00006The human maintainer accepted APG3’s blocked disposition as truthful evidence about its evaluation environment while rejecting the interpretation that APG must remain skill-free until Superpowers can be removed globally or a clean A/B harness exists.
APG4 was authorized to retain Superpowers for other repositories, make it reference-only for APG through root instructions, bootstrap multiple useful but provisional skills, accept structural, scenario, independent-review, and dogfooding evidence without a superiority claim, and defer clean comparative evaluation and decommissioning.
APG3 remains blocked. It stopped before plan freeze, baseline execution,
candidate authoring, or comparison because the exposed environment lacked an
isolated skill profile, observable invocation event, and controlled candidate
availability and loading. APG4 does not rewrite that result as a failed skill,
an adequate baseline, or evidence against later evaluation.
The root repository policy states semantically:
superpowers:* skill governs the workflow unless a human-authorized task
explicitly names one for inspection or comparison;This is a behavioral policy, not proof that the plugin is mechanically unloaded or absent from model context. The policy was already present in the clean fetched APG4 baseline and was preserved unchanged.
Accepted ADR 0003 distinguishes source availability from process authority,
permits provisional authorship before clean comparison, defines bootstrap,
provisional, evaluated, stable, and deprecated maturity, preserves APG3
history, and requires explicit coverage and rollback gates before Superpowers
decommissioning.
The APG v0.1 bootstrap model defines skill acceptance, public/private evidence,
project-owned parameters, independent removal, dogfooding, decommissioning, and
later clean-evaluation requirements. No APG4 skill can become stable until it
has succeeded in more than one real repository and survived explicit post-
Superpowers review.
| Skill | Maturity | Scenario result | Corrections | Independent disposition |
|---|---|---|---|---|
composing-bounded-worker-assignments |
provisional |
Pass | 0 | Accept provisional |
designing-significant-changes |
provisional |
Pass | 0 | Accept provisional |
planning-repository-work |
provisional |
Pass after rerun | 1 | Accept provisional |
implementing-with-test-discipline |
provisional |
Pass | 0 | Accept provisional |
debugging-systematically |
provisional |
Pass after rerun | 1 | Accept provisional after correction review |
reviewing-and-verifying-repository-work |
provisional |
Pass with fixture limitation | 0 | Accept provisional |
Omitted skills: none.
No skill authorizes delegation, expands human-approved scope, requires managed reports from ordinary workers, embeds private conventions, or claims that Codex is incapable without it.
Six fresh read-only workers performed 18 deliberate public-safe walkthroughs: six positive-use cases, six non-trigger cases, and six edge or stop cases. All 18 reached their frozen disposition. The planning and debugging leaves each received one bounded wording correction and a complete three-scenario rerun with no regression or remaining material defect.
This was deliberate skill application rather than automatic-trigger telemetry. No baseline condition was run, no clean causal comparison was attempted, and no superiority claim is made.
Fresh non-author reviewers assessed every skill against trigger precision, authority, proportionality, completeness, project-policy separation, public/private safety, source-expression independence, evidence, rollback, and ADR 0003.
They found one material issue, resolved in the debugging skill: evidence was incorrectly phrased as authorizing correction. The corrected text separates evidentiary support from assignment and project-owned mutation authority, passed all scenario reruns, and passed narrow independent re-review.
Two minor clarity notes remain without correction: the implementation and review skills rely on the controlling bootstrap and project policy for rollback or recovery rather than naming those parameters in every local list. The review positive fixture also tested evidentiary restraint rather than substantive finding quality because it did not include a concrete synthetic diff and test output.
No blocker or material finding remains.
The transition map covers using-superpowers, design, planning, plan execution,
test discipline, debugging, verification, review request and response, parallel
and subagent development, worktrees, branch completion, and skill authoring.
Each routes to an APG v0.1 skill, native Codex capability, repository policy,
intentional deferral, or intentional rejection.
The map preserves useful engineering values while removing universal ceremony, fixed templates, mandatory skill chains, and project-specific mechanics. It does not imply that Superpowers has been decommissioned.
