Index of every page on this site: llms.txt
All skills
wix avatar

/rp-execution-policy

@e2bc48e official
by Wix.comwix/skills40 stars
42
Run it yourself

Applies cross-stage execution safety policy and writes the aggregate migration completion artifact after all delivery-specific terminal evidence validates.

  • 3 files
  • 33 KB
  • Updated last week
  • GitHub
Use this Skill: https://skilld.dev/gh/wix/skills/rp-execution-policy

Nothing lands on disk. Nothing to clean up.

Fork this Skill

Edit a local copy. It keeps the author and licence.

SKILL.md

≈44 tokens for metadata: the name and description. ≈2.2k when used: this file.

Description uses 2.2% of example budget

Before choosing a Skill, your Agent reads its name and description. All available Skills share that space.

  • A shorter description leaves more room for other Skills. This entry exceeds our 1% size suggestion.

In our Claude Code example, all Skill names and descriptions share 8,000 characters. This Skill uses ≈177 characters, or 2.2%.

The 1% threshold is a size suggestion. Longer descriptions can still fit.

Your model, settings, and other Skills decide how much text your Agent can read.

Example settings and source

The example uses a 200k-token context and default Claude Code settings. The count includes the name, description, separators, and when_to_use when present. Codex also counts local file paths.

Skit's source and limits: Codex 0.160.1, Claude Code 2.1.292.

rp-execution-policy

Use this resource for the execution approval gate and aggregate finalization.

For awaiting_execution_approval, present the generated execution plan, wait for the required user acceptance in normal mode, and persist it in orchestration/approvals.json. In explicit user-requested 1-click mode (automationMode=one_click, source=user), use the deterministic agent approval helper only after the plan and any required code-safety review validate. Then route again. A non-user-authored one_click value is normal interactive mode and must wait for user acceptance.

Disclosing what was read from a credential-bearing source surface

If the run read payment-gateway configuration from the source, the plan presented at the approval gate must say so: that gateway configuration was read, which gateways were found, how many are enabled, how many credential fields are configured, and that no credential value was retained. The source adapter produces that line — for WordPress, describeGatewaySummary() in rp-source-wordpress/lib/gateway-summary.mjs.

A run that reports "N gateways, M enabled, K credential fields present, 0 values retained" is auditable. One that read the same thing silently is not, and the user approving the plan has no way to tell the two apart. Never put a credential value, prefix, length or hash into the plan report itself.

Reporting the payments section of the plan

If the run imports payment records, refunds or invoice documents, build that section of the plan with lib/payments-report.js rather than by hand. It enforces three things a hand-written summary has got wrong before, and it BLOCKS the approval gate rather than printing a caveat:

  • The payment outcomes must sum to the discovered source-order population. A report whose numbers do not add up is hiding orders. The builder returns approvable: false with the shortfall named.
  • Zero is a result. Every outcome and every count is emitted even at zero, so "0 refunds found" is reviewable at a glance. An implausible number is cheapest to catch here.
  • Candidate coverage is not writable coverage. Orders carrying a reference-shaped value and orders whose reference can be defensibly attributed are separate lines. Collapsing them overstates fidelity.

The denominator must come from the run's own reconciled reads (from: 'run-reconciled-read'), not from a prior survey. Where a survey counted differently, both numbers are shown and the discrepancy must be reconciled before the pilot.

The provider inventory carries identifier, display name and enabled state only — a credential-shaped field on a provider blocks the gate. The phrase "you will lose your history" is prohibited: it is false for payment records, which are provider-independent.

Run:

node skills/wix-replatform/scripts/migration-completion.js <projectDir>

This command validates backend completion accounting and, for website delivery, the frontend completion receipt. It writes completion/migration-completion.json atomically.

It also refuses, on both delivery modes, unless the run produced its post-execution handoff and its execution log. A management run produces no website and used to require no handoff at all, which is how a backend-only migration could finalize with no account of itself anywhere in the project. The typed refusal reasons are:

Reason Meaning
missing_execution_log no execution-log.md; route back to the import step that owes the chronology
missing_post_execution_handoff no execution/review/migration-handoff.json; generate it from the final outcome record
completion_report_missing_last_write the completion report records no lastWriteAt, so nothing can prove the handoff postdates the import
post_execution_handoff_predates_last_write the handoff was generated before the run's last write — it describes the plan, not the run

The last two compare against the completion report's own recorded last-write timestamp, never a file mtime: an mtime is refreshed by a copy and proves nothing. If validation fails, do not manufacture completion; route again to the module that owns the missing or stale artifact.

