Training
for every role.

Install the agentia binary globally, store your Agentia CI/CD, Testing, and AI credentials in the OS keychain, verify the setup, then create project-local Agent Skills. Expected result: the project contains Agentia-managed Markdown skills, so a compatible agent can load the relevant CICD, Testing, or AI playbook before operating Agentia.

For developers, QA, release managers, and platform engineers. Install or update the CLI, run setup skills create, choose the agents, cursor or claude target, select all packs or only cicd, testing or ai, then inspect the created folders. Ask your IDE agent: "Use the Agentia CICD skill to inspect my assigned Story and explain the local delivery flow. Do not promote anything." The agent reads the CICD skill, checks authentication, lists and reads the Story before any mutation, explains the local path without mixing in cloud commit commands, and does not promote without explicit approval. Refresh later with setup skills update; a file without the Agentia managed marker is never overwritten, and --force is recovery behaviour for a conflicting unmanaged file, not the normal command.

Local and cloud work are separate delivery paths.

Local: work set creates or checks out feature/<story-name>, then you stage and commit with plain git (git add -A, git commit -m "…"). work test --local runs project-owned quality gates. work push registers the Copado Commit (a plain git push does not) and takes --full-metadata and --permissions when those change-list corrections are needed. work submit validates.

Cloud: cloud commit with metadata flags and --metadata-category SFDX for Salesforce metadata, then work status. Never continue a cloud commit with local work push or work submit.

Promotion and deployment (work submit --done, or cloud promote --deploy) require explicit approval; work done is a compatibility alias.

See what changed before you commit, diff a component across orgs, and derive a complete deployment selection straight from your Git diff.

Choose your entry point: start from a Story with cloud promote, or from a Promotion record with promotion run --operation merge. Add deployment only with explicit intent (--deploy or --operation merge_and_deploy). Inspect the promotion and each conflict before resolving, then resume the original Job Execution instead of starting a replacement run.

Inspection first: list with targeted filters, get the Job Execution, then read step logs as CI artifacts. Pause, kill, and restart come after status inspection and require explicit user intent. agentia cicd job manages CI/CD JobExecution records; agentia testing job manages Robotic Testing test/job definitions.

Inspect a template (selected-column graph with get, raw v2 payload with get-detail), validate source record search and destination matching including unmatched or ambiguous results, preview metadata sync and review needsAttention and autoUpdate entries before apply, then create a data commit and read its dataset files. Template, credential, Story, data commit and ContentVersion IDs are all different. If a JobExecution ID comes back, continue with agentia cicd job, not agentia testing job. Conversion and deletion need discovery, inspection, and a saved baseline first; delete is destructive.

CRT authentication is separate from CI/CD: set it with agentia auth set --crt <pak> --crt-domain <crt-domain> --crt-org <crt-org-id>, then gate on ready: true from agentia auth get --crt --json — a credential that is merely set can still be incomplete. Then work the lifecycle: testing project (with project permission for view/edit access), testing robot for where tests execute, testing job for uploaded or Git-backed Robot Framework scripts plus job refs, upload, download and job schedule with a cron expression, timezone and recording policy. Run with testing build run using suite, test, include, exclude, variable, metadata and rerun-failed controls, add --wait-for-result --timeout with --save-artifacts and --xunit, then inspect with build get --full, build latest, build search and build logs --tail-lines --timestamps. Keep secrets out of arguments with testing variable and testing user-parameter value-stdin or value-file, drive matrices with testing data-table and --datatable, and debug interactively with testing live start / execute / inspect / screenshot / stop. Every short-running command supports --json, destructive commands need --yes (MCP uses confirm: true), and none of this is the same as cicd work test --local.

Let the skill choose the mode: ask is the automation and agent-to-agent path with exactly one lifecycle selector (plan, build, test, release, operate, knowledge), while chat is TTY-only for a human terminal. Context flags such as workspace, dialogue, User Story, Salesforce org, and component enrich a clear prompt; they do not replace one. MCP supports buffered single-turn ask, not interactive chat.

MCP gives the agent tools. Agentia Skills give it the operating playbook. Install both when the agent should execute Agentia workflows safely from the IDE, then keep domain boundaries in the tool chain: local work, cloud work, Robotic Testing, and AI each have their own sequence. Promotion and deployment always need explicit user approval, so no work-done tool call is automatic.