discuss-subtask

A discussion workflow for difficult or unclear coding subtasks. It develops at least two possible approaches and can optionally use document review or questioning tools.

In plain words
What is it for?
Brainstorming core algorithms, comparing implementation approaches, defining API contracts, and examining subtasks where errors would be costly.
Why use it?
It helps clarify design choices before implementation when the algorithm, API agreement, or cost of mistakes matters. This reduces avoidable rework on high-risk subtasks.

Skill for Claude CodeCodex

Install

Getting it into your agent

One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.

agentmods
npx agentmods add skills/easyinplay/harnessed/subtask
Any agent
npx skills add easyinplay/harnessed --skill subtask
Clone the repo
git clone --depth 1 https://github.com/easyinplay/harnessed

Made for: Claude Code, Codex.

Per session 133 Skills are progressive disclosure: only the name and description are preloaded; the body loads when the skill is used.
When invoked 1,028 The whole file, excluding the scripts and references it only reads on demand.
Security scan A 0 findings. Scan, not verified.
Origin original No closer match found in the catalogue.
Token cost

What it costs to keep this loaded

Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.

ModelPer sessionOnce invoked
Fable 5 $0.00133 $0.01028
Opus 5 $0.00067 $0.00514
Sonnet 5 $0.00027 $0.00206
Haiku 4.5 $0.00013 $0.00103

Measured 2d ago against content hash 796e3de952be, method: parsed. Prices are Anthropic first-party input rates as of 2026-08-30, from the pricing page.

Security

Grade A, and why

discuss-subtask scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 2d ago.

A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.

Nothing flagged

None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.

workflows/discuss/subtask/SKILL.md · 84 lines

How it starts

The opening of the file, as written. The whole thing — 84 lines — stays where its author put it; the contents beside it link to each section on GitHub.

discuss-subtask workflow (v3)

Overview

1-phase sub-workflow mapping CLAUDE.md "Stage ①.c 子任务层 — superpowers brainstorming" onto harnessed runtime (Phase v3.0-3.4 W0.3 — D-04 Stage ① Discuss 三层 + D-05 phase.invokes_tools conditional fire + Pattern A sub-workflow ship)。

phase id upstream model capability / invokes_tools gate
1 01-brainstorm superpowers opus {{ capabilities.superpowers-brainstorming.cmd }} + conditional invokes_tools[grill-with-docs, grill-me] gate: judgments.subtask-gate.brainstorming.fires

Capability refs

Sister workflows/capabilities.yaml entries:

  • superpowers-brainstorming — Bucket 4 (impl: superpowers, cmd: superpowers:brainstorming)
  • grill-with-docs — Bucket 1 mattpocock (impl: mattpocock-skills, cmd: /grill-with-docs)
  • grill-me — Bucket 1 mattpocock (impl: mattpocock-skills, cmd: /grill-me)

Gate ref

Sister workflows/judgments/subtask-gate.yaml:

  • brainstorming.firessubtask.approaches >= 2 or subtask.core_algorithm == true or subtask.has_api_contract == true or subtask.error_cost == 'high'

Conditional tool invocation (D-05 invokes_tools)

  • phase.spec_ambiguous == true → fire grill-with-docs
  • phase.spec_ambiguous == true AND phase.no_docs == true → fire grill-me

Invocation

  • Slash command: /discuss-subtask <text> (after harnessed setup)

Routing rules

跳过条件 (sister CLAUDE.md "子任务层 ❌ 跳过"):

  • 常规 CRUD / 跟现有模式一致
  • 单一明显实现 (< 20 行 / 一个文件)
  • 标准库直接调用
  • bug 修复且已知根因

How to invoke

Clarification needs real user dialogue, so run this stage directly in THIS session — do NOT spawn it, and do NOT improvise: follow these steps so the locked spec is persisted for the execution stages.

  1. Evaluate the clarification criteria for "$ARGUMENTS":
    • Strategic — new feature / milestone / unclear business scope → gstack /office-hours + /plan-ceo-review
    • Phase — ≥2 open implementation decisions / unclear cross-phase API contract → GSD /gsd-discuss-phase
    • Subtask — ≥2 distinct approaches / core algorithm / API contract design / high error cost → superpowers brainstorming
  2. For each layer that fires, hold the dialogue with the user (use AskUserQuestion for option-style decisions) and lock every open decision.
  3. Transparent-skip layers that don't fire — state which were skipped and why.
  4. Persist the locked decisions to .planning/ via planning-with-files (findings.md / task_plan.md).

Read the full file on GitHub · 84 lines

Files

What ships with it

2 files beside SKILL.md in the same directory: the scripts, references and assets a skill reads on demand. Not counted in the per-session cost; read them before you install if any of them is executable.

Changes

What this file has done since we first saw it

Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.

  1. 2d ago First seen · 84 lines · 133 tokens per session scan A 796e3de952be

Subscribe to this mod's changes

discuss-subtask is a skill published in the GitHub repository easyinplay/harnessed (2 stars, last pushed 6d ago), licensed Apache-2.0. It adds 133 tokens to every session and 1,028 once invoked, about $0.0007 per session on Opus 5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-31.