[brettheap]

Closes #939.
Closes #941.

Lane 2 filed both issues and suggested pairing them, because they share standard.py's ref_fault. One commit covers both, since each edits the same docstring and the same comment. cli.py is unchanged: the checks stand in standard.clone_url and standard.target, which --version (cli.print_version) and prepare already call before any line or git call names a value.

Where the rules come from (the launcher, openRepoShape at 64464d31)

• MISDRAWN_CHARACTERS (:501) and misdrawn_characters (:512): the 41 code points, named as U+XXXX, comma-separated, in code point order.
• checked_repo (:805): drawn as another, on the value as typed; then a leading - or a control character.
• ref_fault (:882, shape line :895) with ref_character_fault (:908): empty, length, drawn as another, control character, shape.
• The shape is setup.sh, install.sh and the openRepoShape command: each keeps its own copy of the old ref rule (*..* | *.lock), so a ref git refuses part by part (a.lock/b, a/.b, a//b) clones the standard before it fails #874's rule and sentence, the same bytes in setup.sh and install.sh. tests/conftest.py states it as REF_PARTS and REF_RULE_WORDS.

#939: what changed, by section

• --version printed a repository holding U+202E or U+200B as typed, exit 0. standard.py now carries the launcher's list as MISDRAWN_CHARACTERS, entry for entry, and misdrawn_characters, which names what a value holds. clone_url asks it first, on the repository as typed, and refuses in checked_repo's words. Then comes its old check, a leading - or a control character. --version calls clone_url before it prints, so it now exits 2 with the launcher's sentence and no byte of the value.
• A ref holding either was refused, but quoted as typed. ref_fault asks misdrawn_characters between the length and the control check, where the launcher asks it, in ref_character_fault's words. The shape refusal, which quotes the ref, is reached only by a value that passed all four checks.
• A run that reached the clone named the repository and handed git the URL. prepare calls target and clone_url before it looks for git, so --doctor, a scaffold run, --shape-ref and the four tools refuse before any git call. Measured by hand on a stamped copy with a logging git first on PATH, for --doctor ., Atlas --org acme and shape-doctor --json: a repository with U+202E, a ref with U+202E and the ref a.lock/b each made 0 git calls, with no E2 80 AE on either stream. A clean acme/r made 1 call (git clone).
• The table is a copy, held equal by a test. The package ships alone and reads no file of the launcher's. The entries are the launcher's hex strings, as written, so the parity test compares text with text, as install.sh's does. That also keeps the package pure ASCII, which test_the_package_is_pure_ascii requires.
• The words. ref_fault's docstring and the comment above CONTROL_RE now state the five checks in order and name the tests that hold the copy to the launcher.
• Decided, as the Direction asks: runs that never read the values are left as they are. A run beside a git checkout and --preflight do not refuse either value. Neither run prints a value or hands one to git (prepare and print_version return first beside a git checkout, and --preflight returns before prepare), so the narrower direction refuses nothing that could reach a line there. A third run reads only half: a scaffold run with a clean --shape-ref never asks $OPENREPOSHAPE_REF anything (target returns the typed ref first), so that value is never printed or sent, though the launcher refuses it. A check in main_openreposhape would also have to be repeated in run_tool for the four tools.
  • The claim holds beside a git checkout, and not beside a source tree with no .git, such as a ZIP, or an sdist installed with pip install -e .. beside_checkout checks for files only, so that tree's setup-project.py self-bootstraps and hands $OPENREPOSHAPE_REPO to git clone without the drawn-as-another check (checked_value, setup-project.py:881-882). That is the same at origin/main, so this PR does not regress it. It is now setup-project.py: under self-bootstrap, an $OPENREPOSHAPE_REPO holding U+202E or another character the launcher refuses is printed and handed to git clone, because its check has no drawn-as-another test #953.

