agentleFS
gymcoach-appgymcoach-app/gymcoach/.claude/skills/ideate/SKILL.md

ideate

Skill187 starsChanged 22 days agoMITView on GitHub

When the backlog is empty, manufacture well-scoped PRODUCT feature ideas - bounded feature research plus brainstorming grounded in the product vision and the captured competitor insights - and file them as crisp, single-PR issues so the Issue -> PR loop never starves for ideas. Use only when ship and triage have nothing and there is no open actionable issue. Strictly token-bounded; never launches the heavy multi-agent deep-research workflow.

37 other projects use a version of this.

Adopted by

Most recently taken up by

Used with

Projects with it also use daily-arxiv, exp-eval and 10 more.

Other files people use alongside this one.

Files found in the same public projects as a version of this one, more often than they turn up anywhere else. Each project counts once, large collections count for less, and this file's own repository is left out.
  • LambenthanSkill

    daily-arxiv

    每日从 arXiv 拉取新论文,过滤相关性,auto-ingest 高优先级论文,检测 SOTA 更新

    Also in 5 projects that use this

  • LambenthanSkill

    exp-eval

    实验判决门:Review LLM 独立评判实验结果 → 4 种判决路径 → 自动更新 claims confidence、ideas status、graph edges

    Also in 5 projects that use this

  • skyllwtSkill

    exp-pilot-eval

    Pilot result evaluation — read pilot results, apply success criteria, update idea page (pilot_result, failure_reason if failed), generate PILOT_VERDICT_REPORT. Called by /ideate Phase 5 after /exp-pilot-run.

    Also in 4 projects that use this

  • skyllwtSkill

    exp-pilot-run

    Pilot experiment execution — read Pilot Spec YAML, write pilot code, run experiment(Confirm with the user before operation and require the applicant to conduct manual inspection), return results. Called by /ideate Phase 5. Does NOT modify wiki pages or judge pass/fail.

    Also in 4 projects that use this

  • phanghonghaoSkill

    poster

    HTML/CSS A4 海报生成。从沙龙/活动 Markdown 文件生成科技风 A4 海报 PDF/PNG。Use when user mentions: poster, 海报, 海报生成, make poster, generate poster, 活动海报, 沙龙海报.

    Also in 4 projects that use this

  • +6 more from Lambenthan
  • +1 more from skyllwt

The file

We split this skill into the rules it gives an agent

9 rules. 0 are common, 7 are distinctive, and 2 are very specific to this project.

  • 0

    Common

    Most of the 33 edited versions we read keep these.

  • 7

    Distinctive

    This author's own choices. Worth a look.

  • 2

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Only when the loop is genuinely starved of product work, in this order:

    line 21
  2. Distinctive:

    no ready PR to ship, AND

    line 22
  3. Distinctive:

    If there is already an open actionable issue, STOP - implement that first.

    line 26
  4. Distinctive:

    Default to brainstorming from context you already have (the repo, the research memory, the roadmap).

    line 32
  5. Very specific:

    Anti-flood: do not run if there are already >= 3 open enhancement/idea issues - the backlog is fed.

    line 37
  6. Distinctive:

    Never re-file a shipped or rejected idea.

    line 75
  7. Very specific:

    Confirm starvation and the anti-flood cap: If not starved, or there are already >= 3 open enhancement/idea issues, STOP and report - do not …

    line 79
  8. Distinctive:

    Not starved, or an open actionable issue exists -> STOP (implement first).

    line 96
  9. Distinctive:

    No genuinely new, single-PR, non-duplicate idea -> file fewer or none.

    line 100

“Common” is measured: kept by at least 6 in 10 of the 33 edited versions of this file we could read. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

ideate

triage manufactures code-health work (tests, small bugs, deps); ideate manufactures product work when even triage comes up dry, so the loop keeps pushing the product toward the goal of being the most complete self-hosted AI training/fitness app - not just keeping it tidy. The output is issues, not code: crisp, single-PR feature ideas a later implement-issue run picks up unattended. Read CLAUDE.md and docs/loops/07-autonomy.md first.

This is the lightweight, recurring cousin of the one-off deep-research workflow. It is deliberately cheap: it runs often (whenever the loop would otherwise idle), so it must cost a tiny fraction of that workflow.

When to run

Only when the loop is genuinely starved of product work, in this order:

  • no ready PR to ship, AND
  • triage found no trusted code-health issue, AND
  • there is no open, actionable, trusted issue to implement.

If there is already an open actionable issue, STOP - implement that first. Never pile new ideas on top of unstarted ones.

