| name | agent-view |
| description | Diagnose Claude Code Agent View readiness safely inside the minmaxing harness. This route is static readiness and troubleshooting only; open the live Agent View TUI manually with claude agents. |
| argument-hint | ["status/troubleshoot"] |
| disable-model-invocation | true |
/agent-view
Diagnose Claude Code Agent View readiness for:
$ARGUMENTS
Agent View is Claude Code's local background-session dashboard. In this harness,
/agent-view is a readiness and troubleshooting route only. It does not open
the live TUI, does not dispatch background sessions, and does not run
claude --bg.
To open the live native Agent View TUI manually, run this from a shell:
claude agents
Native Boundary
Agent View is separate from:
/remote-control: continues a local Claude Code session from claude.ai/code
or mobile. Agent View is a terminal dashboard for local background sessions.
/agents: the interactive Claude Code subagent panel. claude agents opens
Agent View from the shell.
- Subagents: subagents report back to the parent conversation. Agent View
sessions report only to the operator.
- Agent teams: experimental peer coordination. Agent View sessions are
independent unless the operator or another harness contract coordinates them.
/parallel: minmaxing packet orchestration. Agent View can help an operator
monitor independent sessions, but it does not satisfy packet DAG, ownership
matrix, sidecar, aggregation, /introspect, or /verify requirements.
/goal-mode: readiness and safety guidance for native Claude Code /goal.
Agent View can monitor independent sessions where the operator manually uses
/goal, but Agent View plus /goal is still not /parallel and still does
not satisfy parent verification.
Do not build a custom dashboard, websocket bridge, browser backdoor, MCP control
plane, or API-key fallback for this route.
Effectiveness Yield Policy
Use Agent View for effectiveness only when it increases operator awareness over
independent top-level lanes. The high-yield pattern is:
main harness session = orchestrator, judge, and verifier
Agent View sessions = optional independent evidence lanes
artifacts = bridge back into /parallel, /workflow, or /hiveworkflow
Do not optimize for the number of running sessions. Optimize for verified
evidence per operator minute, reduced blocked-session time, fewer context
switches, and cleaner stop/attach decisions. Three scoped sessions with clear
artifacts are better than ten busy sessions that produce merge conflicts,
unverified claims, or stale context.
Agent View is appropriate when each background session has explicit scope,
owned files or read-only surfaces, a required output artifact, stop conditions,
and a parent verification path. Move on to /opusworkflow, /parallel, or
local work when the task is tightly coupled, touches shared files, involves
secrets/security-sensitive authority, or needs one shared reasoning loop.
If a background session uses native /goal, the goal condition must inherit the
same packet contract: owned files or worktree, forbidden paths, expected
artifact, command evidence, stop bound, and parent verification. A /goal
success message is not a verified worker result.
Prerequisites
- Claude Code must be
v2.1.139 or newer.
- Check the local CLI with
claude --version.
- Agent View can be disabled with the shared
disableAgentView setting or
CLAUDE_CODE_DISABLE_AGENT_VIEW.
- Background sessions read settings and permission mode from their target
directory, so this repo's trusted-local
bypassPermissions posture is a real
unattended authority risk.
- Background sessions are local, consume subscription quota independently, are
stopped by sleep or shutdown, and can use
.claude/worktrees/ for file edits.
- Worktrees created by Agent View are not a minmaxing merge strategy. Keep
/parallel aggregation and parent verification as the source of truth for
governed packet work.
Checks
Run the static doctor:
bash scripts/agent-view-doctor.sh --static --json
Run the fixture gate:
bash scripts/agent-view-smoke.sh --fixtures
These checks do not open Agent View, do not dispatch background sessions, do not
read ~/.claude/jobs, do not inspect transcripts, and do not prove paid-account
runtime availability. They only prove that the committed harness documents and
validates Agent View safely.
Runtime Use
Manual runtime check after upgrading Claude Code:
claude --version
claude agents --help
claude agents
Only the operator should open the TUI. A manual runtime result must be reported
as manual evidence, never as CI/static proof. Do not dispatch mutating background
sessions unless the task has explicit ownership, verification, and rollback
boundaries.
Static Artifact
scripts/agent-view-doctor.sh --static --json emits:
{
"artifact_type": "agent-view-readiness",
"runtime_agent_view_started": false,
"runtime_proof_status": "blocked_by_cli_version",
"minimum_required_version": "2.1.139",
"operator_boundary": "manual_operator_monitor_only"
}
Allowed runtime_proof_status values:
blocked_by_cli_version
not_run_static_only
ready_static_only
Static artifacts must not claim that Agent View opened, that a background
session ran, that a paid Claude account connected, or that background sessions
replace minmaxing verification.
Anti-Patterns
- Treating Agent View rows as verified worker results.
- Treating
claude agents as /parallel, /hive, or /opusworkflow.
- Claiming ten background sessions means 10x throughput.
- Claiming background sessions are cloud durable.
- Reading
.env, local Claude settings, ~/.claude/jobs, transcripts, or
session state for a static readiness answer.
- Hiding the
bypassPermissions risk for unattended background sessions.