Skip to main content

1PRIORITY — what to work on next

Triage view across the open work in plans/backlog/. Tiered by what to do next, what is deferred pending a prerequisite, and what is parked. This is a triage tool, not a roadmap — see PLANS.md → "Keeping backlog/1PRIORITY.md current".

Last Updated: 2026-09-07 Snapshot: v1.7.6 · default branch main · plans: 0 active / 5 backlog / 29 completed · open PRs: 0 · open issues: 0

How to use this doc​

Read Tier 1. If Tier 1 is empty, read Tier 2. Each row states the evidence for its rank and the one thing that would unblock or close it. A row moves down only when its blocker is real and named, never because it is uncomfortable.

Nothing is invented here. Every row below is something already in the tree that is either contradictory, stale, or scoped-and-unstarted.

Tier 1 — decide this first (blocks other work)​

The PowerShell surface contradicts itself, and it is shipped​

The repo says three different things about PowerShell at the same time:

SourceClaim
README.md line 103 (v1.7.6)"PowerShell siblings are no longer shipped (Bash is the supported surface)"
CLAUDE.md line 28"Multi-OS by default. Every script ships .sh … and .ps1 … siblings."
The tree12 bin/*.ps1, install.ps1, and 8 lib/*.ps1 are present and tracked

They are not dead files. install.sh's sparse-checkout set is /bin/ /lib/ /templates/ /shell/ /version.txt, so every .ps1 is checked out onto every install, including installs whose README says PowerShell is gone.

Worse, the three that a Windows user would actually run are on the pre-v2 code path:

  • bin/info.ps1:25, bin/logs.ps1:28, bin/shell.ps1:28 all source lib/service.ps1 — the v1 lib, which targets the old single-repo layout.
  • Their Bash siblings moved to lib/service-v2.sh in PLAN-B / PLAN-D (both in completed/).
  • There is no lib/service-v2.ps1. The PowerShell half of the v2 rewrite was never written.

So a Windows user running noclickops info today gets v1 behaviour against a layout that v2 replaced — silently, with no version guard.

Decision needed (a human call, not a refactor): either delete the .ps1 surface and the installer's Windows path, and fix CLAUDE.md line 28 to match README.md — or keep multi-OS, write lib/service-v2.ps1, and port info/logs/shell. Both are defensible. Shipping both claims is not.

Blocks: PLAN-G, whose scope is wrong until this is settled (below).

Status: asked 2026-08-30; a fleet hold is open and escalates to a human daily until answered. Note the decision may already have been taken and never executed — README.md and PLAN-G ("PowerShell was dropped at v2") both state it was, while CLAUDE.md still mandates the opposite and the installer still ships it. If so this is a confirmation rather than a fresh call. The one fact nobody has supplied: whether any user runs native PowerShell without WSL or Git Bash. That answer decides it outright.

Tier 2 — ready once Tier 1 lands​

PLAN-G — v1 cleanup + the 2.0.0 bump​

Two things about this plan are out of date:

  1. Its prerequisite is already met. PLAN-G says it waits on v2 being "demonstrably verified working end-to-end … Currently waiting on full verification (some Reader-access constraints)". But v2-smoke-test.md → Run history records a full clean run on v1.6.5 across all four phases with the verdict PASS. The prerequisite text is stale and should be struck.
  2. The 2.0.0 promotion it assumes never happened. That same run-history entry ends "promoted to 2.0.0 in a follow-up commit". version.txt says 1.7.6. Either the promotion was reverted or the note was aspirational; the doc and the version file disagree and one of them is wrong.

And its scope is wrong. PLAN-G's goal is to "remove the orphan v1 lib files (lib/service.sh, lib/service.ps1)". lib/service.ps1 is not orphan — three bin/*.ps1 commands source it (see Tier 1). Deleting it as written breaks the PowerShell commands outright. lib/service.sh is closer to orphan: no bin/*.sh sources it, only tests/test-PLAN-008-info.sh does, which PLAN-G already plans to retire.

