Using Artifacts with Agents

Loading artifacts into an agent is how you give it the files and skills a task needs. Dana's email asks for the Q3 numbers, which are Acme's files rather than the state of any env, so this page loads them into assistant, the agent working in the email instance, with a skill for sending them. The steps form one task, which Tasks and steps covers.

Load files into the agentCopy Anchor Link

The Q3 numbers are q3-workpapers, the file artifacts from Creating your artifacts. A load_artifact step with agent_name copies an artifact into that agent's container. These steps seed the email instance, deploy assistant next to it and give it the files:

[
  {"id": "deploy", "type": "deploy_env", "env_id": "email", "env_version": 1},
  {"id": "seed", "type": "load_artifact", "env_id": "email",
   "artifact_id": "acme-email", "artifact_version": 2,
   "depends_on": [{"task_step_id": "deploy"}]},
  {"id": "agent", "type": "deploy_agent", "a2a_agent_id": "my-agent",
   "agent_name": "assistant", "env_ids": ["email"],
   "depends_on": [{"task_step_id": "seed"}]},
  {"id": "workpapers", "type": "load_artifact", "agent_name": "assistant",
   "artifact_id": "q3-workpapers", "artifact_version": 1,
   "depends_on": [{"task_step_id": "agent"}]}
]

The files land in /tmp/file_artifacts, as q3-numbers.csv and notes.md, and destination_path puts them somewhere else. The same step takes a single file artifact too, or an environment universe, which an agent gets as one folder per environment name: the demo's second agent, analyst, gets acme as email/emails.json and contacts/contacts.json.

SkillsCopy Anchor Link

Skills are modeled as artifacts: a stored skill is versioned like any other artifact, and a task pins the version it gives an agent. Skills covers writing one, storing it and adding it in a task. This task's add_skills step gives assistant the skill q3-report, and a second skill that says where the q3-workpapers files are:

{"id": "skills", "type": "add_skills", "agent_name": "assistant",
 "skills": [{"skill_artifact_id": "q3-report", "skill_artifact_version": 1}],
 "file_artifact_universe_ids": ["q3-workpapers"],
 "depends_on": [{"task_step_id": "workpapers"}]}