Pre-Execution Review Policy (internal)
One owner for the rules that both pre-execution reviewers apply, so a Product
review and a Plan review cannot drift into two different definitions of
independence, union, or convergence. It states policy; it runs nothing,
writes nothing, and emits no verdict of its own.
evidence-grounding = how an author prepares and self-checks an artifact.
pre-execution-review = how any pre-execution reviewer judges it, and what the
frozen ledgers look like.
lane (design/plan/review steps) = the surviving pre-execution path; feature 61
P8b retired the standalone review-spec / review-plan
verdict skills, so this pack now owns the ledger shapes
the lane and execute-phase consume, not a verdict gate.
When to use
- The lane's
plan/review steps — when writing the ledgers and when a review
comes back failed.
- Nothing else. Candidate-source review (
review-change), merge gating
(audit-pr) and execution (execute-phase) keep their own contracts; this
skill adds no authority over them.
Hard rule — no verdicts, no edits
This skill never prints SPEC-REVIEW-PASS, PLAN-REVIEW-PASS,
SPEC-REVIEW-FAIL, PLAN-REVIEW-FAIL or NEEDS-DESIGN as a result of its own
reading, and it never writes a unit artifact. Quoting a verdict shape here is a
definition, not an issuance. A policy summary that reads like an approval is a
contract violation — the verdict belongs to the reviewer turn that binds the
snapshot.
The three references
| Condition now |
LOAD |
| Running or repairing a pre-execution review |
references/POLICY.md — independence, union, dismissal, diversity labels, author exclusion, untrusted content, critique/synthesis/arbitration bounds, no quorum, no-progress, batch repair, CONVERGENCE-ANOMALY, write-then-report |
| Building or re-checking a snapshot digest |
references/SNAPSHOT.md — the one executable recipe (scripts/pre-execution-snapshot.mjs), what each stage binds, how a consumer re-verifies a receipt, and why a snapshot digest is not a git blob id |
| Writing or validating a Plan-stage artifact |
references/LEDGERS.md — the planning-evidence table, the obligation ledger, the stage-aware planning-findings.md, the durable review mark, and who may write each |
Guardrails
- Single owner. A caller may restate a rule only as a one-line pointer plus
the stage-specific detail it adds. Two copies of the union rule, the dismissal
rule, or the convergence fields is the drift this skill exists to prevent.
- Policy is not a receipt. Reading this file proves nothing about any
artifact; only a reviewer turn that binds a snapshot does.
- No weakening. The bounds here are ceilings and floors: a cycle may be
shorter than the budget allows, never looser than the policy allows.
- Vocabulary is closed. Severity, class, role, verdict and status words are
exactly those named in the references. Inventing
accepted, waived,
mitigated, or a fourth verdict is a contract break.
- Docs-language and commit conventions per the project's Workflow conventions.
Done when
- Every pre-execution reviewer and authoring caller points here for the shared
cycle and ledger rules instead of restating them, and each stage file keeps
only the detail that is genuinely stage-specific.
1---2name: pre-execution-review3user-invocable: false4version: 4.10.05author: "Gabriel Trabanco <1969593+gtrabanco@users.noreply.github.com>"6license: MIT7description: >8 Internal owner of the pre-execution review cycle and the planning ledgers:9 independence, unioned findings, counter-evidence dismissal, no-progress,10 `CONVERGENCE-ANOMALY`, and the evidence/obligation/findings tables. Consumed by11 `execute-phase`, `review-change`, and the lane's authoring steps for the12 **planning-ledger shapes and ownership**; the pre-execution verdict owners13 (`review-spec`/`review-plan`) were retired by feature 61. Not a menu entry.14metadata:15 source: https://skillmd.com/skills/gtrabanco/pre-execution-review16---1718# Pre-Execution Review Policy (internal)1920One owner for the rules that both pre-execution reviewers apply, so a Product21review and a Plan review cannot drift into two different definitions of22independence, union, or convergence. It states **policy**; it runs nothing,23writes nothing, and emits no verdict of its own.2425```text26evidence-grounding = how an author prepares and self-checks an artifact.27pre-execution-review = how any pre-execution reviewer judges it, and what the28 frozen ledgers look like.29lane (design/plan/review steps) = the surviving pre-execution path; feature 6130 P8b retired the standalone review-spec / review-plan31 verdict skills, so this pack now owns the ledger shapes32 the lane and execute-phase consume, not a verdict gate.33```3435## When to use3637- The lane's `plan`/`review` steps — when writing the ledgers and when a review38 comes back failed.39- Nothing else. Candidate-source review (`review-change`), merge gating40 (`audit-pr`) and execution (`execute-phase`) keep their own contracts; this41 skill adds no authority over them.4243## Hard rule — no verdicts, no edits4445This skill never prints `SPEC-REVIEW-PASS`, `PLAN-REVIEW-PASS`,46`SPEC-REVIEW-FAIL`, `PLAN-REVIEW-FAIL` or `NEEDS-DESIGN` as a result of its own47reading, and it never writes a unit artifact. Quoting a verdict shape here is a48definition, not an issuance. A policy summary that reads like an approval is a49contract violation — the verdict belongs to the reviewer turn that binds the50snapshot.5152## The three references5354| Condition now | LOAD |55|---|---|56| Running or repairing a pre-execution review | [references/POLICY.md](references/POLICY.md) — independence, union, dismissal, diversity labels, author exclusion, untrusted content, critique/synthesis/arbitration bounds, no quorum, no-progress, batch repair, `CONVERGENCE-ANOMALY`, write-then-report |57| Building or re-checking a snapshot digest | [references/SNAPSHOT.md](references/SNAPSHOT.md) — the one executable recipe (`scripts/pre-execution-snapshot.mjs`), what each stage binds, how a consumer re-verifies a receipt, and why a snapshot digest is not a git blob id |58| Writing or validating a Plan-stage artifact | [references/LEDGERS.md](references/LEDGERS.md) — the planning-evidence table, the obligation ledger, the stage-aware `planning-findings.md`, the durable review mark, and who may write each |5960## Guardrails6162- **Single owner.** A caller may restate a rule only as a one-line pointer plus63 the stage-specific detail it adds. Two copies of the union rule, the dismissal64 rule, or the convergence fields is the drift this skill exists to prevent.65- **Policy is not a receipt.** Reading this file proves nothing about any66 artifact; only a reviewer turn that binds a snapshot does.67- **No weakening.** The bounds here are ceilings *and* floors: a cycle may be68 shorter than the budget allows, never looser than the policy allows.69- **Vocabulary is closed.** Severity, class, role, verdict and status words are70 exactly those named in the references. Inventing `accepted`, `waived`,71 `mitigated`, or a fourth verdict is a contract break.72- Docs-language and commit conventions per the project's Workflow conventions.7374## Done when7576- Every pre-execution reviewer and authoring caller points here for the shared77 cycle and ledger rules instead of restating them, and each stage file keeps78 only the detail that is genuinely stage-specific.
