AI Agent & Automation Workflows

Inventory Agentforce agents in the project — both source-controlled AgentDefinition files and live agents in the target org. Maps each agent's topics, sub-agents, and bound actions (Apex, Flow, prompt template, MCP tool). Counterpart to /flow-audit for the agent surface.

Persist Salesforce agent evaluation scores over time and surface regressions per agent and per axis (factuality, completeness, tone, refusal-correctness, action-correctness). Sibling to /argo:coverage-trend but for agent evals.

Validate readiness for AgentExchange listing. Checks required metadata, eval coverage, Trust Layer config, license boundaries, and listing artifacts (icon, screenshots, install URL). Reports a severity-graded readiness report so submission isn't a back-and-forth.

Generate or update an Agent Script YAML specification by wrapping sf agent generate agent-spec with project-specific defaults from sf-project.json and project-context. Iterative refinement loop — re-run with --spec to refine an existing one.

Run agent evaluation tests via sf agent test run against the target org's Testing Center. Produces severity-graded findings; CI mode emits SARIF for GitHub Code Scanning. Persists each run to history for /argo:agent-eval-trend.

Decide whether a feature should be built as an Agentforce Agent, a Flow, an Apex Trigger, or other mechanism. Replaces /argo:flow-vs-apex when agents are in scope; otherwise falls through to it. Asks 6–10 questions and recommends with rationale.

Decide whether a piece of trigger logic belongs in before insert/update or after insert/update. Asks 4–5 quick questions, then recommends with rationale.

Compute cyclomatic complexity per Apex method and per LWC JS function. Flags hotspots above a configurable threshold (default 10). Useful for spotting methods that need refactoring before they become unmaintainable.

Persist Apex code-coverage results over time and surface regressions. After each /argo:test-coverage run (or via this skill directly), append the result to a per-project history file and show the trend. PR-mode diffs against the baseline.

Build a Salesforce destructiveChanges.xml interactively from a list of components to delete. Validates that nothing else in source still references them before authoring the file.

Deploy only the metadata that's changed since a git ref (default main) instead of the full project. Reads git diff to compute the changed-file set, intersects with the project's paths, and runs sf project deploy start with --source-dir flags. Useful for fast PR validation deploys.

Find all references to a Salesforce field across LWC, Apex, layouts, validation rules, formula fields, Flows, and reports. Use before deleting or renaming a field, or when assessing the blast radius of a field-level change.

Run a named Apex test class N times against the target org and report tests with non-deterministic results. Useful when a test fails intermittently in CI but passes locally — runs prove flakiness.

Inventory active Flows in the target org, classify them by trigger type and purpose, flag overlaps with Apex triggers in source, and surface flows that lack source-control coverage. Useful before recommending new automation (Flow vs. Trigger decision) and during automation audits.

Decide whether a new piece of automation should be built as a Flow, an Apex Trigger, or another mechanism. Asks the user 6–8 quick questions, then recommends with rationale. No code generated.

Static-analyze Apex source for DML and SOQL operations missing CRUD/FLS enforcement. Reports every insert/update/delete/upsert/merge and dynamic SOQL that doesn't use USERMODE / SECURITYENFORCED / Schema.DescribeSObjectResult checks.

Generate (or validate) the LLM Governance on AI Gateway configuration for the project — token quotas, model allowlist, fallback policy, per-agent rate limits. Output is a .claude/ai-gateway. .json file the org's gateway service consumes.

Static estimate of governor-limit consumption per public Apex method — count of SOQL queries, DML statements, callouts, queries-in-loops, async enqueues. Useful for spotting bulk-unsafe code before runtime.

Decide whether a new UI component should be built as an LWC or a React component on Salesforce. Asks 5–7 quick questions, recommends with rationale and references to RX- / SF- patterns.

Decide whether to expose work as an MCP tool, an Apex REST endpoint, or a Platform Event. MCP tools are a peer integration pattern; this skill maps the choice space.

Diff local Salesforce source vs. the target org and produce a structured drift report (setup-only changes, source-only changes, conflicts). Useful before deploys, after merge, or when investigating "why does the org behave differently from source?

Snapshot the target Salesforce org's schema (objects, fields, profiles, permission sets, installed packages) into a local cache for downstream argo skills. Cache is for offline use, batch reports, and snapshot-vs-snapshot diffs.

Manage Salesforce 2GP / unlocked package versions — create, promote (release), and list versions. Wraps sf package version create/promote/list with conventions for changelog generation and release tagging.

Install, list, or remove argo pattern packs (Agentforce, React, Platform Events, Change Data Capture, External Objects, Big Objects, Field Service, Industries, CMS, Data Cloud). Each pack adds domain-specific patterns, quality-checklist items, and optionally agents/skills to the project.