Skip to content
Use Cases
AGH RuntimeUse Cases

Release Readiness Sweep

Use AGH tasks and sessions to run release checks with durable ownership and completion evidence.

Audience
Operators running durable agent work
Focus
Use Cases guidance shaped for scanability, day-two clarity, and operator context.

A release sweep is a set of checks that should not disappear into chat: docs, build health, integration status, migration notes, and operator risk. AGH models that work as tasks, task runs, and attached sessions.

The model is simple:

Runtime objectMeaning in this use caseTypical owner
TaskThe release concern, such as "Verify release documentation".Operator creates and reviews it.
Task runOne attempt to perform that task, with queue, claim, and close state.Operator or agent claims and closes it.
SessionThe durable agent process that reads files, runs commands, and reports evidence.Agent produces the work evidence.

Setup

Start with a registered workspace:

agh workspace add /Users/you/project --name project

Create one task per release concern. Keep the title concrete enough that an agent can claim it without extra context.

agh task create \
  --scope workspace \
  --workspace project \
  --title "Verify release documentation" \
  --description "Check docs, CLI examples, and known release claims for drift." \
  -o json

Flow

Rendering diagram...

A release readiness sweep moves from task creation through a claimed run, attached session evidence, and an auditable completion decision.

In a small team, the same human operator may enqueue, claim, attach, and close the run. In a more autonomous setup, an agent can claim and close runs while the operator reviews the durable evidence.

agh task run enqueue task_123 -o json
agh task run claim run_123 -o json
agh session new --workspace project --agent reviewer --name release-docs -o json
agh task run attach-session run_123 --session sess_1234 -o json
agh session prompt sess_1234 "Verify release documentation. Report mismatches, missing examples, and commands used."

When the evidence is clear, close the run:

agh task run complete run_123 \
  --result '{"status":"ready","evidence":["agh session history sess_1234"]}' \
  -o json

If the check finds a blocker, fail the run instead:

agh task run fail run_123 \
  --error "CLI examples do not match generated reference" \
  --metadata '{"session":"sess_1234"}' \
  -o json

Evidence to keep

EvidenceCommandWhy it matters
Task definitionagh task get task_123 -o jsonShows scope, workspace, owner, channel, and metadata.
Run stateagh task run list task_123 -o jsonShows whether work is queued, claimed, running, or closed.
Attached sessionagh task run list task_123 --session sess_1234 -o jsonConnects agent work to the task run.
Session historyagh session history sess_1234Preserves the reasoning and commands behind the result.

Useful release sweep tasks

Task titleWhat the agent should prove
Verify CLI examplesEvery command in edited docs exists in generated CLI reference.
Check OpenAPI route claimsDocs distinguish generated schema coverage from implemented routes.
Review provider and agent supportPublic copy lists only implemented provider IDs and agent adapters.
Exercise first-run docsA new operator can install, start the daemon, create a workspace, and start a session.
Collect unresolved release risksAny gap has a task id, owner, and visible next action.

Failure path

SymptomFirst checkLikely fix
No one can claim the runagh task run list task_123 -o jsonConfirm the run was enqueued and is not closed.
Session is doing unrelated workagh task run list task_123 --session sess_1234 -o jsonAttach the intended session or start a new one.
Result JSON is too vagueagh session history sess_1234Re-prompt for concrete evidence before closing.
Release concern keeps recurringTask history and run metadataSplit the concern into smaller tasks.

On this page