Threnody task

Skill timjensgrossinger/threnody/skills/threnody-task

Threnody normal orchestration via plan_task, decompose_task, and fleet_plan. Use for multi-file tasks, wave-based host Agent execution, or /fleet planning without a full execute_swarm contract.From its SKILL.md

Install

npx -y skills add timjensgrossinger/threnody --skill threnody-task

Assembled from the repository path, not quoted from the project. Check it against their README if it does not work.

One thing to look at

2 stars2 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.

SKILL.md

4.2 KB, as published. Nobody here has run it

Threnody task orchestration

Execution phase after planning — see threnody-plan for plan-only vs plan-then-execute routing.

Use this skill for planning + host wave execution without the full swarm persistence contract (swarm_id, budget preview, resume checkpoints).

When to use

Use `threnody-task`	Use `threnody-swarm` instead
`plan_task` / `decompose_task` / `fleet_plan`	`execute_swarm`
Host runs `host_spawn_waves`	Need `swarm_id`, telemetry, resume
No budget preview token flow	Budget preview + `preview_token` confirm
Single planning pass	Coordinator star rounds (delegate mode)

Workflow

Prefer threnody-plan first unless the user already approved a wave plan.
route_task(task) — tier, execution_hint, optional single host_spawn.
Decompose (multi-concern work):
- Prefer decompose_task(task) (alias of plan_task).
- Or plan_task(task) directly.
Read host_spawn_waves from the plan response.
Execute waves in order via host Task/Agent:
- Agents within one wave may run in parallel.
- Respect wave ordering (later waves wait for earlier dependencies).
Reporting — check learning_report_contract.report_mode:
- batch (default): do NOT call report_host_wave per worker wave. Spawn waves; learning is captured automatically (PostToolUse hook) or passed in the single terminal call (learning_capture=model).
- inline (legacy): report_host_wave(host_run_id|plan_run_id, wave, workspace_root, agents[]) after each wave with task_id, spawn_id, success, touched_files, output_excerpt.
Mid-run: expand_host_plan(discovered_files=[...]) when wave 1 discovers files not in the initial plan.
Final: report_host_swarm_complete(outcome=...) once (batch imports + finalizes). Check finalize.swarm_outcome.
Optional: fleet_plan(task) when you want ready-made fleet command strings per wave.

Rules

Host-native heuristic planning fans out one agent per file for webapp/fullstack intent or explicit paths; single-file tasks stay one agent.
Keep handoffs compact: use host_spawn_waves as the executable manifest and avoid inlining duplicate full plans/prompts into user-facing summaries.
When host_spawn_waves or host_execution_contract: spawn_subagents is present, spawn one Task/Agent per agent — never use direct Write/Edit on planned target_files.
Do not call execute_subtask for same-host work — use host_spawn entries.
Do not follow route_task direct_edit while a plan handoff is active; use routing_guard and host_run_id from the plan response.
Utility delegation only when delegation_utilities_enabled: true (see threnody-routing).
For frontend + backend + API in parallel, use the contract-first pattern in threnody-fullstack.
For broad review work, use threnody-fast-review; reserve threnody-swarm-review for explicit deep/specialist review.

Example

route_task(task="Refactor auth across services and UI")
decompose_task(task="Refactor auth across services and UI")
→ host_spawn_waves: [
     { wave: 1, parallel: false, agents: [...] },
     { wave: 2, parallel: true, agents: [...] }
   ]
→ Spawn each agent via host Task tool; wait for wave N before wave N+1.
→ (batch mode, default) report once at the end:
   report_host_swarm_complete(
     host_run_id,
     outcome="accepted",
     workspace_root="<from plan handoff>",
     agents=[{                       # only when learning_capture=model
       task_id, spawn_id, success,
       touched_files: ["relative/path.py"],
       output_excerpt: "short agent summary",
     }, ...],
   )

MCP tools

route_task, plan_task, decompose_task, fleet_plan
report_host_wave, report_host_swarm_complete, expand_host_plan, inspect_swarm
validate_routing_guard (guarded hosts before edits)