One thing that unblocks it: the Tier 1 decision. Then rewrite PLAN-G's Prerequisites and its lib/service.ps1 line, and ship it as the 2.0.0 PR.

Tier 3 — real maintenance, not blocked, not urgent​

14 open Dependabot PRs — done 2026-09-07​

Closed by #43, a batched sweep. The queue had grown to 16 by the time it was worked. All were transitive-only security advisories in website/; website/package.json never changed. npm audit went 45 advisories to 29, critical 1 to 0 (websocket-driver <=0.7.4). The 16 individual PRs were closed as superseded rather than left to rot, so the queue does not silently rebuild.

The website build is not covered by CI​

Found while doing the sweep above, and it is the more durable problem.

.github/workflows/tests.yml runs bash tests/run-all.sh and nothing else. The Docusaurus build runs in .github/workflows/deploy-docs.yml, which triggers on push to main — that is, after merge. Since docusaurus.config.ts sets onBrokenLinks: 'throw', a change that breaks the docs build passes its PR check green and breaks main. All 16 dependency PRs above would have merged green without anyone building them; #43 was verified locally instead, which is not a control that survives the next contributor.

One thing that closes it: a job in tests.yml that runs scripts/generate-docs.sh and npm run build for PRs touching website/. Same steps deploy-docs.yml already uses, moved before the merge instead of after.

29 npm advisories remain, and they are not individually fixable​

What is left after #43 is the @docusaurus/* cascade plus image-size and serialize-javascript, none of which has a fix at the pinned Docusaurus 3.10.1. A 3.10.2 patch is available and may clear part of the cascade — it needs to be tried and built, not assumed. The only fix npm offers for @easyops-cn/docusaurus-search-local is a semver-major downgrade to 0.29.0, which is not safe and should not be taken on audit's advice alone.

One thing that closes it: try 3.10.2 on a branch, rebuild, re-audit, and record what it did and did not fix. Deliberately not bundled into the security sweep, so a build regression would be attributable.

Tier 4 — scoped, ready, deliberately deferred​

INVESTIGATE-plans-and-scaffolding — PLAN-B / C / D​

Status line says "4 PLANs scoped, ready to ship". One of the four has shipped: PLAN-A → PLAN-106-slim-install (it names itself as that entry). Its checklist item at line 177 is still unchecked — a stale box, not stale work.

The remaining three (scaffold-ai-developer, scaffold-docs, update-ai-developer) are fully scoped and unstarted. They add three new user-facing commands, which is a minor-version surface change — landing them before 2.0.0 would widen the very surface PLAN-G exists to narrow.

Deferred until: 2.0.0 ships. Then re-rank to Tier 2.

⚠️ Naming collision to fix while you are in here. There are two unrelated PLAN-A/B/C/D letter series: this investigation's (A = slim install, B/C/D = scaffolding) and INVESTIGATE-new-target-structure's, which shipped as PLAN-A-service-discovery, PLAN-B-info-rewrite, PLAN-C-deploy-rewrite, PLAN-D-logs-shell-rewrite in completed/. "PLAN-B" is ambiguous in this repo today. Re-letter this investigation's remaining three.

Tier 5 — parked​

ItemWhy it is parked
PLAN-watch-live-deployReal user problem (first public-endpoint deploy is ~60 min behind a WAF registration). Needs no new prerequisite — it is parked on value, not blockers: nobody has asked for it since 2026-05-29. Promote on the first real request.
INVESTIGATE-uis-lessonsA survey of a sister project. It has no child PLANs and nothing depends on it. Its findings were meant to feed edits back into INVESTIGATE-noclickops; check whether they did before ever reopening it.
INVESTIGATE-noclickopsThe parent v1 design. Substantially delivered — 12 commands ship. It stays in backlog/ mainly because nobody has decided what "done" means for it. Closing it is a 2.0.0 housekeeping task, not investigation work.

What this file does not cover​

Fleet coordination. That lives in terchris/urb-agents (read remotely, path-scoped), not here. This file is the project-side triage document; fleet/noclickops/status.md in that repo is its one-screen snapshot.