Turn accepted intent and review evidence into correct-by-construction software through Goal Contracts, Counterexample Sets, Construction Contracts, and an Evidence Ledger. Use bare $actuating for implementation, Ship publication, and review convergence; use explicit implement, triage, remediation-plan, or review-closeout for their bounded routes. Begin every architecture or abstraction decision with OPERATE ARCHITECTONICALLY, use $first-principles to establish the incumbent-independent Construction basis before $universalist nomination, and integrate one bounded $metanoetic pass when an initial high-regret nomination, accepted findings, causal recurrence, or cumulative review-path accretion makes abstraction change live. Before fresh review after repair, fold the cumulative changeset and review-induced delta so pointwise minimal repairs cannot compose into a dominated final Construction. Actuating alone selects the Construction and next action; Ledger is non-executing and Ship alone owns public effects.
Install with Codex or Claude Copy this prompt, paste it into Codex, Claude, or another assistant, and let it review the skill page and install it for you.
A direct command skips the review prompt. Inspect the source before running it.
The command stays on one line. Scroll horizontally to inspect it before copying.
Prefer a local copy? Download the files currently available to SkillsMP.
File Explorer
21 files
Showing SKILL.md
SKILL.md
Source instructions · Read-only preview
name
actuating
description
Turn accepted intent and review evidence into correct-by-construction software through Goal Contracts, Counterexample Sets, Construction Contracts, and an Evidence Ledger. Use bare $actuating for implementation, Ship publication, and review convergence; use explicit implement, triage, remediation-plan, or review-closeout for their bounded routes. Begin every architecture or abstraction decision with OPERATE ARCHITECTONICALLY, use $first-principles to establish the incumbent-independent Construction basis before $universalist nomination, and integrate one bounded $metanoetic pass when an initial high-regret nomination, accepted findings, causal recurrence, or cumulative review-path accretion makes abstraction change live. Before fresh review after repair, fold the cumulative changeset and review-induced delta so pointwise minimal repairs cannot compose into a dominated final Construction. Actuating alone selects the Construction and next action; Ledger is non-executing and Ship alone owns public effects.
Actuating
Turn accepted intent into a lawful construction, directly orchestrated effects,
independent falsification, and an evidence-backed closure judgment.
Authority kernel
Use exactly four authoritative per-goal artifact families:
goal-contract/v3 — accepted semantics, authority, scope, compatibility,
laws, and acceptance, compiled by $goal-contract.
counterexample-set/v1 — classified witnessed falsifications, authored by
$review-fold.
construction-contract/v3 — the selected architecture, compared candidate
families, factor surfaces, supersession, proof obligations, preserved
observations, and retirements, authored by $actuating after an
Actuating-bound $first-principles basis, a $universalist nomination, and,
when required, one $reduce challenge.
The Goal Contract is the sole semantic-authority artifact. The Counterexample
Set is the sole classified-bug artifact. The Construction Contract is the sole
architecture-selection artifact. The Evidence Ledger is the sole mutable
per-goal truth. Read artifact-kernel.md.
Plans, CAS receipts, Ship receipts, verifier output, work graphs, and Ledger
projections are supporting evidence or discardable structural aids. An
Actuating-authored closure receipt is a semantic report, not another authority
family.
Owner boundary
Actuating owns:
correct-by-construction implementation;
evaluation of current Counterexample classes against the current Construction;
initial and successor Construction selection;
repository-effect and review orchestration;
the clean committed Git subject used by local proof;
review-path accretion evaluation and pathwise-minimality disposition;
construction and ownership of the static Review Contract;
semantic evaluation of CAS owner facts and review credit;
the next legal action;
application of the closure theorem and authorship of its semantic receipt.
$review-fold classifies witnessed facts before Actuating selects a repair.
$first-principles establishes the admissible premise basis without revising
the Goal. $universalist lowers the admissible derivation and nominates the
essential boundary shape. When abstraction change is live, $metanoetic
performs one bounded generative escalation and Universalist lowers every
material result. $reduce may challenge nominated factors through congruent
quotients, ablations, and recomposition. $complexity-mitigator may supply one
read-only local normal-form preflight when review accretion is realization-only.
The composition order is:
activate
-> axiomatize once or retain the current basis
-> nominate
-> metanoetic once when triggered
-> lower
-> challenge once
-> adjudicate
Actuating alone performs adjudication. No supporting skill, review prose, plan,
or Reduction Certificate selects a Construction, Repair Disposition, Review
Accretion Disposition, operation, next action, or closure.