#941: what changed, by section

• The shape line. It now holds setup.sh, install.sh and the openRepoShape command: each keeps its own copy of the old ref rule (*..* | *.lock), so a ref git refuses part by part (a.lock/b, a/.b, a//b) clones the standard before it fails #874's rule: REF_RE, no .., no empty part (//, or a / at the end), no part that starts with . or ends in .lock, and no . at the end. That is setup-project.py's why_not_a_ref rule. I measured the issue's 3,110 names (every name of up to five pieces from a, 0, ., /, -, .lock that REF_RE takes). This rule disagrees with git check-ref-format --branch on 0 of them, and with the launcher's own ref_fault, lifted out and run by bash, on 0. origin/main's line disagrees with each on 1,122.
• A ref git refuses reached git. a.lock/b, a/.b, a//b, a/b/ and a. are now refused by target, before remote_commit and the clone: 0 git calls (see the tests below).
• The sentence is the launcher's, word for word (REF_RULE_WORDS). --shape-ref gets the same sentence with its own opening.
• The docstring and the comment state the order and the rule, and name the test.
• No test held the rule to git's. test_ref_fault_takes_a_ref_as_git_does_and_refuses_the_rest is now the fourth reader of REF_PARTS. It runs in-process with assert_a_copy_answers_as_git_does("standard.py", ...) and assert_the_refusal_states_the_ref_rule, and git is asked about each name, never a repository.

Tests added (tests/test_openreposhape_package.py), 115 cases

• test_version_refuses_a_repository_or_ref_no_line_can_print (34): the launcher's 12 UNPRINTABLE_VALUES and 5 order rows, each under C and C.UTF-8, through cmd.detached("--version"), checked with assert_refused_saying. The order rows are an escape then U+202E (repository and ref), U+202E then U+200B (named U+200B, U+202E, repository and ref), and a 259-character ref holding U+202E (refused for its length).
• test_version_prints_a_repository_holding_none_of_the_list (14): the launcher's NOT_CONTROLS print as typed, exit 0, under both locales.
• test_a_run_that_clones_refuses_the_value_before_any_git_call (21): --doctor ., a scaffold run and shape-doctor, times a repository with U+202E, a ref with U+202E and setup.sh, install.sh and the openRepoShape command: each keeps its own copy of the old ref rule (*..* | *.lock), so a ref git refuses part by part (a.lock/b, a/.b, a//b) clones the standard before it fails #874's five refs, with a logging git, gh and curl first on PATH: no call, and that one refusal.
• test_a_typed_shape_ref_is_refused_before_any_git_call (2): --shape-ref with U+202E, and with a.lock/b.
• test_the_misdrawn_table_is_the_launchers_list (1) and test_misdrawn_characters_names_the_launchers_and_nothing_else (1, over SHARING_ITS_BYTES).
• test_ref_fault_takes_a_ref_as_git_does_and_refuses_the_rest (36, REF_PARTS).
• test_a_ref_is_checked_before_anything_is_cloned: plus setup.sh, install.sh and the openRepoShape command: each keeps its own copy of the old ref rule (*..* | *.lock), so a ref git refuses part by part (a.lock/b, a/.b, a//b) clones the standard before it fails #874's five refs (5).
• test_help_answers_past_a_repository_and_ref_no_line_can_print, the launcher's, imported unchanged (1).
• call_loggers is factored out of test_a_bare_run_asks_nothing_of_git_gh_or_curl so the new tests share its fakes. That test's behaviour is unchanged.

Left out of scope

• Neither issue has a "Needs a ruling" or a "Documented limit" section.
• openreposhape package: openRepoShape --version prints a repository holding U+202E or U+200B as typed and exits 0, and a ref holding either is refused with the value as typed, where the launcher refuses both before it prints (#866) #939's "Found while measuring, not this issue's" finding is in another file: setup.sh's --shape-ref check has no drawn-as-another check, and quotes the value under C.UTF-8.
• Noticed, and fixed in round 1 (rv947's NIT-1). prepare asked the ref (target) before the repository (clone_url). The launcher's refuse_bad_values and the package's own --version ask the repository first, and openreposhape package: openRepoShape --version prints a repository holding U+202E or U+200B as typed and exits 0, and a ref holding either is refused with the value as typed, where the launcher refuses both before it prints (#866) #939's Direction says "the repository first". It now asks the repository first: see Fix round 1 below.
• Already there, and recorded in openreposhape package: ref_fault keeps the old ref rule, so it takes a.lock/b, a/.b, a//b, a/b/ and a., clones the standard in full and fails at git checkout, where the launcher, setup.sh and install.sh refuse them first (#874) #941. The package's --version refuses a ref the shape refuses (a.lock/b, as it already refused topic.lock). The launcher's --version asks a ref only its characters (openRepoShape: --version and two fetch refusals print $OPENREPOSHAPE_REPO and $OPENREPOSHAPE_REF with no check, so an escape or a bidi override in either reaches the terminal as typed, and a scaffold run sends its fetch with them #866) and prints it.
• openRepoShape, setup.sh and install.sh: under the C locale a $OPENREPOSHAPE_REPO URL or path, or a ref, with a letter past ASCII is refused as "a control character", so --version, --install and --preflight refuse what v1.1.0 ran #940 and openRepoShape: --doctor's shape_checkout runs git clone before checked_ref, so a ref it refuses (a.lock/b, a//b, one over 255 characters) costs a whole clone first, and where the clone fails the ref's refusal is never reached #942, lane 2's, are the launcher and are not touched. This module imports the launcher tests' UNPRINTABLE_VALUES, NOT_CONTROLS, MISDRAWN, SHARING_ITS_BYTES, LOCALES, ESCAPE, RLO, assert_refused_saying, repo_drawn_as_another and ref_drawn_as_another, and reads MISDRAWN_CHARACTERS=(...) out of openRepoShape. A launcher PR that renames or rewords those will fail here too, which is the parity working.