Entity-QA attestation gate

The same finalizer also refuses without a fresh, complete execution/qa/entity-qa-attestation.json that matches the current execution/completion-report.json digest and covers every mapped entity/subtype with matching counts (rp-execute-import/SKILL.md's "Entity-by-entity QA" section), including each mapped entity's archive.<entity> source-archive row when the report carries one. The refusal carries entityQa: { qaStatus, entities, recheck }:

qaStatus Meaning Action
missing no attestation file, it doesn't parse as JSON, its schema is invalid, there is no usable completion report to compare against, or the independent mapped-entity coverage source (mapping/mapping-plan.json or, for quick mode, quick-mode/plan.json) is unavailable run the full entity-by-entity QA pass from rp-execute-import, then scripts/entity-qa-attestation.js
stale the attestation's completionReportDigest does not match the report that is shipping — a later pass rewrote it run the full entity-by-entity QA pass again, then the writer; the shipped prose treats any report change as reopening the whole review
failed the attestation disagrees with the independent mapped-entity list or the report's own counts for one or more entities, named in entityQa.entities dispatch a fresh Developer for the listed entities, then run the full QA pass and the writer

A valid attestation proves import-row freshness and coverage only (scope: 'import-entity-coverage' in migration-completion.json) — never business setup or launch readiness, and never proof the QA pass itself happened rather than a copy of the report's own counts.

Site settings in the execution plan

When the setup plan has import-site-locale, show the owner what the first imported entity will do:

node skills/wix-replatform/scripts/site-regional.js preview <project>

It prints, from the source file alone (the site is read when setup runs), "Currency: set to … (from the store)", the same for time zone, language and locale, and the rule: a site created by this migration, or an existing site with no orders and no products, is set directly; one that holds an order or any product asks the owner first (one question, change or keep, raised by setup — never answered for the owner, one-click included). No sample-product classifier ships today. Nothing regional needs separate approval or a file bound into the approval.

Reporting coupons, currency and plan definitions

lib/coupon-currency-plan-report.js buildCouponCurrencyPlanReport renders the section and decides approvability the same way payments-report.js does: every line with its count including zero; per area written + excluded + skipped = discovered; exactly one currency outcome, and a mismatch or multi-currency history BLOCKS as a halt; the coupon-usage surface must be named. Two phrases are refused wherever they appear — "imported N coupons" (the honest headline is the redeemability breakdown) and "not supported" (zero plan rows is "none found", a measured fact).

Discovered-set status and safe refinalization

For every standard-mode finalization, build execution/discovered-set-status.json with node scripts/discovered-set-status.js build <projectDir> after the completion report, then regenerate the post-execution handoff, retain the current entity-QA attestation, and run node scripts/migration-completion.js <projectDir>. Missing or stale census approval, reconciliation or status is a hold; rebuild the owning artifact and obtain a fresh mapping review where required. The final receipt is version 2. Open discovered-set rows keep a complete backend at complete_with_warnings, and the handoff and receipt must agree.

On reviewOnly: true, refinalization uses the existing import outcomes. Make no Wix client, setup, codegen, import or website dispatch. Rebuild only the discovered-set status and post-execution handoff, then the aggregate completion; the writer archives the original version-1 receipt once. Any other route is held for the Import Operator. Manual work can be recorded only for a manual-open row, by the merchant or Cutover Operator through node scripts/discovered-set-status.js resolve <projectDir> --census-id <id> --by merchant|cutover-operator --evidence <short non-record evidence>. All other open outcomes require their existing mapping/import recovery path. Quick mode requires no discovered-set status.

Source: SKILL.md on GitHub

skilld matched fixed text patterns in SKILL.md and file names. Patterns miss obfuscated code.

skilld run checks every file with the same patterns. It asks for approval before it loads a Skill with a behavior marked Needs approval.

No third-party reports yet.

Signed by skilld at e2bc48e. This ties the file your Agent reads to that commit on GitHub. It does not review the instructions.

Last checked against GitHub yesterday.

Activeupdated last week

README badge

README badge for wix/skills/rp-execution-policy