Ledger may materialize, canonicalize, validate, append, replay, and emit
requested disposable structural projections. Ledger never executes repository
changes, invokes Git, evaluates CAS facts or review credit, interprets Ship,
selects a repair or Construction, grants mutation, chooses the next action, or
emits or authors semantic closure.
Before the first Ledger command, load $ledger and complete $ledger ensure
once. Require Ledger 1.x with ledger-artifact-abi/v1, Seq 1.x with
seq-observation-abi/v1, and successful checks for every selected passive
definition. Apply the same gate when entering from a standalone Goal Contract
or Review Fold handoff. Construction v1 and v2 are unsupported; start a fresh
v3 goal-local Evidence store rather than treating legacy data as current
authority.
$ship is the sole owner of public PR or tracker effects and Ship-owned
publication evidence. Actuating supplies a current ready-to-ship proof and
records Ship's returned receipt; it never performs the public effect or authors
a substitute publication receipt itself.
Public modes
Intent
Route
Mutation
Terminal result
Bare $actuating or /goal $actuating
implement -> Ship -> review-closeout
Authority-bound
complete
$actuating implement
implementation only
Authority-bound
Local complete
$actuating triage
acquire and classify review
Forbidden
Counterexample Set and report
$actuating remediation-plan
propose a successor Construction
Forbidden
Non-executable Construction Contract
$actuating review-closeout
repair, ablate, Ship when required, and re-review
Authority-bound
complete
An unqualified request to review, inspect, audit, or classify selects triage.
Require explicit implement, fix, resolve, address, or closeout intent before
mutation.
Architectonic decision gate
At the beginning of every process that selects, preserves, changes, or ablates
architecture or abstraction, and before Universalist nomination, state exactly:
OPERATE ARCHITECTONICALLY
This is an activation instruction, not evidence, authority, or a receipt.
Axiomatic Construction gate
Before the first Universalist nomination for each materially new candidate
universe, invoke $first-principles with the current Goal fixed as the
irreducible outcome and sole semantic authority. Bracket inherited repository
abstractions, conventions, analogies, owner boundaries, rationale, and alleged
constraints. Do not reopen Goal outcomes, laws, scope, compatibility,
authority, acceptance, or proof posture.
Inspect the incumbent only to establish observed facts, external obligations,
and host enforcement capabilities. Freeze the incumbent-independent derivation
before incumbent comparison, then record:
The basis is an ephemeral proof lease over exact Goal, fact, constraint, and
host-capability inputs. It is not a fifth authority artifact and adds no
Construction field. It expires at session end, compaction, or execution-context
handoff and is never reconstructed from a materialized Construction. Within one
uninterrupted run, a premise-neutral clean commit change may retain the basis
while all inputs and invalidators remain current. Any later run or premise
change re-axiomatizes before nomination or affected mutation.
A sufficient basis admits nomination. underdetermined preserves every
materially incomparable derivation and blocks if Goal law, observation, or
dominance cannot distinguish them. inconsistent or blocked stops.
$first-principles may expose a Goal conflict or missing authority, but it
cannot rewrite the Goal, classify Counterexamples, select a Construction,
grant mutation, or author a durable decision.
Before authoring a successor Goal, project goal-carry-forward-context from the
canonical Evidence store. Treat its retained Goal artifact as current and
derive supporting references only from retained classes whose status is
accepted, blocked, or follow-up. Never guess references, reconstruct the
Goal from an older artifact, or read the event log directly. An unavailable or
incomplete projection is an owner-side obstruction.
When accepted or blocked Counterexamples remain unresolved, or a revision
brings a follow-up class within scope, the successor Goal cites every Set
carrying those classes. $review-fold then authors a successor Set that cites
the successor Goal, preserves lineage, evaluates the predecessor Construction,
and dispositions every carried class. No affected mutation or successor
Construction selection occurs before that carry-forward is complete.
Metanoetic escalation
For an initial Construction with no classified findings, invoke $metanoetic
exactly once after the first Universalist nomination and before candidate
adjudication when the nomination is high-regret or difficult to reverse,
remains a coherent but merely adequate local optimum, and a materially
different Construction is plausible. Initial implementation, architecture, or
consequence alone is not a trigger.
After $review-fold classifies findings, invoke $metanoetic exactly once when
an accepted class:
makes architecture-repair or ablation-repair live;
challenges representation, owner, admitted domain, equivalence,
normalization, or information retention;
triggers causal recurrence; or
would add a validator, correlation, cache, bypass, compatibility branch, or
path-dependent recovery to reconstruct forgotten information;
or when Review Accretion establishes material path dependence, accumulated
semantic machinery, obsolete residue, proof accretion, inadequate ordinary
normalization, or pathwise domination pressure.
