This map records how APG v0.1 preserved useful engineering behavior during the transition away from Superpowers. It remains coverage and disposition evidence after the maintainer’s completed decommission; it is not active Superpowers workflow policy.
Superpowers is external MIT-licensed evidence. APG uses independently written
synthesis and does not copy its templates, slogans, diagrams, mandatory skill
chains, or project-specific mechanics. APG’s six skills remain provisional.
APG7 added project-local projection rollback and a human decommission runbook.
The maintainer subsequently decommissioned Superpowers and completed bounded
post-decommission smoke; preserved source remains reference evidence rather
than installation or workflow authority.
| Superpowers skill | Primary APG disposition | Value preserved | Ceremony removed or rejected | Project-owned or native behavior | Maturity and remaining gap |
|---|---|---|---|---|---|
using-superpowers |
Repository instruction; universal bootstrap intentionally rejected | Explicit awareness that focused procedures may exist | Mandatory skill invocation before every response or action, fixed priority, rationalization policing, and compulsory skill chaining | Current task instructions and APG repository policy select the workflow; Codex discovers and loads applicable skills through its native environment | Repository policy is active. Superpowers was subsequently decommissioned; preserved source has no workflow authority. |
brainstorming |
APG v0.1 replacement: designing-significant-changes |
Resolve consequential uncertainty, compare alternatives, record decisions and non-goals | Universal design gate, fixed question cadence, required option counts, approval loops, fixed artifact locations, and mandatory transition to planning | Decision authority, artifact form, approval, architecture, compatibility, and implementation handoff | provisional; needs real-project trigger and non-trigger evidence. |
writing-plans |
APG v0.1 replacement: planning-repository-work |
Executable sequencing, affected surfaces, dependencies, evidence, acceptance, and handoff | Fixed minute-sized tasks, complete code in each step, universal TDD and commit cadence, fixed template/path, and execution menu | Plan destination, task size, Git workflow, test commands, ownership, and stop conditions | provisional; needs dogfooding across code and mixed-artifact plans. |
executing-plans |
Native Codex capability plus repository policy; dedicated skill deferred | Review a plan before action, track progress, verify units, and stop on real blockers | Required Superpowers chain, forced branch-finishing transition, universal worktree assumptions, and literal execution of every plan step | Codex executes authorized plans; repositories own checkpoints, branch policy, replanning authority, and completion gates | No APG execution-controller skill. Reconsider only after repeated execution-specific failure not covered by plans and project policy. |
test-driven-development |
APG v0.1 replacement: implementing-with-test-discipline |
Behavioral evidence, defect reproduction, regression coverage, minimal implementation, and final verification | Universal red-green ritual, delete-and-restart mandates, test-every-function claims, fixed mocking doctrine, and absolute automation requirements | Test layers, commands, coverage, mock/fixture policy, generated/legacy treatment, and acceptable substitutes | provisional; APG7 supplies one real executable implementation observation; real bugfix, broader feature and refactor, and mixed-artifact evidence remain needed. |
systematic-debugging |
APG v0.1 replacement: debugging-systematically |
Reproduction, evidence gathering, boundary tracing, explicit hypotheses, controlled experiments, and narrow correction | Rigid phase count, unique-root-cause assumption, fixed failed-fix threshold, unsupported effectiveness claims, and automatic skill chaining | Diagnostic access, privacy, observation limits, live-system mutation, containment, rollback, and evidence retention | provisional; needs evidence for flaky, multi-cause, and unreproducible failures. |
verification-before-completion |
APG v0.1 replacement: reviewing-and-verifying-repository-work |
Claim-specific fresh evidence, full output interpretation, independent worker verification, and truthful unrun checks | Moralizing language, same-message evidence rule, command-only proof, and unconditional full-suite gates | Evidence freshness, required checks, environment state, privacy-safe reporting, Git/push verification, and acceptance actor | provisional; needs real completion claims and evidence-expiry cases. |
requesting-code-review |
APG v0.1 replacement: reviewing-and-verifying-repository-work |
