Enhancement: OpenClaw plugin — expose full task CRUD via tools/commands

[Tweakzx]

Problem

The current OpenClaw plugin (plugins/openclaw-agentflow/index.ts) exposes limited capabilities:

Commands:

• agentflow.run — run one task (via mock adapter)
• agentflow.help — show usage guidance

Tools:

• agentflow_status — read board summary
• agentflow_capabilities — show capability map

Missing:

• No task creation tool/command
• No task status update tool/command
• No run history query tool
• No audit trail query tool
• No way to specify a real adapter from within the plugin

This means the OpenClaw agent cannot fully manage AgentFlow tasks through the plugin — it still needs to shell out to CLI or call HTTP endpoints directly.

Suggested Additions

New Tools

Tool ID	Description	Input
agentflow_create_task	Create a task in a project	{ project, title, description?, priority?, impact?, effort?, source?, external_id? }
agentflow_move_task	Move a task to a new status	{ task_id, to_status, note? }
agentflow_run_task	Trigger a run on a specific task	{ task_id, adapter?, agent? }
agentflow_task_detail	Get full task detail with runs and history	{ task_id }
agentflow_recent_runs	List recent runs for a project	{ project?, limit? }
agentflow_audit	Query audit trail	{ project?, limit? }

New Commands

Command ID	Description	Input
agentflow.create	Create a task	{ project, title, ... }
agentflow.move	Move task status	{ task_id, to_status }
agentflow.detail	Show task detail	{ task_id }
agentflow.audit	Show audit trail	{ project?, limit? }

Implementation Notes

The plugin already has runAgentflow() helper that calls the CLI. The new tools/commands follow the same pattern:

api.registerTool?.({
  id: "agentflow_create_task",
  description: "Create a task in AgentFlow",
  inputSchema: { type: "object", properties: {
    project: { type: "string" },
    title: { type: "string" },
    description: { type: "string" },
    priority: { type: "number" },
    impact: { type: "number" },
    effort: { type: "number" },
    source: { type: "string" },
    external_id: { type: "string" },
  }, required: ["project", "title"] },
  async run(input) {
    const args = ["add-task", "--project", input.project, "--title", input.title];
    if (input.description) args.push("--description", input.description);
    if (input.priority) args.push("--priority", String(input.priority));
    // ... etc
    const result = await runAgentflow(args);
    return { content: result.stdout };
  }
});

Why this matters

• Completes the agent → AgentFlow control loop (currently read-only + one write action)
• Enables the tracking skill (Feature: OpenClaw skill for sub-agent task tracking via AgentFlow #6) to work entirely through the plugin rather than raw HTTP calls
• Makes AgentFlow a true "native" capability of OpenClaw rather than an external dependency

Depends on

• None (uses existing CLI commands), but Feature: Add POST /api/tasks endpoint for programmatic task creation #5 would enable a pure-HTTP implementation alternative

[Tweakzx]

Triage decision (2026-03-31): Keep, P2.

Why worth doing:

• Full CRUD tools improve plugin ergonomics and agent autonomy.
• But this is less critical than making execution real first.

Recommended split:

• Phase A: create/move/detail tools.
• Phase B: run history + audit query tools.

[Tweakzx]

Implementation update (2026-03-31): OpenClaw plugin now exposes broader task CRUD/query capabilities.

Delivered in plugin:

• New commands:
  • agentflow.create
  • agentflow.move
  • agentflow.detail
  • agentflow.audit
• New tools:
  • agentflow_create_task
  • agentflow_move_task
  • agentflow_task_detail
  • agentflow_recent_runs
  • agentflow_audit
• Capability document updated with new command/tool surface

Delivered in CLI (supporting plugin surface):

• task-detail --task-id <id>
• recent-runs --project <project> --limit <n>
• audit --project <project> --limit <n>

Validation:

• Added CLI smoke test coverage for new commands
• Full suite green: 33 passed, 4 skipped