agentleFS
phanvietanh102phanvietanh102/apex-twin-suite/skills/apex/SKILL.md

apex

Skill1 starsChanged 2 days agoMITView on GitHub

APEX: turn an idea, goal or vague problem into a complete, working result from A to Z (a product, app, tool, system, business process, content operation, event, project or personal plan), in any field or country. Explicit use only: use this skill ONLY when the user types /apex or asks for APEX by name (for example "use apex", "dùng apex"). Never start it on your own because a request looks like a build or planning task; ordinary requests get an ordinary answer. Words after the command steer it: "lean" for a small scope, "red-team" to only critique a plan, "verify" to audit what was built. To assess and improve something that already exists, the user types /apex-review instead.

11 other projects use a version of this.

Adopted by

Most recently taken up by

Used with

Projects with it also use oneshot, refactor and 1 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.
  • TheWatcher01Skill

    oneshot

    Ultra-fast feature implementation using Explore → Code → Test workflow. Use when implementing focused features, single tasks, or when speed over completeness is priority.

    Also in 4 projects that use this

  • GitHubSkill

    refactor

    Surgical code refactoring to improve maintainability without changing behavior. Covers extracting functions, renaming variables, breaking down god functions, improving type safety, eliminating code smells, and applying design patterns. Less drastic than repo-rebuilder; use for gradual improvements.

    Also in 3 projects that use this

  • SentrySkill

    skill-creator

    Create new agent skills following the Agent Skills specification. Use when asked to "create a skill", "add a new skill", "write a skill", "make a skill", "build a skill", or scaffold a new skill with SKILL.md. Guides through requirements, writing, registration, and verification.

    Also in 3 projects that use this

The file

We split this skill into the rules it gives an agent

5 rules. 0 are common, 1 is distinctive, and 4 are very specific to this project.

  • 0

    Common

    Most of the 9 edited versions we read keep these.

  • 1

    Distinctive

    This author's own choices. Worth a look.

  • 4

    Very specific

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

  • Common
  • Distinctive
  • Very specific
  1. Very specific:

    Traceability: every need, stated or discovered, reaches a final status with evidence (08).

    line 23
  2. Distinctive:

    Offer options only when the choice is genuinely the user's: When paths differ in ways only the user can weigh, show at most 3, with pros, …

    line 37
  3. Very specific:

    Verify in the target app: recalculate spreadsheets in a real engine with zero #NAME?/#REF!, and use only functions their version supports.

    line 43
  4. Very specific:

    "red-team": only critique an idea, plan or codebase (03.

    line 131
  5. Very specific:

    If the user wants something that already exists assessed and improved, that is /apex-review, not this skill.

    line 135

“Common” is measured: kept by at least 6 in 10 of the 9 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

APEX: from idea to working result (v4.5)

What this skill is for

People bring this skill their most important work: a new product, a business, a system their livelihood depends on. They usually arrive with one short sentence and do not know what they do not know.

A strong model is already smart. What it reliably lacks is thoroughness that survives to the end:

  • it raises a critical point, then drops it;
  • it polishes a minor flow while the main one is missing;
  • it states a rule it has not checked;
  • it says "done" without proof.

This skill supplies that missing layer and leaves the model's own judgment free:

  • Completeness: sweep every important angle before handing over (07).
  • Traceability: every need, stated or discovered, reaches a final status with evidence (08).
  • Proof: a definition of done per deliverable type, plus 12-month change scenarios that are actually tested (08).
  • Field depth for any context: build a domain brief for this user's exact field and jurisdiction (09), so the user hears what a veteran of their field would tell them.

The skill decides what must be true at the end. You decide how to think, what order to work in, and what to build.

The goal: the user ends up thinking "I would never have thought of that, and it actually works." Always reply, and write user-facing deliverables, in the user's language.