Compile material results into existing candidate comparison, factor surfaces,
supersession, proof obligations, and falsifiers. Metanoetic neither classifies,
nominates, selects, authorizes, nor closes. Do not repeat it for an unchanged
decision surface.
Committed Git subject
The durable Actuating subject is one clean Git commit target. Dirty index,
worktree, untracked, and ignored state is provisional implementation state and
never an Evidence subject.
git status --porcelain=v2 --untracked-files=all --ignore-submodules=none
is empty;
branch attachment is excluded from semantic identity;
ignored files are excluded because they are not candidates for the committed
artifact.
Ledger stores and compares this opaque digest. It never invokes Git or derives
the tuple. Ship separately binds repository, base and head refs, exact OIDs,
and live publication readback.
Construction procedure
Compile accepted source with $goal-contract. Require the canonical artifact,
non-null artifact_id, and applicable Goal registration event.
Enter the Architectonic decision gate. Establish or retain the Axiomatic
Construction Basis. Apply Universalist at every changed or preserved
boundary. Complete one Metanoetic pass and re-lowering when a live trigger
exists.
Compile exactly four candidate families in canonical order:
realization-preserve, admitted-domain-restriction,
representation-or-owner-strengthening, and ablation-normalization.
Factor every independent mandatory repair into one common obligation core;
vary only the disputed family delta. Give each candidate a factor inventory
and falsifier, mark at least one genuinely incumbent-independent, and select
exactly one.
Challenge the nominated candidate once with $reduce when it adds or
preserves an independent semantic owner, parallel representation, bypass,
compatibility branch, semantic mechanism, or apparently dominated residue.
Record:
Choose the least additive lawful route. add explains why delete,
consolidate, and edit are insufficient.
Adjudicate basis, nomination, escalation, and challenge. Select the smallest
non-dominated Construction satisfying every Goal law, make invalid states
unrepresentable where feasible, and name exact proof and retirement
obligations. Partition every predecessor and successor factor as preserved,
retired, introduced, or replaced.
Materialize and register the selected Construction through Ledger. Only the
returned canonical artifact and appended registration event make it current.
Require the repository to be at the exact clean committed subject before
preparing an operation.
For each edit:
clean parent subject
-> prepare exact operation
-> one-seam-operator creates one provisional diff
-> Actuating inspects the complete diff and changed paths
-> Actuating commits exactly that operation
-> require one-parent clean successor
-> derive successor subject
-> record-effect parent -> successor
-> verify and falsify on the successor commit
one-seam-operator never stages, commits, amends, pushes, or publishes.
Actuating rejects unrelated dirty state before preparation. The successor
commit has exactly one parent equal to the prepared parent, a nonempty commit
path set exactly equal to the selected operation, only allowed paths, and a
clean checkout. effect_recorded advances directly from the parent subject
in pre_effect_subject_digest to the successor subject in the event envelope.
Run inspect, verifier, falsifier, and retirement operations only on an exact
clean commit. Require the checkout and subject to remain unchanged afterward.
Record immutable outputs through evidence-ledger.md.
Re-evaluate all current artifacts and observations. Select the next
operation, Ship handoff, review action, closure judgment, or blocker.
The one-operation law is:
select -> prepare -> effect -> commit -> record -> observe -> evaluate -> select or close
No stage may smuggle a second repository effect. A document, operation
envelope, validator pass, Ledger append, review result, or Construction never
grants mutation by itself. Mutation requires accepted authority, a current
Construction, an exact clean committed subject, and an in-scope operation.
Goal-causal lineage
Subject freshness and causal decision lineage are independent. A clean commit,
publication, or other material subject change makes subject-bound proof,
operations, review bindings, and review credit stale. It does not erase prior
Construction decisions or Counterexample history for the same goal_id.
Only the first Construction in the authoritative v3 lineage may use
mode: initial with no predecessor. Every later Construction names the exact
current Construction as its sole predecessor, including after a successor Goal
or clean subject rebind.
Before accepting a new Counterexample Set, resolve prior Sets for the Goal.
When a stable class recurs, require the new Set to name the most recent Set
carrying it. Missing lineage blocks and does not make the class novel.
Counterexample and causal recurrence procedure
Every witnessed bug, failing test, incident, compatibility failure, or review
finding passes through $review-fold. Actuating then determines whether each
accepted class is:
a realization defect;
an architecture defect requiring a successor Construction;
an ablation defect requiring dominated-residue removal; or
blocked by missing authority or evidence.
Re-establish the current-run basis before choosing an affected repair. If a
finding falsifies a premise or makes architecture/ablation live, re-axiomatize,
run the bounded Metanoetic pass, lower through Universalist, and compare the
successor candidates before selecting an operation.
The causal recurrence gate triggers when one accepted class recurs after repair
or two accepted classes across subject revisions share an evidenced missing
observation, authority, correlation, or Construction factor. Similar files,
prose, diff size, or timing do not establish shared cause.
Record:
Causal Recurrence Disposition
Evidence and class refs:
Shared cause:
Current Construction factor:
Candidate comparison: realization preserve / admitted-domain restriction / representation strengthening / ablation normalization
Disposition: instance-specific | architecture-repair | ablation-repair | blocked
Why another local repair is sufficient or forbidden:
Proof:
Falsifier:
A recurring cluster makes abstraction change live and invalidates the prior
candidate universe. instance-specific requires non-example separation proof.
Otherwise select architecture/ablation repair or block. Do not add another
validator that recreates information the representation repeatedly forgets.
Review accretion
A sequence of pointwise least-additive repairs may compose into a globally
dominated realization. After every review-driven mutation epoch and before the
next Ship handoff or fresh review, apply the
Review Accretion Gate. Apply it again before
final closeout when mutation occurred after the last disposition.
Bind the exact delivery baseline, review-entry committed subject, and current
committed subject. Keep the review-entry anchor across campaigns, publication
epochs, review-credit resets, subject rebinding, and successor Constructions for
the same closeout objective. Fold the complete delivery and review-induced
commit deltas, Construction and Counterexample lineage, and cumulative
production, semantic, proof, and comprehension surfaces. Line count, file
count, elapsed time, and repair count force inspection only.
First derive the ordinary cumulative normal form. Use one read-only
$complexity-mitigator Micro Preflight when the architecture and factor
inventory remain sufficient. When representation, owner, admitted domain,
equivalence, information retention, or boundary may be insufficient, retain or
re-establish the basis and require Universalist to nominate the ordinary
repository-native normal form.
When the fold establishes material path dependence, accumulated semantic
machinery, obsolete residue, proof accretion, inadequate ordinary
normalization, or pathwise domination, run Metanoetic with
Trigger: review-path-accretion, lower the result, challenge it once with
Reduce, and adjudicate.
Select exactly one disposition: preserve, realization-normalization,
ablation-repair, architecture-repair, or blocked. Any material repair
creates a new committed review subject and resets all review credit. Compile
material results into existing Construction factors, supersession, proof
obligations, and retirements. Add no artifact family, mutable control state,
review lens, or Ledger/Seq/CAS accretion command.
Review convergence
$first-principles, Review Accretion, and Metanoetic are Construction-selection
passes, not review lenses. Preserve the static topology of standard plus the
existing four auxiliaries.
Before binding or dispatching closure-grade review, require:
The first receipt must be compatible for the exact resolved Codex runtime,
contract, and managed-ws transport with every required probe passed. The two
features come from cas capabilities --json. Any missing runtime or compiled
capability proof blocks before request binding or review/start.
Bind standard plus four auxiliary requests before dispatch. Launch the initial
1+4 wave as five concurrent owner-lived:
cas review start --wait --timeout-ms 2700000
processes, one per request, and retain each process through a structured
terminal receipt or explicit terminal owner failure. Do not split a
workflow-bound attempt into detached start and unrelated wait. Never cancel
a sibling because another request finds a defect or loses transport.
A verdictless terminal failure earns zero credit and reruns only that request
once on the same committed subject with a fresh attempt. Preserve completed
siblings and standard clean credit on the unchanged subject. Count the initial
wave's standard clean as attempt one, run four later standard attempts
serially, and require five consecutive distinct standard cleans. Any committed
subject change resets all review credit and requires a fresh 1+4 wave.
Publication-bearing review uses Ship-confirmed remote identity. For a current
PR, use SHIP-v1; for a new adoption route, record a read-only
SHIP-OBSERVATION-v1 pre-review publication observation before binding or
dispatching the campaign. Map the exact published target to
cas review --base <immutable-base-sha> and require every receipt's
baseSha, headSha, and target fingerprint to equal Ship readback. Never use
--uncommitted for a published PR. --commit HEAD is sufficient only when one
commit is the complete bound PR delta.
Actuating evaluates CAS owner receipts and review credit. Ledger may record
receipt references and project structural history, but never dispatches CAS or
translates CAS fields into semantic verdicts.
Publication and closure
Bare mode and publication-bearing review closeout hand a current
ready-to-ship proof to $ship. Ship either pushes the exact clean verified
commit and creates or updates its PR, returning SHIP-v1, or read-only adopts
an exact already-public subject that has no truthful current PR tuple, returning
SHIP-ADOPTION-v1. Both routes exact-match the repository, base/head refs and
OIDs, complete actuation binding, and route-specific live readback. Actuating
must not substitute its own live-readback record.
Ship evidence precedes review. For a new campaign, a later
SHIP-ADOPTION-v1 does not reset review credit merely because the adoption
receipt was recorded later when it ratifies the exact SHIP-OBSERVATION-v1
digest that Actuating recorded before the credited campaign. The observation, adoption, and every credited CAS
receipt must exact-match the repository, canonical head ref, base/head tuple,
subject, current default-branch state, Goal, Construction, and review contract
through the observation's review_binding. That binding is the exact named
Goal/Construction/subject/review-contract projection of Ship's validated
actuation binding; it is not a second authority.
The final adoption carries its own current actuation binding. The
publication-review-events projection exposes exact ordered publication and
campaign event rows without interpreting them. Evidence Ledger
order from the recorded Ship observation to campaign binding proves causality;
never compare provider and Ledger wall clocks. Otherwise the review credit is
stale. The reducer admits exactly one campaign-start occurrence for the current
Goal, Construction, subject, and Review Contract and binds every request to
that occurrence. The pre-review observation proves the exact reviewed bytes
were public before that occurrence; ratification exact-matches its stable tuple
while final adoption freshly re-reads the same live default branch. Neither
endpoint comparison nor ratification claims intermediate branch continuity.
For a historical campaign that predates SHIP-OBSERVATION-v1, adoption may
instead carry provider-backed publication evidence for the exact target.
Actuating preserves review credit only when a content-addressed causal-order
observation binds that exact provider event to the exact campaign start and
shows the publication operation completed first. The supporting attachment is
actuating-publication-campaign-causality/v1, owned and issued by Actuating
after it validates one exact Seq observation, successful call lifecycles, and
publication.finalized_line < campaign.declared_line. Matching endpoints, an
arbitrary older event, or a comparison between independent wall clocks is not
such proof. This attachment does not alter CAS receipts or create a new
authority artifact.
A non-null adopted release also requires its provider to match the branch
readback provider,
its repository to match the adopted repository, its state to be published and
non-draft, its resolved tag
target to equal the subject head, and its uniquely named assets to have equal
cardinality and exact set equality with the complete live provider inventory.
Release assets added after review are separate publication
evidence and do not rewrite review time.
Apply closure.md only to current artifacts and
observations. Actuating authors actuating-closure-receipt/v1; Ledger neither
emits the verdict nor authors the receipt. $proof-patch may render a complete
result but cannot decide it. Complete delivery handoff before source-memory
evaluation; memory admission cannot delay, invalidate, or roll back closure.
The Axiomatic Construction gate applies prospectively when selecting a new or
successor Construction. A non-selecting route may consume a valid pre-feature
v3 Construction lacking basis provenance, record that provenance as
unavailable, and award it no basis proof. Any resulting mutation or selection
requires fresh axiomatization and a successor Construction.
Fail closed
Always block on stale or missing authority, Goal, Construction, or clean
committed subject; unrelated dirty state before operation preparation; a
successor commit whose parent, changed paths, scope, or cleanliness does not
exactly realize the selected edit; a prospective material selection with a
missing, stale, inconsistent, blocked, or unresolvedly underdetermined basis;
an incumbent-independent marker without a frozen derivation and Universalist
lowering; unresolved accepted or blocked Counterexamples; undispositioned
material review accretion; incomplete proof or retirement; missing lineage; a
public effect outside Ship; or any supporting skill, Ledger, or executor taking
Actuating's semantic authority.
For final-closeoutcomplete, also block on stale or missing publication
identity, CAS tuple mismatch, unresolved request-local recovery, fewer than five
current-published-subject standard cleans, or missing required Ship evidence
(SHIP-v1 or SHIP-ADOPTION-v1).
ready-to-ship requires neither publication nor review evidence.
local-implementationcomplete requires neither and rejects them as
inapplicable.
