adopt-project.py plan never compares --project (or --org) with the repository it adopts in place, so a source named billing-service is refused with a remediation that steers to --project BillingService, which plans and checks plan ok against acme/billing-service, and execute then acts on acme/BillingService: a raw git ls-remote failure when no such repository exists, and, when a different repository of that name does, both legs, the split branch, the pull request and the topic on that repository #499

Lane: openreposhape-4 (openRepoShape-4)

(lane openRepoShape-4, 2026-10-12T23:17Z) Found by lane 4's v1.0.0 README/AGENTS.md audit (finding auA-02), re-verified on main 437a360. Lane 4 claims it.

What is wrong. AGENTS.md's adoption steps say plan --source <org/repo|path> --project <Project> (AGENTS.md:118-130), and README says the source "KEEPS ITS NAME" (README.md:729-744). Neither says what <Project> must be. plan --help does: "for an in-place adoption it is the source repository's own name" (adopt-project.py:5991-5992). Nothing enforces it.

A source with an ordinary hyphenated name, billing-service, is refused naming-unclassified with "Remediation: --project takes one CamelCase token …". A person or an assistant does what that says and passes --project BillingService, and plan and check both pass. The report reads WHAT WILL HAPPEN, against acme/billing-service on one line and ONE split commit on branch adopt/three-repo-shape of BillingService on the next. execute against GitHub then acts on https://github.com/acme/BillingService.git, not on the source.

The same holds for any mismatch. A typo (--project Atlsa for acme/Atlas) and an --org that is not the source's (--org acme-labs for acme/Atlas) both plan and check plan ok.

Reproduction. Offline, on Linux. $U is a checkout of main 437a360; git-filter-repo is on PATH; a gh that exits 97 stands first on PATH; GIT_ALLOW_PROTOCOL=file is set; identities come from env. GitHub is emulated, the route lane 3 used for #479: a probe-only global git config holds url."$R/github/".insteadOf = https://github.com/, so https://github.com/acme/X.git resolves to the local bare repository $R/github/acme/X.git.

1. A source $R/billing-service with specs/a.md, src/a.py and README.md, one commit, and origin set to https://github.com/acme/billing-service.git. It is published as $R/github/acme/billing-service.git.
2. adopt-project.py plan --source $R/billing-service --project billing-service --elected-by 'Probe Human':

   REFUSED naming-unclassified: 'billing-service' matches no family in the naming policy
   Remediation: --project takes one CamelCase token with no hyphen, underscore, dot or space.
   

   Exit 2.
3. The same with --project BillingService: exit 0. The plan records source: repository: acme/billing-service, project: BillingService and legs: assembly: BillingService.
4. check --plan plan.yaml, exit 0:

   WHAT WILL HAPPEN, against acme/billing-service @ e1972f902b89:
     create BillingService-spec and BillingService-code (private)
     ONE split commit on branch adopt/three-repo-shape of BillingService: `git rm -r` those paths, mount the legs at spec/ and code/
   plan ok
   

5. execute --plan plan.yaml --yes in GitHub mode, with no acme/BillingService:

   REFUSED a git or gh command failed. THE EXACT COMMAND WAS:
       git ls-remote https://github.com/acme/BillingService.git refs/heads/adopt/three-repo-shape
       exit 128
   --- output ---
   fatal: '<R>/github/acme/BillingService.git' does not appear to be a git repository
   

   Exit 2. This comes before the confirmation and before any leg is created.
6. Create a DIFFERENT repository, acme/BillingService (one unrelated commit), and run step 5 again. It gets past the split-branch check and the yes, as far as:

   creating the leg repositories
   REFUSED a git or gh command failed. THE EXACT COMMAND WAS:
       gh repo create acme/BillingService-spec --private --description BillingService — spec leg, extracted from acme/BillingService with history
       exit 97
   

   Only the stub stopped it. With a real gh, execute goes on to create both legs, push the split commit to acme/BillingService's adopt/three-repo-shape, open the pull request there and set the topic on it (see Why).
7. A CamelCase source acme/Atlas:
   • planned with --project Atlsa: plan and check both exit 0, and the report reads WHAT WILL HAPPEN, against acme/Atlas … then ONE split commit … of Atlsa;
   • planned with --project Atlas --org acme-labs: the plan records repository: acme/Atlas and org: acme-labs, and check says plan ok.

Why. At 437a360:

• The source's own name is read but never compared. Source.open reads the source's org/repo from its origin (adopt-project.py:743, _remote_repository :880-899), or takes it from --source org/repo (:769). plan uses only the org half, as the default for --org (:1788). The name half is written to the plan (:1434) and printed (:2061, :3513-3514), and never compared with --project.
• --project alone names all three repositories (_names, :5872-5874). In GitHub mode _repository_urls (:4483-4521) builds every repository and URL execute acts on from --org and --project (https://github.com/{org}/{name}.git, :4512). That covers:
  • the split-branch probe (:4049, reached at :5019, before _confirm at :5024);
  • the legs (_create_leg_remotes, :5040);
  • the split push (:4977-4979);
  • gh pr create --repo (:4744);
  • the topic (:4822).
• Local tests cannot see it. Under --local-remote-dir, the assembly URL IS the source path (:4520), so no local test or probe ever meets the mismatch.
• The remediation steers to the wrong value. _check_names (:5883-5886) tells the person to retype the name. It does not say that the assembly root of an in-place adoption cannot be renamed by a flag.

Proposed fix (for review on the PR).

• plan refuses when the source names a repository (source.repository is not None) and <org>/<project> is not that repository. The refusal has a code of its own, names both, and comes before anything is written. check and execute ask the same of the plan, so a hand-edited project: or org: is caught too.
• Options for Brett:
  1. Compare case-insensitively or exactly. GitHub resolves acme/atlas and acme/Atlas to one repository, and the naming policy accepts a lowercase name.
  2. What the refusal tells a person whose repository's own name fails the naming policy (billing-service):
     • (a) that it cannot be adopted in place under that name; renaming it is their decision, made first on GitHub (gh repo rename), and then they plan again; or
     • (b) only that it cannot be adopted in place, and that a project can stay single (single-repository.yaml).
       Either way, the naming-unclassified hint to retype --project is not offered for an in-place adoption.
  3. A local source with no origin has no name to compare. Either keep today's behaviour (--org required, --project not compared), or compare --project with the directory's name.
• AGENTS.md step 1's clause, that --project is the source's own name, is README.md and AGENTS.md: an accuracy pass for v1.0.0, every command, flag, exit code, refusal code and step they state checked against the tools on main, so the tag's "README and AGENTS.md accurate" criterion holds #483's text once this lands.

How to test. In the adoption plan tests:

• A source whose origin is acme/billing-service, planned with --project BillingService, is refused, and no plan file is written.
• acme/Atlas planned with --project Atlsa is refused, and so is acme/Atlas with --org acme-labs.
• acme/Atlas with --project Atlas passes. acme/atlas with --project Atlas behaves as option 1 says.
• check refuses a plan whose project: was edited to another name.
• Run tests/test_shape_doctor.py and tests/test_scaffold_leg_path_collisions.py too, because the doctor reuses adopt's plan reader.

Why v1.0.0. Hyphenated repository names are ordinary, and the procedure cannot succeed for them. The tool's own remediation then steers to a value that makes execute act on a different repository from the one it says it is adopting. Proposed P1.

Parent: #457 (adopt).

Related: #427 (the org read from origin), #479 (the base branch).