Public provenance identifies generalized maintainer-authored practices, RepoMap, Superpowers under the MIT License, and the public Agent Skills specification as applicable. APG4 uses independently written synthesis and introduces no copied or adapted Superpowers expression or new notice requirement. APG’s own distribution license remains deferred.
Exact repository snapshots, source mappings, scenario artifacts, hashes, correction details, and reviewer dispositions remain in publication-excluded evidence. Public files do not link into that tree and remain independently understandable.
The final documentation and skill gate records:
| Check | Result |
|---|---|
| Development and read-only source baselines | Passed. Development main matched its fetched remote before editing; the source repository remains clean and equal to its fetched remote. |
| Authorized APG4 scope and changed-file classification | Passed. All 22 staged paths are authorized APG4 documentation, publication-excluded evidence, indexes, or the six skill leaves; no unstaged change remains. |
| Superpowers policy and no mechanical-disable claim | Passed. The root reference-only policy is present and is characterized only as behavioral. |
| Exact skill count, direct-child shape, frontmatter, names, trigger descriptions, non-triggers, and local links | Passed. Six exact leaves passed the pinned leaf validator and the repository structural gate; no support directory exists. |
| Authority, worker-report, and project-parameter boundaries | Passed. Semantic inspection and independent per-skill review found no delegation authorization, scope expansion, or ordinary-worker managed-report requirement. |
| Markdown headings, fences, and local links | Passed across all 52 Markdown files. |
| Public confidentiality and public-to-private dependency scan | Passed. Public files contain no private source identity, snapshot, local path, username, or link into private/. |
| ADR and exit namespaces and indexes | Passed. ADR sequence is 0001–0003; exit sequence is 00001–00006; filenames, titles, and indexes agree. |
| Provisional maturity and prohibited-claim scan | Passed. All six skills remain provisional; clean causal, statistical, universal, stable, and superiority claims are explicitly denied or deferred. |
| No prohibited plugin, profile, adapter, framework, dependency, runtime, scheduler, registry, taxonomy, publication, license, or extra-skill artifact | Passed by exact staged-path and artifact review. |
| External and project license boundaries | Passed. Superpowers remains external MIT-licensed evidence; APG’s own distribution license remains deferred. |
| Independent complete staged-diff review | Passed. One fresh non-author reviewer independently reproduced the structural, scenario, transition, sequence, privacy, parity, and whitespace evidence; found no blocker, material finding, or new minor finding; and accepted the candidate staged diff. |
git diff --check |
Passed. |
git diff --cached --check |
Passed. |
| Source tests and compilation | not run; documentation and skill content only |
No decommissioning occurred. Readiness remains false until every material workflow is mapped, APG succeeds in this repository and another real project, no authority, privacy, destructive-action, or completion-evidence regression remains, an uninstall rollback plan and preserved source snapshot exist, the human explicitly approves decommissioning, and post-decommission smoke passes.
APG4 supplies transition coverage and a preserved source snapshot, but it does not satisfy the real-use, regression, rollback-plan, human-decision, or smoke requirements.
Deliberately use each retained skill when its trigger naturally occurs in APG
and at least one additional real repository. For every use, record the trigger
decision, project-owned parameters, artifact or stop disposition, correction,
authority and privacy result, verification run, and checks intentionally not
run. Include real non-trigger observations. Add a concrete public-safe review
fixture and a schema or contract-change implementation case before requesting
any maturity beyond provisional.
External review is requested to accept APG4 as Complete — six provisional
skills bootstrapped, retain all six skills at provisional, and authorize only
the recorded dogfooding and external-review next action.
That disposition does not promote a skill, decommission Superpowers, resume APG3, publish APG, choose a license or taxonomy, or authorize another implementation phase.
External review returned correction-required: APG4 authored and validated six
canonical leaves under skills/ but did not install the .agents/skills/
repository projection required for Codex discovery. The omission did not change
the six procedures, 18 scenario dispositions, two corrections, independent
reviews, maturity, provenance, or no-decommission result.
APG4A later added the six-link projection as a forward-only integration correction. No APG4 scenario or review is rewritten as though the projection existed during that work, and no APG4 or APG4A record claims that explicit or automatic invocation occurred.