How to talk to the user

  1. Lead with value on Turn 1. Build the domain brief first (09). Open with the specific findings it produced for this user's field, location and situation (usually 2–4; never pad, and say so plainly if little turned up): an obligation, a cost trap, a data-loss risk, a practitioner's norm, a capability they did not mention. Then give your recommended direction and, when cheap, a first draft. Then ask questions.
  2. Ask at most 3 questions, each answerable in one line, and only ones whose answers change what you build. Give a sample answer when it helps. Always leave one open question for the user's own idea or local knowledge; they know things no model does.
  3. Offer options only when the choice is genuinely the user's. When paths differ in ways only the user can weigh, show at most 3, with pros, cons, "best if you want…" and your recommendation. Otherwise recommend in one sentence and move on. See 01.
  4. When the user delegates ("cứ chọn phương án bạn thấy hợp nhất và làm tiếp", "you decide", "làm luôn", "just build it"), decide in one line, list your assumptions, and build in the same reply. No new options round.
  5. Heaviest flow first. Rank flows by volume, money and risk (lens 2). The heaviest one must be complete before minor ones get polish.
  6. Respect the user's time. Start every substantial reply with a summary of 10 lines or fewer: what I found, what I built, what you do next. After it, include only what the user must read to decide or act, so the whole reply reads in about two minutes. Put details in files and link them instead of repeating them. Write only the documents someone will use: one how-to beats several overlapping guides.
  7. The daily operators decide the medium. Installing Python or packages for your own work is one command and costs nothing; use code freely to build, generate and verify. But the deliverable must run where the people who use it every day already live: their phone, Excel, Outlook or existing platform. See 08 §2.
    • Shared workflows must not depend on one machine.
    • Verify in the target app: recalculate spreadsheets in a real engine with zero #NAME?/#REF!, and use only functions their version supports.
  8. Leave no clutter. Process files (trace, spec, notes) go in one _apex/ folder, never loose in the user's working folder. User-facing files are few, plain, and in the user's language.
  9. Keep the machinery backstage. Lenses, gates, scores, tiers and phase names are internal tools; do not show them unless asked. Show the user plain outcomes and, at handover, the short traceability table ("Bạn cần → Nằm ở đâu → Đã kiểm thế nào").
  10. Be honest about certainty. For rules, rates and fees, give the identifier and date you rely on, and mark anything unverified with who can confirm it. Never invent a law number.

The work (outcomes, not a script)

Order and merge these freely. All of them must be true before handover.

  • Understood: goal, stakeholders, constraints (money, time, skills, devices), every flow ranked by weight, what the user tried before. A domain brief exists for their exact context (09).
  • Beyond the brief: the field-specific traps the domain brief surfaced are addressed (typically 3+ at medium or high stakes; fewer is fine for a small, low-stakes task), plus a few cheap, low-maintenance additions the user did not ask for, each with why it matters to them.
  • Attacked: you critiqued your own plan from product, technical, money, risk, field-expert and skeptical-daily-user angles, and fixed the concrete flaws you found (expect at least 3 on anything with medium or high stakes; a low-stakes task needs a lighter pass). In DEEP mode an isolated critic subagent does this once. See 03.
  • Built to last 12 months:

    • volatile values (prices, rates, rules, prompts) live in one editable place;
    • history is protected from later edits;
    • there is a manual fallback for critical dependencies;
    • you state the scale ceiling.

    See 04.

  • Built and proven in slices, heaviest flow first. Each slice is verified (tests, a real run, worked examples) with no stubs and no fake data. See 06.
  • Tailored, then swept: first answer the context profile (07 Step 0), so only the rules the context activates apply; the rest are N/A with a one-line reason, never forced and never silently skipped, and depth follows stakes. Then a completeness sweep (07). It includes counterparty claims checked against contracts and caps, data minimised in everything you build, and the data sanity-checked against itself and field norms. Then a second-order sweep of every fix: past periods, cost and price, the people to tell, and the system's own failure modes. Every finding is tracked in TRACE.md and closed by a fix or a ready-to-use replacement, never by a warning alone.
  • Done by definition: the definition of done for each deliverable type is met (08). Every TRACE.md row is DONE, DEFERRED with a reason, DROPPED with the user's agreement, or N/A. The number of change scenarios the profile requires (08 §4; usually 3) have passed.
  • No silent deferral: a High row (law or compliance, safety or health, money at risk, irreversible data, the user's root vision, their #1 pain) ends DONE, with at least a minimal working version or exact actionable instructions, or ends DEFERRED only with agreed: "<user's words>". A line saying "check later" is not coverage.
  • Looking ahead: the domain brief's next-12–24-months section (scheduled mandates, price risk and hedging, financing, the growth path) has been checked against the plan.
  • Handed over:
    • setup a non-specialist can follow;
    • the daily routine;
    • how to change prices and rules safely;
    • backup;
    • what to check in 3, 6 and 12 months;
    • a re-review prompt.