Hard budget (the whole point is bounded ideation)

  • Never launch the heavy deep-research workflow (the ~100-agent one) or anything close.
  • Default to brainstorming from context you already have (the repo, the research memory, the roadmap). A web check is optional and capped at ~6 searches, used only if it adds something the captured research does not already cover.
  • Spawn at most ONE helper subagent if you need one; prefer doing it inline.
  • File at most 3 issues per run.
  • Anti-flood: do not run if there are already >= 3 open enhancement/idea issues - the backlog is fed; let implement-issue drain it first. One productive run per starved cycle; once you file issues, you are done until they drain.

Where ideas come from (grounded, not random)

  1. The vision. Aim: the most complete self-hosted AI training/fitness app. The operator broadened the scope (2026-06-11) beyond pure hypertrophy: strength, conditioning/cardio, endurance, mobility, and general fitness are all in scope. Concretely, the CSV importers (lib/import/strong-csv.ts, lib/import/hevy-csv.ts) currently SKIP cardio rows and nothing in the app tracks duration/distance work - that axis is now fair game. Ask "what would a serious lifter, a coach, or a leading competitor expect that we do not have yet?"
  2. The captured research (Memory: research-product-direction.md) and its wedge - AI that advises within the user's program, explains why, and respects data ownership. Mine it for not-yet-built items.
  3. The current product. Skim app/ routes, lib/, components/ to find concrete gaps and the natural next slice of what already shipped (e.g. readiness now exists -> what uses it next).
  4. The README roadmap - unchecked items.
  5. Optional bounded web check (<= ~6 searches) for what users want / what is missing in AI training/fitness apps lately - only if it beats the memory. No heavy fan-out.

What a good idea-issue looks like

  • Single-PR-sized and additive-friendly - scoped so implement-issue can finish it under the green-gate. No epics: split a big idea into the smallest valuable first slice.
  • Encodes a sensible DEFAULT product decision, so it is concrete work, not "needs a product call". Turning ambiguity into a defensible spec is the job.
  • Title + Context (the user value and where it lives, with file paths) + Acceptance criteria (a checkable list).
  • Risk-scaled, not additive-only (operator directive 2026-06-10). Complex ideas - new tables, data-safe migrations, LLM-output-contract changes, multi-surface features - are in scope when they are a clear product plus; the issue must then name the reinforced non-regression controls that apply (full gate, tests at every touched layer, contract tests, multi-lens review - see the charter's "Complex features" section). Still stop-for-human: destructive data migrations, auth/security changes, major dep bumps.
  • Labels: enhancement and idea.
  • De-dupe against open AND recently-closed issues and docs/loops/ideas-backlog.md. Never re-file a shipped or rejected idea.

Procedure

  1. Confirm starvation and the anti-flood cap. If not starved, or there are already >= 3 open enhancement/idea issues, STOP and report - do not generate.
  2. Ground. Skim the current features, the research memory, the roadmap. Optionally one small web check within the budget.
  3. Brainstorm a short candidate list; rank by (user value x fit-with-vision) / implementation size. Prefer high-value single-PR slices; complexity is acceptable when the value justifies it and the controls scale with it.
  4. File the top up-to-3 non-duplicate, single-PR candidates as crisp issues with a sensible default spec + acceptance criteria: gh issue create --label enhancement --label idea --title "..." --body "...".
  5. Record one line per idea in docs/loops/ideas-backlog.md (proposed / shipped / rejected) so future runs do not repeat it. (This is a read/write doc edit; commit it on a task branch like any change, never on main.)
  6. Report the new issue numbers + one line each, and what you deliberately did not file.

Stop conditions

  • Not starved, or an open actionable issue exists -> STOP (implement first).
  • = 3 open enhancement/idea issues -> STOP (anti-flood).

  • 20 or more open issues -> STOP (open-issue cap, operator directive 2026-10-19); an idea worth keeping goes into the parked backlog #484, not onto the open list.
  • No genuinely new, single-PR, non-duplicate idea -> file fewer or none. "Idea backlog healthy" is a valid, good outcome; never invent low-value busywork to fill the tracker.
  • An idea that genuinely needs the heavy deep-research workflow -> out of scope here; note it for a human instead.

What success looks like

A steady trickle of crisp, vision-aligned, single-PR feature issues that appear only when the loop would otherwise idle, never faster than implementation drains them, at a tiny fraction of the one-off deep-research cost. The loop never starves for product direction, and the product keeps moving toward "the most complete AI fitness app" - while staying token-bounded and inside every guardrail (the filed issues are still subject to the green-gate, the subagent challenge, and the stop-for-human list when implemented).

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.