B2a evidence — Shared procedures
Purpose
Record the local verification of TASK-002 on 2026-11-07. B2a provides deterministic lookup and preparation of eight procedures; it does not certify agent execution nor close B2 or the v1.0.0 launch.
Result
make verify finished with exit code 0: 71 passing tests in a 24.111-second suite. Environment: macOS 27.2 arm64, Python 3.15.2, Setuptools 84.2.1 previously verified and pinned by hash. No new dependencies.
| Set | Cases | Scope |
|---|---|---|
| B1 regression | 43 | Original CLI, filesystem, links, metadata and installation. |
| B2a procedures | 24 | Eight es/en recipes, inputs/targets, graph, limits, manifest, repetition and absence of writes/network/execution. |
| Additional cases | 3 | Unknown fields excluded, invalid Unicode request and shared ancestor without duplicates. |
| B2a distribution | 1 | Installs the wheel and the one rebuilt from sdist in clean environments; prepares plan in es/en with both. |
Wheel/sdist were built, and syntax, es/en parity and whitespace were checked. The frozen B1 and B2a tests remain intact. The human-readable output of workflow show discover was inspected: purpose, inputs, steps, expected outputs, gates, checks and safeguards in Spanish.
After updating the documentation, the structure and local links of 30 Markdown files were verified, and the inherited rules of AGENTS.md were confirmed to be fully preserved. The syntax/catalogs/whitespace check was repeated; implementation and resources did not change after the final build.
The updated wheel was installed in .venv with --no-index --no-deps. A fictitious notes project created in .context/ made it possible to prepare the eight operations from that executable. This is a preparation demo: the specs generated by init remain drafts and are not presented as completed artifacts of the procedure.
Test sequence
- Prior baseline: 43 passing B1 tests, recorded in
.context/b2-baseline.log. ca43f17: 25 new cases frozen before implementation. Red set against the CLI withoutworkflow; log.context/b2-red.log.d4d6f93: three additional cases before fixing the propagation of unknown target fields. The malicious case triggeredUnicodeEncodeError; the fix projects onlyid,kind,pathandparents. Log.context/b2-edges-red.log.- Final result:
.context/b2-verify.log. Installed demo:.context/b2-demo-result.json. Private logs and projects are not distributed.
The test commits carry no human sign-off or attestation. Implementation and documents remain in the working tree for review. No push, merge or publication was performed, and nobody acted as approver.
Artifacts of the final run
Local version 1.0.0.dev0; the hashes identify this build and supersede the previous local artifacts. Bit-for-bit reproduction is not promised.
| File in dist/ | Bytes | SHA-256 |
|---|---|---|
dixi_spec_kit-1.0.0.dev0-py3-none-any.whl | 28 503 | 045b73f69fa1970d18b4b375348d965bc8ee13d374c421bbaca6502692b89d7c |
dixi_spec_kit-1.0.0.dev0.tar.gz | 23 471 | 59ad2030b1d52be30aeccb2364f6a57682a47c3959b9493533ae5407f645e71e |
The explicit packaging list was extended with workflows.py, resources/workflows.json and the workflows.es.json/workflows.en.json recipes. The resources are new to the product; no private rules were copied. The package manifest keeps excluding context, history and the workspace's operational configuration.
Limits and next step
Only lookup/preparation is verified: compatible ID, declared references, hashes and resources. Semantics, consent, design quality, execution of the steps and enforcement of safeguards by an external LLM are not verified. The limits and read conditions are in the contract. There is no transactional snapshot; sources may change after being read.
No adapter evaluations, tests on other systems/Python versions, coverage, mutation, gitleaks or semgrep were run. The controls required before publishing, rights/license and release remain pending.
The next increment must connect the same recipes to optional integrations and evaluate artifacts, missing questions, conflicts and gates on the example. Those evaluations will complete the B2 evidence; TASK-001 and the guided walkthrough remain open.