Decompose
AI builds faster than anyone, with no skin in the game and no memory of yesterday. How do you govern that? With X2, decisions stay in files people own, unbreakable rules fail the build, code is disposable, and every human turn is counted. Done is an audited report, never the agent's word.
npx -y skills add teklabsdigital/x2-method --skill decomposeAssembled from the repository path, not quoted from the project. Check it against their README if it does not work.
3 things to look at
- 26 days oldThe repository was created 26 days ago. New is not bad, but a brand new repository carrying a familiar-sounding name is the shape a typosquat arrives in, and there has been no time for anyone else to find a problem with it.
- no licenseNo license file was found in the repository. Code published without one is not open source by default, so using it at work is a question for whoever answers licensing questions where you are.
- 13 stars13 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.
What its author says it does
Copied from the file, not written here
Use after the repo is seeded to run the D-000 five-lens decomposition, record persistent schema decisions, define slice one, and start the deltas file. Also use when a stays-ahead decision (decomposition, schema) must be revisited. Do not use for per-slice implementation decisions; those are D-0xx record-gaps raised during implement.
SKILL.md
3.4 KB, as published. Nobody here has run it
X2 Decompose (D-000)
The one place where the model's whole-solution-space view beats the human's slice-by-slice view, and the decisions a slice cannot cheaply reverse. Everything else stays reversible and waits.
The five-lens session
Decompose the story set under five imposed lenses, one pass each, no code: data flow, failure
domain, change frequency, team boundary, trust boundary. Look at what disagrees between the five
decompositions; the disagreements are where the real shape decision lives. Choose a shape. Record
the choice and the four rejected shapes with reasons as D-000-decomposition.md in
docs/decisions/.
D-000 also records:
- The story set from gate 1. D-000 is the source of truth for it; there is no separate requirements document (the pilot's turn 8b defect was creating one).
- The recorded product decisions from ideation.
- The persistent schema decisions. Schema is the second stays-ahead asset; approving it is a human turn, and every later migration is an unconditional human turn, every time.
- The slice-one definition: the smallest slice that ships real user-visible value end to end. Build scope and design scope are separate rulings: the build is slice-first, the design is whole-product (MET-03); record both.
The launch and exposure pre-flight
Applies when the product will ship to a public audience; rule each concern IN or OUT now, so none of it arrives as ad-hoc directives at the end (record: NS-2, NS-3):
- The estate question: does this greenfield app land inside an existing brand estate (styles, domains, policies, analytics) that is not greenfield?
- Exposure values: serving domain, sender addresses, public identifiers, entry points. These are ruled before first use, never assumed; an assumed value propagates into green artifacts and surfaces late.
- Launch concerns: brand chrome and legal pages, discoverability (robots, sitemaps, social and AI-readability), analytics, abuse pricing, and edge hardening.
Also read the cross-project churn table in the method's record/negative-space.md: it names the
decision types history says will churn. Pre-decide the pre-decidable ones here; budget the
iteration-inherent ones (story refinement, model selection, content voice) instead of fighting
them. For the marked chronic churn types, record a confidence and a "wrong-if" line on the
decision so reality can trip it later.
The deltas file
Start deltas.md: how this project differs from the kernel edition. It grows as the build
surfaces differences; it never restates what the edition already says.
Human-turn contract
- Human turns here: the shape choice, schema approval, slice-scope ruling. These are stays-ahead decisions and are not counted in the per-slice metric.
- No MET-07 gate of its own; gate 2 belongs to the design step that follows.
What this skill must NOT produce
- No screens, no prototype, no acceptance criteria, no tests, no code.
- No detailed per-module design; a single-service module is legal and the shape decision is the module boundary, not the internals.
Next
design, carrying D-000 and the behaviour spec.