Skip to main content
OrchestKit v7.88.0 — 109 skills, 37 agents, 196 hooks · Claude Code 2.1.139+
OrchestKit
Skills

Agents View

Wraps the Research Preview `claude agents` CLI (CC 2.1.139+) and `claude plugin details ork` for live observability of parallel agent sessions. Surfaces running/blocked/done state, per-session token cost, and the 188-hook plugin's runtime footprint. Use when debugging multi-agent workflows, projecting cost on a long-running orchestration, or auditing which hooks fired during a run.

Command low
Invoke
/ork:agents-view

/ork:agents-view — Parallel Agent Observability

Research Preview — UI may change before GA. Pinned to CC ≥ 2.1.139.

1. What it does

Thin wrapper around two CC 2.1.139 CLI surfaces:

CommandPurpose
claude agentsResearch Preview — lists parallel agent sessions with state (running / blocked / done), elapsed time, token usage, and per-session cost projection.
claude plugin details orkPlugin runtime footprint — number of hooks loaded, skills available, agents registered, and which hooks fired in the current session.

Flags:

  • --plugin — show only claude plugin details ork (skip the agent list).
  • --watch — re-poll every 2s until interrupted (uses claude agents --watch under the hood).
  • --json — emit raw JSON for piping into downstream skills (e.g. budget gates).

2. When to use

ScenarioWhy
Parallel-agent debugging — a /ork:explore, /ork:brainstorm, or /ork:implement run feels stuckclaude agents shows which child session is blocked, on what, and how long. Faster than tailing logs.
Observability for the 188-hook pluginclaude plugin details ork reports per-hook invocation counts; useful when a hook is suspected of slowing the loop or firing more often than expected.
Per-session cost projectionThe Research Preview view ships token + dollar projections per agent. Use before kicking off a long xhigh-effort run.
Post-mortem after an xhigh orchestration--json snapshot lets you diff hook fire counts and agent durations across runs.

3. Sample output

$ /ork:agents-view
claude agents — Research Preview (CC 2.1.139)
─────────────────────────────────────────────
ID      AGENT                          STATE     ELAPSED   TOKENS   COST(proj)
a1f3    ork:explore                    running   00:04:12  142.3K   $0.42
a1f4    └─ ork:backend-system-architect blocked   00:01:08   34.1K   $0.09   (waiting: db-schema)
a1f5    └─ ork:frontend-ui-developer   running   00:02:55   71.8K   $0.21
a1f2    ork:code-quality-reviewer      done      00:00:48   18.4K   $0.05
─────────────────────────────────────────────
4 sessions · 2 running · 1 blocked · 1 done · projected total $0.77

claude plugin details ork
─────────────────────────
skills:   107 loaded · 12 invoked this session
agents:   37 registered · 4 spawned this session
hooks:    188 total (120 global · 46 agent · 22 skill) · 31 fired
top hooks: skill/repo-structure-indexer (8×) · global/stale-import-detector (5×)

--json mode emits the same data as a single JSON object — see claude agents --help for the schema (Research Preview, subject to change).

The following skills all spawn background agents that surface in this view — /ork:agents-view is the canonical observability surface for any of them:

  • ork:brainstorm — multi-agent ideation; check --watch while it runs to see which angle is blocking.
  • ork:audit-full — long-running parallel audit; cost projection helps decide whether to let it finish.
  • ork:explore — 4–5 parallel explorer agents; claude agents shows which explorer is still searching.
  • ork:dev — boots agent-browser as a background session; appears alongside Claude agents in the list.

Anti-patterns

  • NEVER parse the human-readable table format — use --json for any programmatic consumer. The Research Preview output schema is documented; the table layout is not.
  • NEVER assume claude agents is available on CC < 2.1.139 — fall back gracefully and surface the install hint.
  • NEVER call this inside a tight loop without --watch; each invocation pays cold-start latency on the CLI.
Edit on GitHub

Last updated on