Test runs

Each run was a separate pytest process, one at a time, with a stub gh (exit 97) first on PATH, GH_TOKEN=bogus, an empty GH_CONFIG_DIR and GIT_SSH_COMMAND=false:
timeout 2400 /usr/bin/python3 -m pytest -q -p no:cacheprovider --basetemp=<scratch>/bt-<name> -o tmp_path_retention_policy=failed <file>

• tests/test_openreposhape_package.py at origin/main (64464d31): 58 passed in 20.37s
• tests/test_openreposhape_package.py at this head (21558e44): 173 passed in 28.66s
• tests/test_openreposhape_command.py: 447 passed in 46.44s
• tests/test_agents_md_codes.py: 47 passed in 1.75s
• tests/test_cli_reference.py: 23 passed in 136.99s (0:02:16)
• tests/test_update_add_codes_documented.py: 6 passed in 0.48s
• tests/test_repo_hygiene.py: 81 passed in 3.41s
• tests/test_no_auto_maintenance.py: 60 passed in 12.58s
• tests/test_validator_child_encoding.py (it names cli.py): 163 passed in 71.79s (0:01:11)
• tests/test_release_workflow.py (it names src/openreposhape/): 56 passed in 3.88s
• tests/test_setup_sh.py (a REF_PARTS reader, unchanged): 125 passed in 88.01s (0:01:28)
• tests/test_install_sh.py (a REF_PARTS reader, unchanged): 270 passed in 40.76s
• tests/test_shape_doctor.py: not run, because neither shape-doctor.py nor a helper it imports changed.

No new refusal code was added (the package's refusals carry none), so no code doc changes. No doc quoted the old ref sentence: at origin/main, git grep finds it only in standard.py.

Mutants

Each mutant was applied to standard.py and run against the whole of tests/test_openreposhape_package.py. Each is killed.

#	mutant	failed	killed by (among others)
m00	origin/main's standard.py	65	the U+202E and U+200B rows and 4 order rows under both locales (16), the 21 no-git-call runs, the 2 --shape-ref cases, the 2 table tests, 19 REF_PARTS rows and the 5 added refs. The control rows, the length row, NOT_CONTROLS, the help case and the 17 rows git takes pass there as well, as they should
m01	clone_url without the drawn-as-another check	11	test_version_refuses_...[repo-u+202e-C], test_a_run_that_clones_...[doctor-repo-u+202e]
m02	ref_fault without it	12	test_version_refuses_...[ref-u+202e-C], test_a_typed_shape_ref_...[u+202e]
m03	ref_fault asks the control check first	2	test_version_refuses_...[ref-an-escape-then-u+202e-*]
m04	clone_url asks the control check first	2	test_version_refuses_...[repo-an-escape-then-u+202e-*]
m05	the table without U+200B	10	test_the_misdrawn_table_is_the_launchers_list, ...[repo-u+200b-C]
m06	the range U+2028 to U+202E cut at U+202D	21	test_misdrawn_characters_names_the_launchers_and_nothing_else
m07	names in the value's order, not code point order	5	test_version_refuses_...[repo-two-at-once-*], [ref-two-at-once-*]
m08	a range's last code point left out	25	the table tests, every U+202E case
m09	code points named in lower case	25	the table tests, every U+202E case
m10	drawn-as-another asked before the length	2	test_version_refuses_...[ref-over-long-and-u+202e-*]
m11	the old shape line back, with the new sentence	34	test_ref_fault_takes_...[a.lock/b] and 12 other rows, the 5 refs in each run
m12	no empty-part check	10	test_ref_fault_takes_...[a//b], [a/b/]
m13	no check for a part that starts with .	8	test_ref_fault_takes_...[a/.b], [a/b/.c/d]
m14	.lock asked of the end of the name only	8	test_ref_fault_takes_...[a.lock/b], [release.lock/v1]
m15	no check for a name that ends in .	6	test_ref_fault_takes_...[a.], [a/b.]
m16	a part that ends in . refused (over-refusal)	1	test_ref_fault_takes_...[a./b]
m17	the first character lower case only (over-refusal)	1	test_ref_fault_takes_...[Feature/ABC-123]
m18	a part that starts with _ or - refused (over-refusal)	2	test_ref_fault_takes_...[fix/_tmp], [fix/-wip]
m19	the sentence without its empty-part clause	35	every refused REF_PARTS row, every shape refusal in a run

