[dependabot]

Bumps the shared group with 2 updates in the / directory: @wirecat/cli-base and @wirecat/cli-core.

Updates @wirecat/cli-base from 0.26.0 to 0.43.0

Release notes

Sourced from @​wirecat/cli-base's releases.

0.35.0

Added

• Bounded batches share parsing, correlation and sequential execution. prepareBatch, readBatchInput and runBatch accept explicit registered handlers, cap input/count/read reservations and retained output, and keep a result for every input on budget refusal. Failed reads never refund their reserved ceiling; it is not a byte consumption measurement. Hosts retain account authorization, cancellation, final rendering and resource cleanup. projectBatchAnswers selects nested answer fields while retaining ids, errors and completeness.

Full changelog

0.31.1

Fixed

• Nested field selection traverses arrays and retains report coverage. Item/list envelopes, continuation, missing data and operation identifiers are handled by one shared projector; consumers no longer need to unwrap item answers.

Full changelog

Changelog

Sourced from @​wirecat/cli-base's changelog.

0.43.0 — 17.10.2026

Added

• Released changelog sections may keep their old headings: legacyHeadingsThrough: "<version>" in ChangelogRules (@wirecat/cli-base/release) skips the heading-name and repeat checks for every section at or below that version, a pre-release of it included. A CLI that moved to the shared heading set sets it to its last version released with the old names, instead of renaming lines before the check. A value that is not a version is reported as a problem.
• One test setup file for every CLI: setupFiles: ["@wirecat/cli-base/testing/setup"] in vitest.config.ts gives each test file a fresh folder for TMPDIR, HOME, the XDG folders and every <APP>_CONFIG_DIR, _STATE_DIR and _CACHE_DIR already set, removed when the file is done; replaces @napi-rs/keyring with one that throws; and refuses every connection to anything but this machine, naming the host, from net, tls, http(s), fetch and WebSocket. A test about the network itself calls onTestFinished(allowNetwork()). It needs vitest, now an optional peer dependency. A CLI's own setup file, listed after it, keeps only what is the CLI's own.

Changed — may break callers

• cli-dev audit compares wa with the other CLIs on the shared roots of the all parity group (commands, doctor, skill, mcp, upgrade, search, notes, meetings). An audit run that passes --cli for wa may find differences it did not report before; the weekly scorecard adds wa once wa is published.

0.42.0 — 17.10.2026

Added

• mcp: false works on a single option: annotateOption(option, { mcp: false }) from @wirecat/cli-base/commands marks it, and commands --json carries mcp: false on that option and nothing on the rest, so an MCP tool can leave it out of its input while the command stays offered. The first use is messages delete --for-everyone: agents never delete for everyone. No MCP server reads it yet.

0.41.0 — 17.10.2026

Added

• The WireCat CLI standards handbook ships in the package, at docs/standards/README.md: an index, 16 area pages with 254 rules, from the toolchain to releases, and a checklist for a new CLI. Each rule has an id, a reason, the check that enforces it or "review only", an example, its exceptions and the CLIs it applies to. A rule lives on one page; a rule that moved or merged keeps its old id as a pointer, so a citation still resolves. pnpm docs:check checks the ids, the rule counts in the index, and that every "Enforced by" entry names a real check: one in this repository, or one in another WireCat repository written as <repo>: <check>.

0.40.4 — 17.10.2026

Added

• scan is a known verb: read stored messages for what a rule finds, such as tasks, and list it; record it only with --yes. For tasks scan <chat> in cli-core, the name the owner chose over tasks apply.
• mcp: false keeps a command off MCP servers: set on a command or a whole group with annotate, it covers every command below it. commands --json carries mcp: false on each command it reaches and nothing on the rest, so MCP

... (truncated)

Commits

• 42f7728 chore(release): fold Unreleased into 0.43.0 (#207)
• c063ee9 feat(docs-check): legacy changelog headings through a version; wa in parity; ...
• 8a5d00b chore(release): 0.43.0 (#205)
• 70ab9e2 feat(testing): one setup file — sandbox, throwing keyring, no network (#204)
• d54d621 chore(release): 0.42.0 (#203)
• cd9ec9f feat(commands): mcp: false on an option keeps it off MCP tools (#202)
• a970d1d docs(standards): cli-core's import rules apply again (#201)
• 490692a chore(release): 0.41.0 (#199)
• 2f18bdf docs(standards): one home per rule, one format, cross-repo checks (#197)
• 40af6f0 docs(standards): MCP, accounts, permissions and security rules (#196)
• Additional commits viewable in compare view

Updates @wirecat/cli-core from 0.262.0 to 0.325.0

Release notes

Sourced from @​wirecat/cli-core's releases.

0.293.0

Added

• store.mail.emails and emailRefs take attachmentErrors: only emails with an attachment whose text extraction failed with one of those errors, so a retry reads those emails again instead of a whole mailbox.
• Meeting reads can run as an explicit bounded batch. The shared CLI builder registers word search and passage handlers, keeps independent answers/cursors/errors, reserves stored-text read ceilings and performs one bounded index pass. The host selects one account and owns input, cancellation and final output; no model or remote write is dispatched. The existing single-input commands keep their grammar and behavior.