Bounded review context, requirement-aware inspection, evidence-backed findings, and explicit disposition | Mandatory review after every task, compulsory subagent use, fixed SHA templates, and branch or pull-request assumptions | Review trigger, reviewer independence, artifact identity, severity, hosting platform, and correction authority | provisional; review initiation and artifact packaging remain project-parameterized. |
receiving-code-review |
APG v0.1 replacement: reviewing-and-verifying-repository-work |
Clarify ambiguous feedback, verify against the repository, correct incrementally, and disagree with evidence | Conversational-style mandates, fixed response script, and source-specific finding workflow | Finding ownership, response channel, re-review threshold, hosting mechanics, and final acceptance | provisional; needs conflicting-reviewer and false-positive dogfooding. |
dispatching-parallel-agents |
Manager-worker protocol plus native Codex capability; separate skill deferred | Independent problem domains, exclusive write ownership, focused assignments, and integrated verification | Same-response dispatch rules, fixed worker thresholds, and parallelism as a default | Codex provides the agent harness; the project and manager own authorization, worker count, isolation, cancellation, integration, and partial outcomes | Bounded assignment composition is provisional; parallel coordination remains policy/native behavior. |
subagent-driven-development |
composing-bounded-worker-assignments, manager-worker protocol, and native Codex capability |
Bounded briefs, blocker return, explicit worker outcomes, independent manager integration, and review of evidence | Fresh worker per task, fixed multi-review loops, per-task commits, model-tier heuristics, progress ledgers, and forced branch completion | Delegation selection, dispatch, harness lifecycle, commit ownership, integration, review depth, and final reports | Assignment skill is provisional; an execution-controller skill is intentionally deferred. |
using-git-worktrees |
Native capability and repository policy; universal worktree workflow intentionally rejected | Detect existing workspace ownership, avoid fighting the harness, and verify a clean baseline when material | Automatic dependency setup, mandatory worktree creation, fixed directory selection, default .gitignore mutation, and universal baseline test assumptions |
User preference, repository branch/worktree policy, native worktree tools, setup commands, dependency authority, and cleanup | No APG worktree skill planned. Harness-specific guidance may be evaluated only on demonstrated need. |
finishing-a-development-branch |
Repository policy plus native Git capability; delivery skill deferred | Fresh verification, workspace ownership checks, explicit destructive authority, and post-integration verification | Fixed option menus, main/origin assumptions, unqualified pull, automatic prune/delete, and one branch model |
Branch, pull request, merge, push, release, cleanup, and destructive confirmation policy | Review/verification is provisional; a general delivery skill remains a later candidate. |
writing-skills |
Project model, provenance policy, bootstrap model, and skill index; reusable authoring skill deferred | Precise triggers, progressive disclosure, positive/non-trigger scenarios, source provenance, maintenance, and rollback | Universal skill-TDD doctrine, fixed sample and word thresholds, persuasion catalogues, deployment checklist, and push/PR workflow | APG decisions own maturity and evaluation; repositories own installation and publication | APG4 supplies a phase-specific contract, not a general authoring skill. Reconsider after repeated validated need. |
The materially used engineering loop is covered as follows:
Project-local APG discovery can now be installed, adopted, checked, and removed through one focused command with Git-local ownership. This coverage removes no plugin and creates no skill runtime, scheduler, registry, adapter, or generalized execution controller.
The maintainer has decommissioned Superpowers. The preserved workflow map, successful APG and RepoMap use, source snapshot, tested project-local rollback, human runbook, explicit human decision, completed action, and bounded fresh RepoMap smoke satisfy the v0.1 closeout gate. The smoke discovered six skills, successfully applied the review skill, passed managed checks, preserved the target repository, and did not use Superpowers as authority.
This disposition does not delete the historical source, authorize restoration,
prove automatic selection or comparative superiority, or promote a skill.
Every skill remains provisional; ADR 0006 assigns individual post-Superpowers
review to APG13.