Skip to content

Adopt feature-branch workflow with reviewed PRs - #1

Merged
EvanWinget merged 1 commit into
mainfrom
claude-md-branch-workflow
Jul 27, 2026
Merged

EvanWinget merged 1 commit into
mainfrom
claude-md-branch-workflow

Conversation

@EvanWinget

Copy link
Copy Markdown
Owner

What changed

CLAUDE.md gains a Branch workflow section: all development on feature
branches, PRs reviewed by Evan before merging, CI green required, PR
descriptions must include a review guide, and spec/implementation/vector
commits stay separate within a PR.

No workflow file change was needed: .github/workflows/ci.yml already
triggers on pull_request, so this PR itself demonstrates that CI runs.

Review guide

  • One commit, one file, documentation only.
  • Check the section says what you intended: branch discipline, review
    gate, PR size expectations, the review-guide requirement.
  • Verify CI ran on this PR (the checks section below this description).

Suggested follow-up (GitHub settings, only you can do it)

Settings > Branches > add a protection rule for main: require a pull
request before merging and require the CI status checks to pass. That
makes the discipline mechanical rather than remembered.

All development moves to feature branches landing on main through
reviewed pull requests, matching the bll-consensus discipline. PR
descriptions carry a review guide so review doubles as Evan building
expertise in the codebase. CI already runs on pull requests, no
workflow change needed.
@EvanWinget
EvanWinget merged commit 9a802fc into main Jul 27, 2026
2 checks passed
@EvanWinget
EvanWinget deleted the claude-md-branch-workflow branch July 27, 2026 15:00
EvanWinget added a commit that referenced this pull request Jul 30, 2026
vectors/conditions/encoding.json pins CONDITIONS.md section 1 and
MATCHING.md rule 6: 35 cases covering every rejection rule
(structure, opcode tiers, CREATE_COIN arity and argument
strictness, reserved declared-cost rules) plus the acceptance
boundaries (zero amount, MAX_MONEY, the 10,000-byte script, the
cost floor, shapeless reserved arguments).

vectors/matching/create-coin.json pins MATCHING.md rule 1 with the
duplicate-CREATE_COIN theft case as vector #1, the honest batching
counterpart, batching-wallet shapes at three inputs, mixed
transactions with non-BitLisp inputs and unmatched outputs, the C3
divergence cases (duplicate claims from one input), metamorphic
neighbors (amount off by one, script byte flip), and the
oversized-unclaimed-slot case pinning that output slots carry no
size bound.

vectors/README.md documents both case shapes.
EvanWinget added a commit that referenced this pull request Aug 9, 2026
Section 10.3's verdict still carried the post-steelman 70% figure
without the 2026-08-07 raise to 80% recorded in section 8. The
verdict now states the full trail: 72% at original scoring, 70%
after the steelman exercise, 80% after the section 11 evidence.
No confidence judgment changes, the stale sentence catches up to
the ratified one.

The D8 decision text in the VM record said the caps question closes
with the Phase 3 budget mapping while the D8 table row and decision
5 both say Phase 4. The weight mapping is Phase 4 work per the
execution plan and COSTS.md section 9, so the decision text now
says Phase 4.

The Phase 2 checkboxes lagged the landed state. Checked: the
CONDITIONS.md v0 vocabulary (complete through the seal family,
decision 24), the invariant suite (value conservation landed as a
tx-model construction precondition, noted inline), and the
adversarial corpus (theft case is vector #1 of create-output.json,
241 validation cases, verified against the tree). Rules 1 and 4
gained landed notes matching the style of rules 2 and 3 (decisions
4 and 19). Deliberately left unchecked: VALIDATION.md (rule 5
costing pends) and the Chia cross-check translation (no translated
vectors exist in the tree, the box is honest).
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant