Accepted
2026-07-20
The human maintainer’s APG17 assignment accepts this decision because the candidate met the frozen-scenario, bounded-dogfood, privacy, ownership, and review retention threshold. Acceptance authorizes the APG17 development artifacts only. It does not authorize source migration, root reduction, private skill removal, a language profile, manager-prompt capability, publication, maturity promotion, or APG18.
Repository and private instruction corpora often combine universal authority, reusable procedure, language guidance, exact project commands, machine state, personal preferences, externally sourced material, and stale rules. Neither a source file nor a private skill boundary establishes a coherent public capability. Rewriting first risks weakening authority, copying expression without established rights, publishing private context, duplicating an existing owner, and removing the only current source before discovery and rollback are proven.
Ordinary reasoning correctly classified all sixteen APG17 scenario families, but was less repeatable for mixed-unit splitting, coverage accounting, migration dependencies, rollback, and explicit source preservation. Native Codex skill-authoring guidance constructs a technically valid skill after an owner, trigger, scenario contract, and destination are accepted; it does not own the preceding corpus-classification decision.
APG17 therefore evaluates one bounded owner for this pre-rewrite decision.
synthesizing-repository-guidance owns classification and disposition of a
dense, duplicated, mixed-scope, private, or source-derived guidance corpus
before rewrite or migration. Its conceptual boundary is:
classify and disposition; do not rewrite or migrate
Related owners remain distinct:
The synthesis skill proposes owners and prerequisites. It does not replace any of these owners or expand their authority.
The skill splits mixed content into coherent units and records, for each unit:
The accepted vocabulary is:
retain-root
route-existing-owner
synthesize-existing-owner-correction
candidate-new-skill
candidate-language-profile
core-documentation
deterministic-tool-candidate
project-owned
keep-private
evidence-only
defer
reject
split-required
blocked-source-or-rights
A Markdown table plus bounded detail is sufficient. APG does not require JSON, a registry, a database, or a universal migration schema.
Source files, private skill directories, and repeated prose do not establish public capability boundaries. Repetition is evidence to inspect, not adoption proof. Existing owners are preferred, and a new skill must satisfy the accepted new-capability threshold. Mixed units are split before owner selection.
Every source instruction remains in place until a later, separately authorized migration supplies shadow behavior, discovery, override, cutover, rollback, and restoration evidence. A disposition is not implementation authority.
Public APG receives no medical, cognitive, diagnostic, deeply personal,
accessibility, productivity, machine-specific, or private-integration detail.
Raw sensitive guidance is not retained even in APG’s publication-excluded
evidence. keep-private is a valid terminal disposition.
A reusable public principle motivated by private context requires independent engineering evidence and independently written expression that does not depend on or reveal the private detail.
Source identity, derivation, reuse eligibility, and required notices are established before copied or adapted work. Missing identity or rights blocks that work. Bounded independently written analysis from observable facts may continue only when allowed, and no actor invents provenance or permission.
Source packaging and authoritative tone do not create an APG owner. Existing owners or independently written synthesis are preferred.
The synthesis skill produces a reviewable disposition map and public-safe aggregate. It does not author the final skill, root file, language profile, policy, tool, or migration. It does not rewrite, remove, publish, decommission, cut over, or authorize a candidate. A later accepted phase owns each proposed implementation or migration.
The APG workflow router may route a mixed-guidance ownership or migration-plan
case to synthesizing-repository-guidance. The synthesis leaf does not route
ordinary repository work, duplicate router behavior, or become a mandatory
session entry point. Known skill authoring, simple documentation edits, and
accepted implementation or debugging tasks remain synthesis non-triggers.
The historic manager-prompt sample shows a plausible distinct boundary for constructing a top-level Codex assignment from an already human-approved roadmap. APG17 records that as a future capability candidate only. It creates no skill, final name, prompt, or implementation and does not preauthorize later work. Human-approved roadmap authority remains external.
Viable for simple or one-off cases. Rejected as the only owner for mixed corpora because the baseline showed inconsistent coverage, migration, restoration, and source-preservation fields across the materially complex families.
Rejected. File and private-skill boundaries mix capabilities, project policy, machine state, preferences, and duplicate existing procedure.
Rejected. Native technical authoring applies after the semantic owner and contract are accepted. Moving it earlier confuses valid file construction with capability adoption.
Rejected. A broad standards owner would combine unrelated triggers, duplicate repository authority and existing leaves, and encourage private or language-specific detail to become universal procedure.
Rejected. APG17 needs a bounded decision record, not runtime, registry, state, automatic rewrite, or destructive cutover machinery.
Accepted. One narrow pre-rewrite procedure supplies stable unit splitting, ownership, rights, privacy, migration, rollback, and source-preservation fields while leaving implementation with current owners.
Safe and valid for personal, machine, sensitive, and private-repository units. Rejected as a universal disposition because some independently evidenced reusable behavior already has public owners or may justify a future capability.
Accepted as the migration policy. Existing owners remain preferred; only coherent uncovered problems enter the separately governed new-skill lifecycle.
provisional synthesis leaf and one additional
routable capability-map entry.keep-private or project-owned without a public counterpart.APG17 performs no guidance migration. A future migration must accept exact source scope, replacement owners, shadow behavior, discovery, overrides, cutover evidence, and byte-restorable rollback before removing any source.
If the provisional synthesis leaf becomes unsound, remove its canonical leaf, checked projection, catalog row, capability-map entry, and focused scenario contract. Restore the preceding current-development known-unmanaged set and every current-state catalog count, project-integration description, and distribution-compatibility statement to the prior seven-skill development shape. Preserve this ADR, evaluation, exit, and their historical outcome as evidence. Source guidance requires no restoration because APG17 does not change it.
provisional;