Context
Measured on main before this change:
|
|
lib/ |
1.0 MB, 165 files |
docs/verification/ |
31 MB |
total docs/ |
32 MB |
So 97% of the documentation this template ships is acceptance evidence about
issues closed in this repository. Someone who clicks "Use this template"
inherits 31 MB of logs about work they never saw.
The same applies to several guards. tool/check_epic_coverage.dart enforces
that every directory maps to an epic:* label; scripts/dev/check_issue_refs.sh
plus .github/workflows/issue-refs.yml require Refs koniz-dev/flutter-starter#N
in every commit and pull request body. A fork has neither epics nor this
repository's issue numbers, so those are not merely dead weight - the issue-refs
workflow would reject every pull request the adopter opens, and the epic tool
shells out to scripts/bootstrap-issue-labels.sh, so a fork's first
flutter test would go red.
The app code itself is not the problem and is not in scope here.
The rule this applies
A starter template proves the mechanism is correct; it does not prove the
product runs, and it does not carry the maintainer's own process. The test,
added to CLAUDE.md by this change:
Would an adopter notice if this were missing?
Yes -> it ships. No -> it belongs to this repository only, and the strip
variants should drop it.
Acceptance criteria
dart run tool/strip_sample_features.dart --apply produces a tree with no
docs/verification/, no tool/check_epic_coverage.dart, no
scripts/dev/check_issue_refs.sh, no .github/workflows/issue-refs.yml,
no .claude/agents/, no scripts/bootstrap-issue-labels.sh and no
docs/issue-workflow.md. The same holds for --remove-tasks and
--remove-feature-flags.
- Guards that protect the adopter's own app survive the strip:
tool/check_env_assets.dart, tool/check_docs.dart,
tool/doc_signatures.dart, tool/doc_symbols.dart, ci.yml,
strip-smoke.yml and the git hooks.
- All three stripped trees pass
flutter pub get, flutter analyze and
flutter test cleanly, and dart run tool/check_docs.dart reports zero
broken links on a stripped tree - nothing removed may leave a dangling
reference.
strip-smoke.yml asserts the removals actually happened. analyze and
test stay green if the strip silently stops removing, so the assertion is
what makes criterion 1 enforceable. Demonstrate it fails on an unstripped
tree.
- Sections stripped from inside a file a fork keeps are marked in place
rather than tracked by a second list that can rot. Unbalanced markers are a
hard error reported before anything is deleted.
- Nothing is deleted from this repository - only the stripper learns to drop
it. This repo keeps its own process.
CLAUDE.md carries the scope test above, and it is itself marked
process-only.
./scripts/dev/audit_template.sh exits 0.
Context
Measured on
mainbefore this change:lib/docs/verification/docs/So 97% of the documentation this template ships is acceptance evidence about
issues closed in this repository. Someone who clicks "Use this template"
inherits 31 MB of logs about work they never saw.
The same applies to several guards.
tool/check_epic_coverage.dartenforcesthat every directory maps to an
epic:*label;scripts/dev/check_issue_refs.shplus
.github/workflows/issue-refs.ymlrequireRefs koniz-dev/flutter-starter#Nin every commit and pull request body. A fork has neither epics nor this
repository's issue numbers, so those are not merely dead weight - the issue-refs
workflow would reject every pull request the adopter opens, and the epic tool
shells out to
scripts/bootstrap-issue-labels.sh, so a fork's firstflutter testwould go red.The app code itself is not the problem and is not in scope here.
The rule this applies
A starter template proves the mechanism is correct; it does not prove the
product runs, and it does not carry the maintainer's own process. The test,
added to
CLAUDE.mdby this change:Yes -> it ships. No -> it belongs to this repository only, and the strip
variants should drop it.
Acceptance criteria
dart run tool/strip_sample_features.dart --applyproduces a tree with nodocs/verification/, notool/check_epic_coverage.dart, noscripts/dev/check_issue_refs.sh, no.github/workflows/issue-refs.yml,no
.claude/agents/, noscripts/bootstrap-issue-labels.shand nodocs/issue-workflow.md. The same holds for--remove-tasksand--remove-feature-flags.tool/check_env_assets.dart,tool/check_docs.dart,tool/doc_signatures.dart,tool/doc_symbols.dart,ci.yml,strip-smoke.ymland the git hooks.flutter pub get,flutter analyzeandflutter testcleanly, anddart run tool/check_docs.dartreports zerobroken links on a stripped tree - nothing removed may leave a dangling
reference.
strip-smoke.ymlasserts the removals actually happened.analyzeandteststay green if the strip silently stops removing, so the assertion iswhat makes criterion 1 enforceable. Demonstrate it fails on an unstripped
tree.
rather than tracked by a second list that can rot. Unbalanced markers are a
hard error reported before anything is deleted.
it. This repo keeps its own process.
CLAUDE.mdcarries the scope test above, and it is itself markedprocess-only.
./scripts/dev/audit_template.shexits 0.