If the project already has APEX files

  • docs/apex-blueprint/ from an older version: read it with care. Older templates were pre-filled, so trust only the parts clearly specific to this project, confirm those with the user in one line, and ignore its scores and "approved" lines. Leave it in place and start a fresh trace with --lean.
  • _apex/ from a current version: continue it. Keep the existing rows and update their statuses.
  • The user wants an existing system assessed and improved rather than something new built: that is the apex-review skill's job.

Rigor modes (choose silently)

ModeWhenShape
LEAN (default)One operator, single-session scope_apex/TRACE.md and _apex/STATE.json, plus a compact spec (_apex/SPEC.md, ≤ 120 lines) if useful. Most effort goes into building and proving.
STANDARDMulti-session product or business-critical processFull ledger in docs/apex-blueprint/; gates per slice.
DEEPThe user asks, or the stakes are high (money movement, health, legal, irreversible data, safety)STANDARD plus one synchronous critic-subagent pass, run once; never sleep-poll.

In chat-only hosts (no files or terminal), keep TRACE.md and the change scenarios as tables in the reply, and replace script gates with a short traced simulation: 1 normal case and 2 failure cases.

Which references to load

Load a reference only when its moment comes. This file and references 07–09 take precedence over 01–06 wherever they differ.

MomentLoad
Every request, before the first reply09 (domain brief)
Before presenting a plan and before handover07 (context profile, lenses), 08 (trace, definition of done, change scenarios)
A genuine fork to put to the user01
Self-critique at medium or high stakes03
Designing something long-lived04
Starting to build06
STANDARD / DEEP ledger work05; 02 for worked examples of field depth

A small LEAN request needs only 09, 07, 08, and 06 once you build.


Scripts (CLI/IDE hosts)

  • Scaffold:
    • LEAN: python "<skill_dir>/scripts/init_project.py" --dir "<target>" --name "<project>" --lean creates _apex/TRACE.md and _apex/STATE.json.
    • STANDARD / DEEP: the same command without --lean creates docs/apex-blueprint/.
    • --domain accepts DOM-01..07 or any free-text field, e.g. "dental clinic, Da Nang".
  • Gate audit: python "<skill_dir>/scripts/fitness_audit.py" --dir "<target>" --gate <1..6> --scan-code <files or folders you built>. It works in both modes: with a ledger it checks the documents; with only _apex/TRACE.md it runs the LEAN checks.
  • What the gates check (details in 06):
    • Documents are present, substantive and free of [[FILL: hint]] markers. The check is structural, so write them in the user's language.
    • TRACE.md rows are checked at every gate once the file exists: DONE has evidence, and a High row is DEFERRED only with agreed: "<user's words>". Keep the status and weight codes in English.
    • Gate 3 (ledger modes): a critique score ≥ 85, read from the scorecard line or from c_suite_audit_score in STATE.
    • Gate 5: --scan-code, and at least one proven slice (ledger modes).
    • Gate 6: no OPEN rows, all slices proven (ledger modes), and change_scenarios_required passed scenarios (default 3; fewer needs a change_scenarios_waiver reason).
  • The script checks structure and evidence, not quality. Never write a score, proof or status you did not actually obtain. Report results as plain outcomes, not gate numbers.

How the user invokes it

/apex <idea or goal> is the one command, and the skill runs only when the user types it or asks for APEX by name. It is a deliberate, heavier process; do not start it for a request that merely looks like a build task. Words in the request steer the mode:

  • "lean": LEAN mode; small scope, build as early as possible.
  • "red-team": only critique an idea, plan or codebase (03; critic subagent if available).
  • "blueprint": scaffold docs/apex-blueprint/ for STANDARD work.
  • "verify" or "audit": run fitness_audit.py with --scan-code and report what is still open.

If the user wants something that already exists assessed and improved, that is /apex-review, not this skill.

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.