Merges

• git merge-tree --write-tree of this head with origin/main (64464d31): clean, tree f4a9282699321be04716eaf639a234dd4cd8261d, exit 0.
• At 2026-10-18T16:47Z, none of the 10 open PRs touched src/openreposhape/standard.py, src/openreposhape/cli.py or tests/test_openreposhape_package.py, so there was nothing else to merge with.

Fix round 1 (head c1de4c86, after rv947's review of 21558e44)

rv947 found 21558e44 READY: one SHOULD, which is this text, and NITs. The sections above describe 21558e44; this round changes the head as follows. One commit, 3 files, 38 insertions and 10 deletions.

• SHOULD-1 (text only). The "Decided" item claims "beside a git checkout" and no more. Beside a source tree with no .git (a ZIP, or an sdist installed with pip install -e .), that tree's setup-project.py self-bootstraps and hands $OPENREPOSHAPE_REPO to git clone without the drawn-as-another check. That is the same at origin/main, and it is now setup-project.py: under self-bootstrap, an $OPENREPOSHAPE_REPO holding U+202E or another character the launcher refuses is printed and handed to git clone, because its check has no drawn-as-another test #953.
• NIT-2 (text only). The "Decided" item names the third run that never reads a value: a scaffold run with a clean --shape-ref never asks $OPENREPOSHAPE_REF.
• NIT-1. In standard.py's prepare, repo = repository() and url = clone_url(repo) now come before wanted = target(stamp, typed_ref), with a comment and a docstring line that say the order. Nothing between them depended on the old order: clone_url reads only $OPENREPOSHAPE_REPO, and target reads the stamp, the typed --shape-ref and $OPENREPOSHAPE_REF. With both values bad, --doctor, a scaffold run and the tools now name the repository's fault first, as the launcher and the package's --version do.
  • Test: test_with_both_values_bad_a_run_names_the_repository_first in tests/test_openreposhape_package.py, mirroring the launcher's test_bad_values_are_refused_in_the_order_the_check_states. It runs --doctor ., Atlas --org acme and shape-doctor --json (3 cases) with a repository holding U+202E and a ref holding an escape, and wants exit 2, REFUSED: OPENREPOSHAPE_REPO holds ..., no mention of OPENREPOSHAPE_REF, no git call and neither value echoed.
  • Mutant: the old order back in prepare. It fails the 3 cases ([doctor], [scaffold] and [shape-doctor]) and nothing else: 3 failed, 174 passed.
• NIT-3. ("a/./b", False) is in REF_PARTS in tests/conftest.py, and the comment above the table counts the names the old line took as fourteen. git check-ref-format --branch a/./b answers fatal: 'a/./b' is not a valid branch name. The four files that read REF_PARTS (package, launcher, test_setup_sh.py, test_install_sh.py) each gain one case, and all pass.
  • Mutant: a part that is exactly . is taken (rv947's own2). Before this round it survived, with 173 passed. With the new row deselected it still survives here (176 passed, 1 deselected, the 176 including the 3 new cases above). With the row, it fails test_ref_fault_takes_a_ref_as_git_does_and_refuses_the_rest[a/./b] and nothing else: 1 failed, 176 passed.
• NIT-4. test_version_prints_a_repository_holding_none_of_the_list carries NO_NON_ASCII_INTO_GIT_BASH, imported from the launcher's tests, as the launcher's twin test_version_prints_a_path_a_utf_8_locale_prints does: it needs a UTF-8 stdout in the child, so it is skipped on Windows.
• NIT-5 needs no change here: whoever lands second reconciles, if openRepoShape, setup.sh and install.sh: under the C locale a $OPENREPOSHAPE_REPO URL or path, or a ref, with a letter past ASCII is refused as "a control character", so --version, --install and --preflight refuse what v1.1.0 ran #940 adds rows to NOT_CONTROLS or another imported table.

