メニュー
Use when someone asks what an X2 skill does, which skill to run next, how the gates work, or where an artifact lives. A quick in-flight reference; for the method's rationale, read X2.md at the method repo root.
Codex または Claude でインストール この Prompt をコピーして Codex、Claude、または他のアシスタントに貼り付けると、Skill ページを確認してインストールできます。
SOC 職業分類に基づく
| name | help |
| description | Use when someone asks what an X2 skill does, which skill to run next, how the gates work, or where an artifact lives. A quick in-flight reference; for the method's rationale, read X2.md at the method repo root. |
X2 in one paragraph: the kernel enforces the invariants in the build, so the skills only exist to capture human decisions at the right moments. The flow runs through three gates, each a prompted human turn that blocks the next phase: stories approved; the prototype approved at Claude Design and then approved again after copy-down (the lock); then ratified green, the audited exit report read and accepted. Green is declared by the slice-exit report and audited, never by prose. The sole metric is human turns per shipped slice, kept in a ledger file in the repo.
| Skill | One line | Gate it requests |
|---|---|---|
| stories | Problem discovery to epic-level stories, one line each | Gate 1: stories approved |
| seed | Settle which edition, create the repo from it, arm the CI gate, write CLAUDE.md constraints, start the ledger | none (setup) |
| decompose | D-000 five-lens decomposition, persistent schema, slice one, deltas file | none (stays-ahead turns) |
| design | Whole-product prototype and design system, in its own context | Gate 2a: prototype approved |
| lock | Copy the prototype and full design system into the repo, verify, stamp provenance | Gate 2b: the lock |
| derive-tests | Fidelity ledger and acceptance scenarios derived from the locked portion | none |
| implement | Build all of the locked portion to green autonomously | none (the exit report checks the contract) |
| slice-exit | The exit report: the definition of green, audited by intake | Gate 3: ratified green |
On demand: problem-solve (runtime defect, cause unknown, fewest observation turns), adopt (an existing project not built with X2 is migrated to run under it), extract (after a project completes: the quality feedback loop that folds the ledger's evidence back into the method), kernel (change the claims catalog, manage an edition, measure conformance, or check the whole kernel is still coherent), and this skill.
| Artifact | Home |
|---|---|
| Standing constraints, turn-report shape | the repo's CLAUDE.md (written by seed) |
| Turn ledger (append-only, one row per human turn) | docs/work/ |
| D-000 decomposition, story set, schema decisions | docs/decisions/D-000-decomposition.md |
| D-0xx decisions from record-gaps | docs/decisions/ |
| Deltas from the edition the project was seeded from | deltas.md |
| Behaviour spec (if the product has one) | design/ |
| Prototype, design system, provenance README | design/prototype/ |
| Per-slice fidelity ledgers | design/ledger/ |
| Slice-exit reports | docs/work/ |
Every human turn is one of: decision or green-but-wrong (sanctioned), edition-defect, invariant-gap, or methodology-defect (defects, each recorded as a candidate for the invariants pass), or admin (uncounted). Intake's classification governs.
X2.md at the method repo root: the methodology and its rationale.GLOSSARY.md at the method repo root: every record code the skills cite (MET-n, INV-n,
CONF-n, X-n, B-codes, and the record-layer codes NS-n, CF-n, PC-n), defined in one line each.kernel/claims/README.md: what the build enforces, which is what the skills no longer say.record/ at the method repo root: the sanitized cross-project record (negative-space
register and churn table, queued candidates, confirmations, metric history).Use when an existing project that was not built with X2 must be migrated to run under it, with documentation sprawl to dissolve, stories to harvest at epic altitude, an as-built design to baseline, and existing enforcement to map and arm. Runs the migration on a dedicated branch with three human gates. Do not use for a new project (that is seed), and do not use it to redesign the product or refactor code beyond what arming the enforcement requires.
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.
Use immediately after gate 1 (stories approved) to settle which kernel edition the project is built on, create the project repo from it, and make the method durable inside it. Also use when an existing X2 project is missing its CLAUDE.md standing constraints or its turn ledger. Do not use before stories are approved, and never use it to modify the kernel itself.
Use to change the claims catalog (add, amend, retire a claim, cut a catalog pass), to manage editions (add, retire, recompose), to measure an edition's conformance honestly (plant, prove, audit, lower a row), to clear accumulated decisions with the owner, or to check that the whole kernel is still coherent. Also use when a claim or edition change may have left counts, scope or statements in the documentation stale. Do not use it to seed a project from an edition; that is seed's job.
Use when an X2 project completes (or reaches a major milestone) to run the extraction loop, the quality feedback that folds the project's evidence back into the method. Reads the turn ledger and decision records, computes the metric and the per-decision churn table, distils sanitized candidates, confirmations, and negative-space nominations, and holds everything for a rulings pass. Do not use mid-slice (that is the normal intake flow), and never let raw project material cross into the public method.
Use when the prototype is approved at Claude Design (gate 2a) and a slice's portion must be copied down into the repo and locked. Also use when a spine revision touches an already-locked portion and it must be re-locked. Requests gate 2b (the lock). Do not use before gate 2a, and do not derive tests here; that is derive-tests, after the lock.