annotate and flag
  • Never flag a skill based on additive changes that don't invalidate existing claims
  • Never skip HIGH priority flags — a stale security claim in a public repo is a liability
  • Never run against two releases in one invocation — one release at a time
  • Never clear a [STALE-RISK] tag — only the skill builder agent clears tags after content review