Task Context Profiles

Task context profiles let a host tell Aionis what kind of work the next Agent turn is doing.

They do not change memory storage, recall, lifecycle state, admission authority, or feedback attribution. They only change how the governed guide context is rendered for the Agent.

Use profiles when one Runtime serves different Agent loops:


const guide = await aionis.guide({
  query_text: "Answer from the retained source evidence.",
  mode: "full_power",
  context_mode: "compact_agent",
  task_context_profile: "long_qa",
  consumer_agent_id: "qa-agent",
  run_id: "run-001",
});

If omitted, the profile is general.

Who Selects The Profile

The host or adapter selects the profile when it calls /v1/guide.

That can be:

your application code;
a custom Agent host;
a multi-agent orchestrator;
an SDK adapter;
an MCP, AIFS, or native plugin integration.

Aionis does not need to guess the task type. The profile is explicit, auditable, and visible on agent_context.task_context_profile.

Profiles

Profile	Use when	Agent context emphasis
`general`	Default path	No extra task-specific posture.
`coding_verifier`	Coding tasks with tests, builds, validators, or acceptance checks	Keep non-excluded acceptance checks visible; do not complete by narrowing the verifier.
`document_integrity`	File movement, extraction, migration, archiving, or document QA	Preserve original file identity, names, bytes, and output requirements unless transformation is explicit.
`long_qa`	Long-memory QA, factual recall, source-grounded answers	Answer from covered evidence; prefer source spans; rehydrate missing evidence before final.
`multi_agent_handoff`	Planner, worker, verifier, reviewer, or team handoff	Preserve role ownership, current handoff state, and reviewer/verifier boundaries.
`loop_engineering`	Plan, act, validate, repair loops	Preserve plan, iteration index, validator result, repair attempt, and stop reason.

SDK Examples

Coding Verifier


const guide = await aionis.guideAgentContext({
  query_text: "Continue the implementation and run the required checks.",
  mode: "full_power",
  context_mode: "compact_agent",
  task_context_profile: "coding_verifier",
  consumer_agent_id: "worker-1",
  run_id: "run-coding-001",
});

Long QA


const guide = await aionis.guide({
  query_text: "What does the retained evidence say about the migration date?",
  mode: "full_power",
  context_mode: "compact_agent",
  task_context_profile: "long_qa",
  consumer_agent_id: "qa-agent",
  run_id: "run-qa-001",
});

Multi-Agent Handoff


const guide = await aionis.execution.guideForRole({
  agent_id: "reviewer-1",
  team_id: "checkout-team",
  role: "reviewer",
  run_id: "run-handoff-001",
  task_signature: "checkout-migration",
  query_text: "Review the worker handoff and continue from current state.",
  context_mode: "compact_agent",
  guide: {
    task_context_profile: "multi_agent_handoff",
  },
});

HTTP Shape


{
  "query_text": "Continue the current loop from validator feedback.",
  "mode": "full_power",
  "context_mode": "compact_agent",
  "task_context_profile": "loop_engineering",
  "consumer_agent_id": "loop-worker",
  "run_id": "run-loop-001"
}

The response includes:


{
  "agent_context": {
    "task_context_profile": "loop_engineering",
    "prompt_text": "AIONIS_CTX v2\n..."
  }
}

Boundaries

Profiles are not task solutions. Do not put repository-specific fixes, expected answers, verifier shortcuts, or benchmark-specific instructions into a profile.

Profiles are for task posture only. Aionis still decides memory influence through the same governed surfaces:

use_now
inspect_before_use
do_not_use
rehydrate