Test runs at c1de4c86, each a separate pytest process under the stub gh, GH_TOKEN=bogus, an empty GH_CONFIG_DIR and GIT_SSH_COMMAND=false, wrapped in timeout 2400:

• tests/test_openreposhape_package.py: 177 passed in 23.31s (173, the order test's 3 and the new row)
• tests/test_openreposhape_command.py: 448 passed in 36.76s
• tests/test_setup_sh.py: 126 passed in 69.01s (0:01:09)
• tests/test_install_sh.py: 271 passed in 27.20s
• tests/test_repo_hygiene.py: 81 passed in 1.84s
• tests/test_no_auto_maintenance.py: 60 passed in 5.79s

Merges: git merge-tree --write-tree of c1de4c86 with origin/main (64464d31) is clean, tree 0c09d9510683a514d5fa6ef2856b82928de35a4e. With #949's head (8c5d6051), which also edits tests/conftest.py, it is clean too, tree 1951cc0bbca9ed5634ee8c0a2d2b521310b42660.

Lane: openreposhape-3 (openRepoShape-3)
🤖 Generated with Claude Code

[sourcery-ai]

Reviewer's Guide

This PR hardens repository and ref validation to match the launcher and git: visually ambiguous Unicode is rejected by code point before values can be printed or passed to git, and refs are validated component-by-component before cloning. Extensive parity and integration tests verify exact ordering, wording, locale behavior, no-leak guarantees, and zero external calls on refusal.

Sequence diagram for preflight repository and ref validation

sequenceDiagram
    participant CLI
    participant Prepare
    participant Standard
    participant Git

    CLI->>Prepare: prepare()
    Prepare->>Standard: target(ref)
    Standard->>Standard: ref_fault(ref, what)
    alt ref_fault returns refusal
        Standard-->>Prepare: refusal
        Prepare-->>CLI: exit 2 without value or git call
    else ref passes
        Prepare->>Standard: clone_url(repo)
        Standard->>Standard: misdrawn_characters(repo)
        alt repository is unsafe
            Standard-->>Prepare: refusal
            Prepare-->>CLI: exit 2 without value or git call
        else repository passes
            Prepare->>Git: clone URL
            Git-->>Prepare: result
        end
    end

Loading

Flow diagram for ordered repository and ref checks

flowchart TD
    Start([Input repository or ref]) --> Kind{Value type}
    Kind -->|Repository| RepoUnicode[Check misdrawn_characters]
    RepoUnicode -->|Found| RepoReject([Refuse without echo or git])
    RepoUnicode -->|None| RepoLegacy[Check leading dash or control character]
    RepoLegacy -->|Invalid| RepoReject
    RepoLegacy -->|Valid| RepoAccepted([Repository accepted])
    Kind -->|Ref| RefEmpty[Check empty]
    RefEmpty -->|Invalid| RefReject([Refuse])
    RefEmpty -->|Valid| RefLength[Check length]
    RefLength -->|Over limit| RefReject
    RefLength -->|Valid| RefUnicode[Check misdrawn_characters]
    RefUnicode -->|Found| RefReject
    RefUnicode -->|None| RefControl[Check control character]
    RefControl -->|Found| RefReject
    RefControl -->|None| RefShape[Check REF_RE and component rules]
    RefShape -->|Invalid| RefReject
    RefShape -->|Valid| RefAccepted([Ref accepted])

Loading

File-Level Changes

Change	Details	Files
Rejects launcher-defined visually ambiguous characters in repositories and refs before any output or external command can expose them.	Adds an ASCII copy of the launcher's misdrawn-character code-point table and a helper that reports matching code points in canonical order. Checks repositories in clone_url and refs in ref_fault, preserving launcher validation order and refusal wording. Adds coverage for locale behavior, mixed invalid characters, length precedence, and no-leak/no-git-call guarantees.	src/openreposhape/standard.py tests/test_openreposhape_package.py
Aligns ref validation with git and the launcher's per-path-component shape rules.	Rejects empty path components, dot-prefixed components, components ending in .lock, and refs ending in a dot. Keeps validation before cloning and uses the launcher’s ref-rule sentence for failures. Cross-checks the package against git behavior and launcher fixtures, including previously accepted invalid refs.	src/openreposhape/standard.py tests/test_openreposhape_package.py
Expands package-level integration and parity tests to verify validation boundaries across commands.	Tests --version, --shape-ref, doctor, scaffolding, and shape-doctor with logging command stubs. Factors reusable command loggers and imports launcher fixtures for exact parity checks. Adds table-parity, refusal-wording, no-value-echo, and pre-git-call assertions.	tests/test_openreposhape_package.py

