Open work for this cycle only, most important first. Someday work is in BACKLOG_LATER.md; the research behind an item is in BACKLOG-NOTES.md. What developers can build on today and what comes next is the roadmap.
- An id is permanent and never reused. Take one with
bin/next-id CORE: one counter for every worktree on the machine, under a lock, and never below the highest number on any branch. - One item: id · priority · title · where the work starts (
path:line, a plan or a pull request) · details. One to three lines; longer analysis goes to BACKLOG-NOTES.md or a plan beside this page. - Priority: P1 broken, untrue to users, an install failure or a security problem: done before WireCat is advertised · P2 this cycle · P3 someday, in BACKLOG_LATER.md.
- Mark: none — not started · 🚧
<branch>— taken · 🟡 — half done, the rest named · ⏸️ — deferred by the owner · 🚩 — waits on an owner decision. - Claim before code: put
🚧 <branch>on the line in the first push of the branch. - Close in the pull request that ships the work: delete the line (and its note). The changelog keeps history.
None open.
-
CORE-42 · P2 · Take cli-base ≥ 0.66.1 so a bare number counts as a word of a
--discoverquestion: today any number ("after 21 days") sends the question to strict search, which usually finds nothing (cli-base#409). Moves the@wirecat/cli-basepeer range to 0.66, so the tools built on Core move with it.src/services/messages-discovery.ts. -
CORE-36 · P2 · 🚧
fix/shared-contract-auditEnforce shared command contracts by default; exact, reasoned overrides only. Export actual schemas/reasons for complete cross-CLI audits. -
CORE-40 · P2 · 🚩 Start a command without loading every command:
./clistill loads ~880 files (~400 ms) for any command but a bare--version, and no single cut frees more than 119 of them. Owner chooses: lazy action bodies in waves, or a per-tool bundle. Evidence and the two options. -
CORE-3 · P2 · 🚩 A smaller install: a fresh
npm i @wirecat/cli-coreis 238 MB. Draft plan waits for the owner: plans/2026-10-18-install-size.md. -
CORE-4 · P2 · 🟡 🚧
fix/standards-vocabulary-pathsThe audited personal-account result entities, examples and list queries are merged. Canonical command paths and removal of compatibility spellings ship in #1837 (owner, 2026-10-24); every CLI adopts them in the same-day cli-core raise. New bot query/schema gaps are separate from the original personal-account audit. Notes. -
CORE-5 · P2 · Let a new source join without editing this package: the provider list is fixed (
src/domain/providers.ts:14) and Google Drive is hard-coded in the document code (src/services/document-source.ts:8,17). Needed before the Slack, Google Meet, Notion and Calendar CLIs. No separate SDK: cli-core is the shared base (owner, 2026-10-23). Plan first. -
CORE-6 · P2 · 🟡 Search start-up measured: model open 1.44 s, empty-store reopen 8.4 ms, bare Node 31.7 ms (medians). Next compare one-shot CLI and long-lived MCP on the same synthetic corpus before optimizing. Runner and limits · notes.
-
CORE-7 · P2 · 🟡 One
--sourcefor every search: left are document-only selection, provider and kind coverage in answers, and memo's work item 5. source-everywhere-plan.md. -
CORE-8 · P2 · Typo-tolerant search by default (approved option B); waits on plain search covering transcripts, file names and file text. typo-search-default-plan.md · notes.
-
CORE-9 · P2 · Generated tags for notes, documents and messages (approved, not built:
taggingshas no status column,src/store/sqlite/schema.ts:1689). generated-tags-plan.md · notes. -
CORE-10 · P2 · 🚧
feat/reviewed-tag-applyReviewed tag consolidation (draft #1577, store version 30 reserved in COORDINATION.md). Foundation: TAG-CONSOLIDATION.md. -
CORE-11 · P2 · 🚧
feat/transcript-indexVoice-note transcripts searchable in any word form (#1761, store version 29). -
CORE-12 · P2 · 🚧
feat/media-archiveVerified media archives and daily timers (draft #1758). -
CORE-33 · P2 ·
stats messages counters refreshbecomesstats messages counters sync:refreshis the word the vocabulary rules out. tg, max and wa. After #1837.