Planning and tracking for agent-assisted maintenance work across the PrismarineJS ecosystem (mineflayer, prismarine-* libraries, node-minecraft-protocol, and friends).
This repo holds the thinking behind maintenance work — analysis, plans, and implementation records — as plain Markdown so it can be reviewed, versioned, and picked up by humans or agents.
PRINCIPLES.md records the maintainer's six principles. Apply them to existing systems as well as proposed changes. Start with system review for a capability-wide assessment, or PR review for a bounded change. Assessments distinguish current evidence, proposed targets and accepted decisions. The first application is world block state and reusable queries.
Work is organized into phases. Each phase has a planning side and an execution side:
design/
phase1/ ← analysis & plans: what's the current state, what should we do
phase2/
...
implementation/
phase1/ ← execution records: what we actually did, decisions, outcomes
phase2/
...
PRINCIPLES.md ← stable architectural direction
skills/ ← reusable PR and system review instructions
policies/ ← shared repository settings
scripts/ ← read-only policy audits and CI gate generation
tests/ ← offline checks for maintenance tooling
design/phaseN/— Investigation and planning documents. Status reports, triage, proposals, and the rationale for what to work on. Written before the work.implementation/phaseN/— Records of the work itself: what changed, links to the PRs/commits that resulted, decisions made along the way, and follow-ups. Written during/after the work.skills/— Reusable review skills. Start with prismarine-review, then select the relevant protocol/data, lifecycle/actions, items/inventory, geometry/movement, world/rendering, or architecture/public API skill. Code quality and regression checks are included in the domain where they matter. Use PR prioritization to select the most important work or the next ready merge batch. These files can be read directly; this repository does not install them globally.
A design/phaseN/ document typically motivates one or more
implementation/phaseN/ documents in the same phase.
| Phase | Focus | Design | Implementation |
|---|---|---|---|
| 1 | mineflayer PR backlog triage | design/phase1 | implementation/phase1 |
| 2 | U9G PR classification and inline reviews | design/phase2 | implementation/phase2 |
| 3 | Historical review survey, skills, evaluation, and U9G re-review | design/phase3 | implementation/phase3 |
| 4 | Default-branch policy and repository settings | Policy design | Rollout and remaining prerequisites |
| 5 | Principles and whole-system architectural review | Adoption design | Implementation and pilot |
The current domain-specific skills are described in the revision design and implementation report. The original survey/evaluation report records the previous five-skill set.
The September 21 standards update strengthens maintainability review, upstream data consistency and packet-test selection using extremeheat's review decisions.
The policy separates required CI from the review exception: repository admins may bypass approval through a PR only, while CI, current-base validation, force-push and deletion protections have no bypass. See the design and rollout report for actual coverage and outstanding CI prerequisites. Repositories with only the PR rule enabled are reported as partial; their new CI rule remains disabled.
The audit uses Python 3's standard library and an authenticated gh. It only
issues GET requests and has no apply mode. It reports drift, not proof that a
repository's tests are ready for enforcement. A read error gives a nonzero exit;
setting differences are recorded in the report. Reports redact private repos.
python3 -m unittest discover -s tests -v
python3 scripts/audit_branch_policy.py --live --output /tmp/pjs-policy
python3 scripts/audit_branch_policy.py --snapshot implementation/phase4/default_branch_policy.json --output /tmp/pjs-policy-replayEach audit writes a Markdown diff and replayable JSON evidence at the output
prefix. No dependencies or global skills need installing in this repository.
scripts/required_ci.py renders the tested final gate used by rollout PRs;
it does not modify repositories or contact GitHub.
- One topic per Markdown file; use descriptive
snake_casenames (e.g.mineflayer_pr_status.md). - Lead each document with a short header noting its date and source of data, since status snapshots go stale.
- Keep documents self-contained and link out to the relevant GitHub PRs/issues.