Assessment against linked issues

Issue	Objective	Addressed	Explanation
#939	Apply the launcher's misdrawn-character check to repository values, environment refs, and --shape-ref, using the same character list, refusal wording, and check order.	✅
#939	Refuse affected values before --version prints them or a package run prints, clones, or passes them to git, without echoing the values; decide whether runs that never read the values should also refuse them.	✅
#941	Update the package’s ref_fault to apply the launcher’s part-by-part Git ref rules and refuse invalid refs before any Git call, using the launcher’s refusal wording.	✅
#941	Add package tests that compare its ref acceptance and refusal wording with Git and the launcher, including refs the old rule accepted but Git rejects.	✅
#941	Document the corrected rule and its ordering in the package’s ref_fault docstring and related comment.	✅

Possibly linked issues

• openRepoShape: --version and two fetch refusals print $OPENREPOSHAPE_REPO and $OPENREPOSHAPE_REF with no check, so an escape or a bidi override in either reaches the terminal as typed, and a scaffold run sends its fetch with them #866: The PR adds the requested character table and checks in clone_url and ref_fault, refusing affected values before they are printed or passed to git.
• openreposhape package: ref_fault keeps the old ref rule, so it takes a.lock/b, a/.b, a//b, a/b/ and a., clones the standard in full and fails at git checkout, where the launcher, setup.sh and install.sh refuse them first (#874) #941: The PR implements this issue’s requested ref rule and sentence in standard.py, adds Git/launcher parity tests, and refuses listed refs before cloning.

---

Tips and commands

Interacting with Sourcery

• Trigger a new review: Comment @sourcery-ai review on the pull request.
• Continue discussions: Reply directly to Sourcery's review comments.
• Generate a GitHub issue from a review comment: Ask Sourcery to create an
  issue from a review comment by replying to it. You can also reply to a
  review comment with @sourcery-ai issue to create an issue from it.
• Generate a pull request title: Write @sourcery-ai anywhere in the pull
  request title to generate a title at any time. You can also comment
  @sourcery-ai title on the pull request to (re-)generate the title at any time.
• Generate a pull request summary: Write @sourcery-ai summary anywhere in
  the pull request body to generate a PR summary at any time exactly where you
  want it. You can also comment @sourcery-ai summary on the pull request to
  (re-)generate the summary at any time.
• Generate reviewer's guide: Comment @sourcery-ai guide on the pull
  request to (re-)generate the reviewer's guide at any time.
• Resolve all Sourcery comments: Comment @sourcery-ai resolve on the
  pull request to resolve all Sourcery comments. Useful if you've already
  addressed all the comments and don't want to see them anymore.
• Dismiss all Sourcery reviews: Comment @sourcery-ai dismiss on the pull
  request to dismiss all existing Sourcery reviews. Especially useful if you
  want to start fresh with a new review - don't forget to comment
  @sourcery-ai review to trigger a new review!

Customizing Your Experience

Access your dashboard to:

• Enable or disable review features such as the Sourcery-generated pull request
  summary, the reviewer's guide, and others.
• Change the review language.
• Add, remove or edit custom review instructions.
• Adjust other review settings.

Getting Help

• Contact our support team for questions or feedback.
• Visit our documentation for detailed guides and information.
• Keep in touch with the Sourcery team by following us on X/Twitter, LinkedIn or GitHub.

[brettheap]

(lane openRepoShape-3, 2026-10-18T17:08:26Z) RV947's independent review of #947 at 21558e44 (origin/main 64464d31): READY. No MUST, one SHOULD (the body's text, plus a follow-up issue), five NITs. RV947 is lane 3's opus reviewer, not the writer.

1. Parity, measured by RV947's own scripts, which read both files:

   • MISDRAWN_CHARACTERS equals openRepoShape:501 entry for entry, in order. Both expand to the same 41 code points.
   • misdrawn_characters, run over all of Unicode but the surrogates, names exactly those 41.
   • Run by bash under LC_ALL=C, the launcher's function and the package's agree on all 81 edge values: first-2 to last+2 of every range, all at once, and two at once. Names come in code point order.
   • The --version refusals are byte-identical to the launcher's in 11 cases, under C and C.UTF-8.

2. The ref rule (openreposhape package: ref_fault keeps the old ref rule, so it takes a.lock/b, a/.b, a//b, a/b/ and a., clones the standard in full and fails at git checkout, where the launcher, setup.sh and install.sh refuse them first (#874) #941):

   • Over the issue's 3,110 names, the head disagrees with git (--branch) on 0 and with the launcher's ref_fault, run by bash, on 0. The refusal text is identical. origin/main disagreed on 1,122.
   • Over 5,450 more names, the only disagreement with git is HEAD, which every copy takes by design.
   • Hard cases:
     • @{, a@{b, ., .., a/, a/b/, -x, refs/heads/.., a/.lock and the rest are refused by all three.
     • refs/heads/x and a 255-character name are taken by all; 256 is refused for length.
     • @, a@b and non-ASCII letters are taken by git and refused by both copies, in the same words.

3. Paths:

   • --version, --doctor, a scaffold run and shape-doctor refuse before any git call. Measured: 0 calls, and no E2 80 AE bytes.
   • --preflight prints neither value.
   • Beside a git checkout, neither value is printed.

4. Tests, one process at a time under a stub gh:

   File	Result
   package	173 passed
   command	447 passed
   test_repo_hygiene.py	81 passed
   test_agents_md_codes.py	47 passed
   test_no_auto_maintenance.py	60 passed

   The new tests read MISDRAWN_CHARACTERS out of openRepoShape and import the launcher's tables and words, so parity is enforced and not copied.

5. Mutants (failed tests):

   Mutant	Failed
   m01	11
   m08	21
   m11	34
   m14	8
   U+202C dropped from the table	2
   U+202C skipped by the function alone	2
   no .. check (a..b only)	1

   One survives: a part that is exactly . is taken (NIT-3).

6. Merges: clean with origin/main (tree f4a92826, the same as the writer's) and with setup-project.py and validate-repository-naming.py: step (4) reads and names what step (5) does, --link-remedy reads a link holding =, strips its words and notes an unused one, and the Windows hand clone passes -c core.autocrlf=false (#907, #926) #949 (tree 9399716b). No other open PR touches these files.

7. Commit and body: in the house form, "Closes openreposhape package: openRepoShape --version prints a repository holding U+202E or U+200B as typed and exits 0, and a ref holding either is refused with the value as typed, where the launcher refuses both before it prints (#866) #939." and "Closes openreposhape package: ref_fault keeps the old ref rule, so it takes a.lock/b, a/.b, a//b, a/b/ and a., clones the standard in full and fails at git checkout, where the launcher, setup.sh and install.sh refuse them first (#874) #941.", with the right author and opus trailer.

SHOULD-1. The body says the runs left alone never print a value or hand one to git. That is false beside a source tree with no .git, such as a ZIP, or an sdist installed with pip install -e ..

• beside_checkout checks for files only, so that tree's setup-project.py self-bootstraps.
• Measured: stdout holds git clone --quiet -- acme/r[U+202E]forged ..., and git is called with it. setup-project.py:881-882's checked_value has no drawn-as-another check.
• It is the same at origin/main, so this PR does not regress it.
• Fix: narrow the claim to "beside a git checkout". Lane 3 files the follow-up for setup-project.py's self-bootstrap check.

NITs:

• NIT-1: with both values bad, --doctor, a scaffold run and the tools name the ref first. The launcher and the package's own --version name the repository first, and openreposhape package: openRepoShape --version prints a repository holding U+202E or U+200B as typed and exits 0, and a ref holding either is refused with the value as typed, where the launcher refuses both before it prints (#866) #939's Direction says "the repository first". Fix: in prepare, move repo = repository(); url = clone_url(repo) above wanted = target(stamp, typed_ref), and add one --doctor case. Measured: the swap keeps 173 passed.
• NIT-2: a scaffold run with a clean --shape-ref never asks $OPENREPOSHAPE_REF anything. Measured: it is never printed or sent, but the launcher refuses it. Name it in the "Decided" paragraph.
• NIT-3: add ("a/./b", False) to REF_PARTS. Git, the launcher and the head all refuse it, and the mutant that takes a lone . part survives today.
• NIT-4 (Windows, reasoned): test_version_prints_a_repository_holding_none_of_the_list needs UTF-8 stdout in the child. Skip it on Windows, as the launcher's twin is skipped.
• NIT-5: if openRepoShape, setup.sh and install.sh: under the C locale a $OPENREPOSHAPE_REPO URL or path, or a ref, with a letter past ASCII is refused as "a control character", so --version, --install and --preflight refuse what v1.1.0 ran #940 adds rows to NOT_CONTROLS, or another table imported here, saying a lone 9B or E9 passes, the package prints those bytes as ? (measured), and this module's test would fail. Whoever lands second reconciles. openRepoShape: --doctor's shape_checkout runs git clone before checked_ref, so a ref it refuses (a.lock/b, a//b, one over 255 characters) costs a whole clone first, and where the clone fails the ref's refusal is never reached #942 touches none of the imported names.

Next: lane 3 runs a short fix round for SHOULD-1's text and NITs 1-4, and RV947 re-checks the delta. Then #947 lands under Brett Heap's v1.2.0 merge word, once tests is green.

Lane: openreposhape-3 (openRepoShape-3)

[sonarqubecloud]

Quality Gate passed

Issues
0 New issues
0 Accepted issues

Measures
0 Security Hotspots
0.0% Coverage on New Code
0.0% Duplication on New Code

See analysis details on SonarQube Cloud

[brettheap]

(lane openRepoShape-3, 2026-10-18T17:24:15Z) RV947 round 2, on the delta 21558e44..c1de4c86 (fix round 1): READY. No new MUST or SHOULD, and one new cosmetic NIT.

1. The delta, read in full: three files (standard.py, tests/conftest.py, tests/test_openreposhape_package.py) and nothing else.

   • prepare now asks repository() and clone_url() before target().
   • "first fourteen" is right: the old line takes the first 14 rows and refuses topic.lock.
   • The NO_NON_ASCII_INTO_GIT_BASH mark matches the launcher's twin.
   • The Sonnet trailer is correct.

2. The order: with the repository holding U+202E and the ref holding ESC, the package's --doctor ., Atlas --org acme and shape-doctor --json now give the launcher's refusal byte for byte (REFUSED: OPENREPOSHAPE_REPO holds ... (U+202E) ...). That holds under C and C.UTF-8, with exit 2, 0 calls, and no bytes of either value.

3. a/./b: all four REF_PARTS readers (setup.sh's, install.sh's, the launcher's and the package's) run the row and refuse it, as git does. No reader has a skip.

4. Tests, one process at a time under a stub gh:

   File	Result
   package	177 passed
   command	448 passed
   test_setup_sh.py	126 passed
   test_install_sh.py	271 passed
   test_agents_md_codes.py	47 passed
   test_repo_hygiene.py	81 passed

5. Mutants:

   • The old order fails 3: test_with_both_values_bad_a_run_names_the_repository_first[doctor], [scaffold] and [shape-doctor].
   • own2 fails 1: test_ref_fault_takes_a_ref_as_git_does_and_refuses_the_rest[a/./b]. It survived in round 1.

6. The PR body: the SHOULD-1 and NIT-2 text is true; setup-project.py: under self-bootstrap, an $OPENREPOSHAPE_REPO holding U+202E or another character the launcher refuses is printed and handed to git clone, because its check has no drawn-as-another test #953 exists and matches; the last two lines are the Lane line and the robot line.

7. Merges: clean with origin/main 64464d31 (tree 0c09d951) and with setup-project.py and validate-repository-naming.py: step (4) reads and names what step (5) does, --link-remedy reads a link holding =, strips its words and notes an unused one, and the Windows hand clone passes -c core.autocrlf=false (#907, #926) #949 8c5d6051 (tree 1951cc0b), the same trees the body names.

NIT-6 (new, cosmetic): one line of prepare's docstring (standard.py:698) is 104 characters, where the rest of the paragraph wraps at about 76. No test enforces line length. Lane 3 leaves it for the next PR that touches standard.py, rather than restart tests here.

#947 lands under Brett Heap's v1.2.0 merge word once tests is green on c1de4c86, in its turn on #473.

Lane: openreposhape-3 (openRepoShape-3)

[sourcery-ai]

Sorry @brettheap, you've used your own review budget of 250,000 diff characters for the last 7 days.

You can request another review in 3 days by commenting @sourcery-ai review. Upgrade to get a review now.