Partial — Bash, Bats, and Zsh profiles implemented; ZUnit deferred.
APG18 and APG18A completed their accepted decisions, correction, review, commit, push, remote equality, and managed reports before APG19 began. APG19 accepts ADR 0014, retains three provisional profiles, and records one truthful disposition for each candidate:
| Candidate | Disposition | Candidate corrections |
|---|---|---|
bash-language-profile |
retained-provisional |
1 |
bats-test-profile |
retained-provisional |
1 |
zsh-language-profile |
retained-provisional |
1 |
zunit-test-profile |
deferred-source-or-version |
0 |
Bash and Zsh retain distinct language owners. Bats owns test evaluation, capture, assertions, hooks, TAP/file descriptors, isolation, parallelism, and harness cleanup without duplicating Bash semantics. ZUnit retains a future test-harness ownership reservation but no current operational skill. No generic all-shell profile is adopted.
A task selects process, language, and test owners independently. Ordinary Bash or Zsh work does not require a test profile. Bats work adds the Bash profile only when production-language semantics are material. No profile grants action authority.
Sources were inspected on 2026-07-21. GNU Bash 5.3 documentation and patches through patch 15, ShellCheck 0.11.0 factual rule evidence, bats-core 1.13.0, and the official Zsh 5.9.2 release and manual support the retained profiles. Bash is GPL-3.0-or-later; its manual is GFDL-1.3-or-later without invariant sections or cover texts. ShellCheck is GPL-3.0. bats-core is MIT. Zsh uses its permissive distribution notice with file-specific terms for contributions.
ZUnit’s latest release is v0.8.2 from 2018. APG19-ZUNIT-SOURCE-02 identifies
the mutable default-branch observation from 2020; it was documentation-only and
has no semantic revision. Historical upstream runtime evidence ends at Zsh
5.3.1. Local use cannot establish current stable-Zsh support. Re-entry requires
current maintained upstream/runtime evidence and fresh APG verification, or
separate authorization for a legacy-only version-bounded profile.
APG uses independently written synthesis and copies or adapts no source or private expression or code. Versions and tools are evidence, not universal project requirements. Each retained profile defines refresh, deprecation, independent removal, and no-private-restoration boundaries.
| Signal | Green | Yellow | Orange | Red |
|---|---|---|---|---|
| Script physical lines | <= 150 |
151–300 |
301–500 |
>= 501 |
| Function/top-level command count | <= 15 |
16–25 |
26–40 |
>= 41 |
| Decision paths | <= 4 |
5–7 |
8–12 |
>= 13 |
| Nesting depth | <= 2 |
3 |
4 |
>= 5 |
| Fixed positional parameters | <= 4 |
5–6 |
7–9 |
>= 10 |
| Mutable globals/cross-function state | 0–1 |
2–3 |
4–6 |
>= 7 |
| Pipeline/process graph breadth | 0–2 |
3–4 |
5–7 |
>= 8 |
| Independent responsibility families | 1 |
2 |
3 |
>= 4 |
| Signal | Green | Yellow | Orange | Red |
|---|---|---|---|---|
| Test-file physical lines | <= 200 |
201–350 |
351–600 |
>= 601 |
| Test count | <= 12 |
13–24 |
25–40 |
>= 41 |
| Maximum test-body commands | <= 15 |
16–25 |
26–50 |
>= 51 |
| Maximum setup/teardown/bootstrap commands | <= 12 |
13–20 |
21–35 |
>= 36 |
| Maximum helper commands | <= 20 |
21–35 |
36–50 |
>= 51 |
| Shared fixture/global-state owners | 0–1 |
2–3 |
4–5 |
>= 6 |
| Maximum concurrent background child groups | 0 |
1, owned |
2–3, owned |
>= 4, or any unowned/leaking child |
| Independent responsibility families | 1 |
2 |
3 |
>= 4 |
| Signal | Green | Yellow | Orange | Red |
|---|---|---|---|---|
| Script physical lines | <= 200 |
201–300 |
301–500 |
>= 501 |
| Commands in one function or top-level region | <= 15 |
16–25 |
26–40 |
>= 41 |
| Decision points | <= 3 |
4–6 |
7–10 |
>= 11 |
| Maximum control/subshell nesting | <= 2 |
3 |
4–5 |
>= 6 |
| Positional parameters | <= 3 |
4–5 |
6–9 |
>= 10 |
| Distinct option mutations in one scope | <= 2 |
3–5 |
6–9 |
>= 10 |
| Mutable global/cross-function parameters | <= 2 |
3–5 |
6–9 |
>= 10 |
| Autoload/module/hook/ZLE/completion breadth | <= 2 |
3–5 |
6–10 |
>= 11 |
| External process/pipeline families | <= 2 |
3–5 |
6–9 |
>= 10 |
| Independent responsibility families | 1 |
2 |
3 |
>= 4 |
The following research bands are not an implemented APG contract:
| Signal | Green | Yellow | Orange | Red |
|---|---|---|---|---|
| File physical lines | <= 150 |
151–300 |
301–500 |
>= 501 |
| Tests per file | <= 8 |
9–15 |
16–24 |
>= 25 |
| Commands in one test body | <= 12 |
13–20 |
21–35 |
>= 36 |
| Largest setup/teardown/bootstrap body lines | <= 25 |
26–50 |
51–80 |
>= 81 |
| Largest helper function lines | <= 20 |
21–35 |
36–50 |
>= 51 |
| Mutable shared fixture/state domains | <= 3 |
4–6 |
7–10 |
>= 11 |
| Concurrent cleanup-owned children/jobs | <= 1 |
2–3 |
4–5 |
>= 6 |
| Independent responsibilities | 1 |
2 |
3 |
>= 4 |
Every retained profile passed its twelve shared structural cases, twelve domain-specific cases, exact first values for every threshold band, and the upper-Orange composite case after one bounded correction. One Red remains Red; three coupled Yellow signals normally become Orange; two coupled Orange signals over one owner are presumptively Red absent an accepted cohesive exception.
The correction closed a composite-monster loophole, made state/process
measurements reproducible, added Bats run/pipeline false-confidence behavior,
and tightened Zsh’s initially unsupported size disparity. Semantic Red stops
cover dynamic execution, destructive targets, protected-data disclosure,
unsafe process/cleanup ownership, false-pass assertions, startup or option
mutation, and meaningful crisis-level growth as applicable.
Generated, vendored, fixture, snapshot, migration, compatibility, and data-driven artifacts are classified before structural action. A coherent matrix may receive only the bounded exception stated by its profile. Existing Red legacy code may receive the smallest safe fix without a new responsibility or meaningful growth; a major feature remains Red.
Each retained profile used seven read-only cases across APG and at least one additional maintained repository, with classified RepoMap examples supplying non-executable semantic and false-positive evidence. The 21-case retained ledger covers Green, Yellow, Orange, Red or near-Red, legacy-small-fix, classified/coherent-large-artifact, and semantic safety or compatibility behavior. No target source or test was executed or changed as part of dogfood.
Private development now contains twelve canonical skills, twelve relative checked projections, six stable process rows, six provisional rows, and eleven routable non-router capability-map entries. The schema remains version 1. Focused tests cover exact map/catalog alignment, routes, profile headings and thresholds, process/domain pairing, stops, known-unmanaged handling, and the twelve-skill checker count.
apg-project-skills, apg-user-skills, and public release continue to manage
exactly the six stable v0.2 process skills. Public v0.2.0 and the active
public-backed integration remain 6/6/6. No project/user state schema changes.
Focused red/green evidence preceded implementation. The resulting tree passes 22 Bats cases; 25 skill-library, 8 release, and 8 user unit cases; and 28 project, 58 checker, 32 release, and 39 user integration cases. Both exact public-v0.1-conditioned suites ran without a skip. Development checker text and JSON pass at 12/12/12; the fresh public-v0.2 checkout remains 6/6/6. Python compilation, Bash syntax for the Bash commands and helper, and all six command help modes pass. Documentation, privacy, schema, external-state, and complete-diff results are recorded in the APG19 exit after fresh resulting-tree review.
Independent source calibration covered Bash, Bats, Zsh, and ZUnit. A separate private-guidance synthesis preserved project/private ownership. The shared design critic found no rejection blocker and produced the one bounded correction applied to each retained candidate. Final non-author semantic, ownership, integration, and complete-diff reviews govern terminal acceptance.
The three retained profiles begin provisional; no existing maturity changes.
ZUnit remains deferred. Thresholds are calibrated guidance, not lint rules.
Exact tools, versions, runners, helpers, commands, platforms, CI, coverage, and
exceptions remain repository-owned.
No root guidance was removed, no private skill was decommissioned, and no public release or active-integration update occurred. No Go, Ruby, Nix, PostgreSQL, SQLite, or roadmap-to-manager-prompt skill was implemented. Application smoke was not run and remains deferred to APG23/APG24. APG20 did not begin and is not authorized by this evaluation.
APG19A accepted every substantive APG19 disposition and reproduced one material Bats fallback-count defect: bats-core v1.13.0 recognizes a supported comment function form that the APG19 wording could exclude as a comment. APG19A applies one forward correction so the test-count band uses runner-recognized native and supported comment-form declarations, while excluding incidental static text and not authorizing target evaluation merely to count tests. Thresholds, all other profile semantics, maturity, catalog shape, ZUnit deferral, distribution, and application-smoke boundaries remain unchanged.
APG22B later closes the APG19 ZUnit deferral by retaining a provisional profile only for exact ZUnit v0.8.2 with Zsh 5.9.2 in the tested environment. The exact ZUnit v0.8.2 with Zsh 5.3.1 pair is unsupported, and no version range is claimed. This later disposition does not rewrite the historical APG19 result.