Changed — may break callers

• Requires @wirecat/cli-base >=0.35.0 <0.36.0. Update the consumer's direct Base pin to 0.35.0 alongside Core; the shared batch APIs use exports introduced in that Base release.

Full changelog

0.283.0

Added

• Meeting search hits include nested source context. Shared CLI/MCP results carry the meeting title, start date and evidence reference, plus a stored source-document URL when available. Context reads share the text byte budget; absent URLs stay null.

Changed — may break callers

• Requires cli-base 0.31.1 within the 0.31 series. Nested JSON field projection and skill command effects come from the shared implementation. Upgrade the CLI’s Base pin together with Core.

Fixed

• Nested --json paths accept the items.ref list shorthand. Array order and report coverage, pagination, missing data and execution metadata are retained across CLIs.

Full changelog

Changelog

Sourced from @​wirecat/cli-core's changelog.

0.325.0 — 17.10.2026

Added

• Agent tools come from the commands themselves, from @wirecat/cli-core/mcp:
  • toolsFromCommands(commands, { root, exclude, levels, keyOf, omit, offline }) turns the command tree that commands --json describes into one tool per leaf command under root. It leaves out a command marked mcp: false (on itself or a group above it), a command with a lifetime (an agent call must return), a command the profile's levels refuse, and an option marked mcp: false. Arguments are the command's arguments and options in snake_case: a switch is a boolean, choices an enum, a variadic argument and a list flag an array, a count flag (--limit) a whole number from 1, and anything else a non-empty string that is not - alone.
  • runCommandForAgent(program, path, input, ctx) runs one command through the CLI's own run, in this process, and answers { result } with the envelope the command rendered, or { error } with the error object the CLI would print — no stdout or stderr text is parsed. A write at level ask goes ahead: nobody is at a terminal to answer. ctx.levels refuses what the profile denies before the run, and ctx.overrides (key=level, stricter only) reach the command's own guard too. With ctx.connection, the command borrows the server's open connection through the relay seam serve uses: it never opens one of its own and cannot close the server's.
• ownerOnlyOption(flags, description) from @wirecat/cli-core/cli: an option MCP tools leave out of their input. BaseEnvironment.agent (type AgentRun) is how the runner receives a command's answer and error.

Changed — may break callers

• The bot MCP server's tools are generated from the bot commands and run through them, in place of the hand-written table. Same three tools and the same command names; what an agent sees changes:
  • arguments carry the command's own names: messages send and messages edit take md / html in place of format, messages delete takes messages (a list) in place of message, search messages takes query (a list of words) in place of text, chats members remove takes person in place of user;
  • new arguments: offline on every command that reads the messenger, contacts show --sync-first, search messages --page --newest, messages pin --notify, chats moderate --no-block; new command: chats rules show;
  • bounds the hand-written schemas had and the commands check themselves now, at run time: limit up to 100, the message-id and callback-id patterns, the ISO form of since_time, at least two people, and text on messages send. createBotServer takes program: { run, commands } (the CLI's run and describeProgram of its tree) and an optional levels read before every call and keyOf for a CLI's own command guarded by another key (also BotMcp.keyOf), in place of commandAt, run and tools; offered lists command paths, and tools the catalogue. The CLI's run must pass on the whole environment it is given (max's botMcpRun keeps only streams, tty and answer; the runner then refuses with "the CLI's run must pass its whole environment on"). BotMcp.tools is gone: a CLI's own bot commands are offered by being commands, and it marks with mcp: false what agents must not get.
• Commands and options MCP servers no longer offer (mcp: false; commands --json shows it): store, server, schedule, config, mcp, models (and images models), skill, flood, runs, replies, doctor report, upgrade, complete, commands schema, chats delete, chats clear, account sessions end; on the bot side bot auth, bot list, bot webhooks, bot api, bot store, bot mcp, bot commands set|clear, bot chats leave, bot chats admins add|remove; every command marked grantsAuthority (recipient lists, rules, consents, schedules — what changes a later run's authority); and the options messages delete --for-everyone, every --allow-dangerous, bot chats moderate --max-actions, and bot messages send --file --photo --voice --as-file --allow-any-file. The personal server still serves its hand-written tools; its generated catalogue is pinned in src/mcp/__snapshots__/personal-catalogue.snap.
• An agent never deletes for everyone: messages delete run for an agent refuses --for-everyone,

... (truncated)

Commits

• 022f69f chore: release 0.325.0 (#1472)
• 32322bc fix(parity): wa's switch absences are reasons now that wa's main sets them (#...
• ff4e34f feat(tasks): in a watched group any member's reply closes a question (#1469)
• 0149f9d docs(store): reserve annotation transfer migration (#1468)
• 92521dc fix(lint)!: import boundary rules apply again (#1467)
• bab6ab9 chore: release 0.324.0 (#1466)
• b8fb75f fix(parity): wa's new absences are planned until wa sets the switches
• 21a38e3 fix(meetings): preserve communications coverage in compact output (#1464)
• d17a123 feat(store): open compatible stores without writes or maintenance (#1463)
• 981afab test(steps): a step that follows an option a CLI lacks need not name a comman...
• Additional commits viewable in compare view