Skip to content

Branch Enforcement #703

Open
0 of 4 issues completed
Open

Description

@loganfinney27

Summary

Establish enforced branch-management schema for LAF-US/IDAHO-VAULT — a verifiable check on which source branches may target which targets — replacing the current reliance on convention + agentic imitation. This is the umbrella; implementation is split into two phased sub-issues: Advisory Guard (flag, don't block) → Enforcement Guard (promote to required once proven quiet).

Why

Convention without enforcement rots. Observed live on 2026-06-29: ~153 branches, a 55-PR backlog (oldest ~27d) the merge stack isn't clearing, ~20 ad-hoc prefixes, stale registries, and drift. An enforced (source→target) rule is self-correcting — violations surface immediately — which is the cure for that decay, not another doc to keep in sync. At the largest scale this is the same principle as the vault's persistent portable sovereign identity aim: a rule that holds because it's enforced, not because everyone remembers it.

Repo context (verified, read-only)

  • Single trunk: main (protected). No develop. Active merge queue (gh-readonly-queue/main/*).
  • Real prefixes in use (not GitFlow): claude/ codex/ mistral/ manus/ copilot/ dependabot/ automation/ bot/ agent/ feat/ fix/ test/ + bare names. No feature/ release/ hotfix/.
  • Classic GitFlow does not fit (squash-only, merge queue, no develop). The transferable idea is (source → target) governance as a check, not the main+develop model.

Design constraints (load-bearing)

  • Identity-independent. Keys on the branch name, not the actor — so it works even though agents currently commit under Logan's account (the identity collapse; see Stable cross-platform signed-commit solution (replace 1P-desktop-dependent chain) #398). This is why it is the right first brick: it needs neither identity separation nor new infrastructure nor budget.
  • Queue-safe. The merge queue re-runs required checks; a naive required check that fails on unknown prefixes would jam the queue. Hence advisory-first.
  • No new parallel script. Extend the existing classifier .github/scripts/branch_garden_report.py rather than spawn another workflow. The repo already carries ~5 overlapping merge workflows; adding a 6th parallel script would repeat the exact anti-pattern this work is meant to end.
  • Lane-list = small, Logan-curated source of truth, grounded in LIVE merge history, kept separate from the stale narrative registries.

Proposed lane model (Logan to ratify)

Permitted into main: human lane (logan/ — reconcile vs loganfinney27/) + registered per-agent lanes (claude/ codex/ mistral/ …, kept, not collapsed) + system lanes {dependabot/**, copilot/**} + always-allow gh-readonly-queue/**. Everything else: flagged.

Sequencing

Enforcement was parked behind branch pruning; current view is that the Advisory Guard doubles as the pruning lens (flags non-conforming/legacy branches without blocking), so the two overlap rather than strictly sequence.

Tracked as sub-issues: Advisory Guard and Enforcement Guard (linked below).


Provenance

  • Author: Hyperagent "Developer" agent (Claude Opus 4.8) — an external agent, not Logan.
  • Authority: Logan Finney (human), who directed this filing on 2026-06-29.
  • Account of record: filed via the GitHub MCP under loganfinney27 because agent identity separation is not yet in place (see Stable cross-platform signed-commit solution (replace 1P-desktop-dependent chain) #398). Per CONSTITUTION §I, author and authority are not to be assumed as Logan by default — this footer is the disambiguation.

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Type

    No type

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions