murmur-it 0.2.2 → 0.2.4

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
package/CHANGELOG.md CHANGED
@@ -1,8 +1,48 @@
1
1
  # Changelog
2
2
 
3
- All notable changes to murmur are documented here. The format follows
4
- [Keep a Changelog](https://keepachangelog.com/en/1.1.0/) and the project uses
5
- [Semantic Versioning](https://semver.org/).
3
+ All notable changes to murmur are documented here. The format follows [Keep a Changelog](https://keepachangelog.com/en/1.1.0/) and the project uses [Semantic Versioning](https://semver.org/).
4
+
5
+ ## [0.2.4] - 2026-10-21
6
+
7
+ ### Changed
8
+
9
+ - Requires Node 22 or newer (`engines` was `>=20`). Node 20 reached end of life on 2026-04-30, and vitest 5 needs Node 22.12 to run the tests, so CI now runs on Node 22 and 24 instead of 20 and 22.
10
+ - Dependencies: zod 4 (runtime), and for development TypeScript 7, `@types/node` 26, vitest and `@vitest/coverage-v8` 5 and `actions/setup-node` 7.1.0.
11
+
12
+ ### Fixed
13
+
14
+ - One reply that resolves two waits on the same thread answers each wait once. Before, handing over the first wait's replies re-entered the waiter check, which answered the second wait, and the outer loop then answered it again from its stale copy (`src/broker.ts` `checkWaiters`).
15
+ - A reply held for the asker's open `murmur_wait` no longer sends the replier a `queued_delivered` note when the wait hands it over. The note reached the replier as a new turn (a live check woke the replying session a second time for it) although nothing had waited on a session that was down. A reply that was not held for the wait, for example one that stayed queued while the asker's session was closed, still sends the note when a later wait hands it over.
16
+ - The 200 inbox cap (`cap_inbox`) counts only deliveries not yet proven delivered to the recipient's session (queued or sent). It counted delivered ones too, which stay delivered until an ack, so the replies an asker received in waits filled it: after 200 consults every further reply to that asker failed with `cap_inbox`. The unread count in status and inbox is unchanged.
17
+
18
+ ### Security
19
+
20
+ - vitest 5 drops the vulnerable test-only dependencies `npm audit` reported (tinypool GHSA-5gmw-xhrv-c9v3 and GHSA-85c8-ppgw-ccpr, critical; vitest and `@vitest/mocker` GHSA-82fw-gwwq-j7x9, moderate). The published package was not affected: `npm audit --omit=dev` found nothing.
21
+
22
+ ## [0.2.3] - 2026-10-20
23
+
24
+ ### Changed
25
+
26
+ - Murmur is a group chat: a message is posted once and everyone in the thread can read it, but nobody is woken for a message that is not for them. A reply is stored once and delivered (with a push, in the parent's mode) only to whoever is waiting for it: the author of the message it answers plus the session an owner message was typed in (`via`); a reply inside the replier's own thread goes to the parent's recipients. Other participants get no delivery row and no push and read the reply with `murmur_thread` / `murmur thread`. Measured reason: `make bench-consult` (haiku, three sessions) cost 546,811 tokens per consult for murmur against 265,771 for native `SendMessage`, because every reply woke every participant and the Implementer received each reply twice (once in the wait, once as a pushed turn).
27
+ - `murmur_wait` returns the text of the replies addressed to the waiting handle (`replies`: id, seq, sender, verdict, delivered body with the truncation pointer). A reply that arrives while its recipient has a wait open on that thread is not pushed; it stays queued, is handed over in the wait result (condition met or timeout) and marked delivered. A reply that arrives with no wait open is pushed as before.
28
+ - `murmur_thread` takes an optional `messageId` (CLI `murmur thread <id> --message <m>`): the result is only that message and the replies to it, with full delivered bodies; a message not in the thread is a `bad_request`. The footer of every delivered message now points at it (`replies to this: murmur_thread <t> messageId <m>`) instead of the whole thread, and the wait handover is limited to replies to the thread's latest ask.
29
+ - The MCP instructions, the `murmur_reply` and `murmur_wait` descriptions, the opencode and Hummin adapter descriptions and the docs describe this routing.
30
+ - Claude Code never defers murmur's tools: `murmur install claude` writes `"alwaysLoad": true` into `mcpServers.murmur` (idempotent; re-run it on an existing install) and the MCP server lists every tool with `_meta: { "anthropic/alwaysLoad": true }`. Measured reason: each deferred tool cost a ToolSearch call, and a Claude Code call re-reads the whole context (about 38k tokens on haiku).
31
+ - Teaching the reply: the footer on every delivered message and the MCP instructions say plainly that the answer goes through `murmur_reply` with that messageId and that plain text never reaches the sender.
32
+ - `murmur_send` takes optional `wait` (`all_replied` or `any_replied`) and `timeoutMs` (same default and cap as `murmur_wait`): it sends and waits on that thread in the same call and returns `{ sent, wait }` with states, verdicts and the replies' text. Asking is one tool call.
33
+ - `to` also accepts a session's display name (its CLI title, case-insensitive) or a name whose slug is a handle (`Peer A` resolves to `peer-a`). Handles and groups win; an ambiguous name is a `bad_request` listing the candidates.
34
+ - The `/murmur` command file for Claude Code and zcode sets `disable-model-invocation: true`: it is the owner's command, and a model that invoked it through the Skill tool sent nothing. Re-run `murmur install claude` / `zcode` to get it.
35
+ - `make bench-consult` counts a run as completed only when the Implementer's final answer holds both peers' verdicts and their answers really reached it, and reports API calls and ToolSearch calls per session. Result (haiku, 3 runs per setup, medians): plain native `SendMessage` completes 0 of 3 (268,272 tokens, 7 API calls), because its peers answer in plain text that never reaches the Implementer; native with the peers told to answer with `SendMessage` completes 3 of 3 (510,931 tokens, 13 API calls, 13.8 s); murmur completes 3 of 3 (362,879 tokens, 9 API calls, 10.1 s), 29 percent fewer tokens than the native setup that finishes.
36
+
37
+ ### Fixed
38
+
39
+ - A reply held back for an open `murmur_wait` is now only one the wait hands over (a reply to the thread's latest ask); a reply to an older ask, or to a reply, is pushed as usual instead of staying queued with no path to its recipient. A wait remembers the replies held for it and hands them all over even if the thread's latest ask changed meanwhile. Replies held for a wait whose connection closes before it resolves are pushed at once. A delivery whose push is already in flight is never handed over a second time by a new wait or the hook pull.
40
+ - The live rig (`scripts/live/lib.mjs`) sets `MURMUR_AUTOSTART=0`, so a hook that outlives cleanup cannot respawn a broker and none survives a live run.
41
+
42
+ ### Added
43
+
44
+ - `make bench-consult` (`scripts/live/bench-consult.mjs`): three real Claude sessions per run, plain native `SendMessage`, native with the peers told to reply with `SendMessage`, and murmur, with tokens, API calls, copies received and wall time from the sessions' own transcripts; results in `docs/bench/consult.md` and `consult.json`.
45
+ - Docs and repository files: README hero, demo and linked CLIs; a Diataxis docs navigation (getting started, using murmur, consult, design, concepts, reference); `CODE_OF_CONDUCT.md`, `SUPPORT.md`, `MAINTAINERS.md`, `CODEOWNERS`, issue forms, a pull request template, Dependabot, an OpenSSF Scorecard workflow, a docs CI workflow (markdownlint, typos, lychee) and SHA-pinned GitHub Actions.
6
46
 
7
47
  ## [0.2.2] - 2026-10-20
8
48
 
@@ -81,6 +121,7 @@ All notable changes to murmur are documented here. The format follows
81
121
  - Limits: send rate, duplicate window, thread and inbox caps.
82
122
  - 283 tests at 100 percent lines and branches, and an end-to-end rig that replayed the live 3-peer scenario.
83
123
 
124
+ [0.2.3]: https://github.com/rennf93/murmur/releases/tag/v0.2.3
84
125
  [0.2.2]: https://github.com/rennf93/murmur/releases/tag/v0.2.2
85
126
  [0.2.1]: https://github.com/rennf93/murmur/releases/tag/v0.2.1
86
127
  [0.2.1-rc.1]: https://github.com/rennf93/murmur/releases/tag/v0.2.1-rc.1
package/README.md CHANGED
@@ -1,35 +1,79 @@
1
- <img src="docs/assets/murmur-512.png" alt="murmur" width="96">
1
+ <div align="center">
2
+
3
+ <img src="https://raw.githubusercontent.com/rennf93/murmur/master/docs/assets/murmur-512.png" alt="murmur icon" width="160">
2
4
 
3
5
  # murmur
4
6
 
5
- Murmur is a local message bus for AI CLI sessions that share one machine and one user account. One broker process owns a SQLite store and one Unix domain socket; every agent session joins as a named peer, and peers talk in threads. One send produces one message that fans out to any mix of handles and groups, so no session writes the same announcement twice and no session relays another's words by hand. A reply reaches everyone in the thread. Senders always see, per recipient, whether a message is still queued, was handed to the session, was proven to have reached it, or was answered.
7
+ A local message bus so AI CLI sessions talk in group threads.
6
8
 
7
- It works with Claude Code, Codex, Gemini CLI, zcode, opencode and Hummin. Each joins through the deepest hook its CLI offers (see the truth table), and any process that can run a shell command can use the `murmur` CLI.
9
+ [![npm](https://img.shields.io/npm/v/murmur-it)](https://www.npmjs.com/package/murmur-it)
10
+ [![CI](https://github.com/rennf93/murmur/actions/workflows/ci.yml/badge.svg)](https://github.com/rennf93/murmur/actions/workflows/ci.yml)
11
+ [![License](https://img.shields.io/npm/l/murmur-it)](https://github.com/rennf93/murmur/blob/master/LICENSE)
12
+ [![Docs](https://img.shields.io/badge/docs-rennf93.github.io%2Fmurmur-blue)](https://rennf93.github.io/murmur/)
13
+ [![OpenSSF Scorecard](https://api.scorecard.dev/projects/github.com/rennf93/murmur/badge)](https://scorecard.dev/viewer/?uri=github.com/rennf93/murmur)
8
14
 
9
- Full documentation: https://rennf93.github.io/murmur/ (concepts, per-CLI pages, command, MCP and protocol reference, releasing). See also [CHANGELOG.md](CHANGELOG.md), [CONTRIBUTING.md](CONTRIBUTING.md) and [SECURITY.md](SECURITY.md).
15
+ [Full documentation](https://rennf93.github.io/murmur/) | [Changelog](https://github.com/rennf93/murmur/blob/master/CHANGELOG.md) | [Contributing](https://github.com/rennf93/murmur/blob/master/CONTRIBUTING.md) | [Security](https://github.com/rennf93/murmur/blob/master/SECURITY.md)
10
16
 
11
- ## Install
17
+ </div>
18
+
19
+ Murmur is a local message bus for AI CLI sessions that share one machine and one user account. One broker process owns a SQLite store and one Unix domain socket; every agent session joins as a named peer, and peers talk in threads. One send produces one message that fans out to any mix of handles and groups, so no session writes the same announcement twice and no session relays another's words by hand. A reply goes to whoever is waiting for it, and everyone in the thread can read it. Senders always see, per recipient, whether a message is still queued, was handed to the session, was proven to have reached it, or was answered.
20
+
21
+ It works with Claude Code, Codex, Gemini CLI, zcode, opencode and Hummin. Each joins through the deepest hook its CLI offers (see the [truth table](#delivery-truth-table)), and any process that can run a shell command can use the `murmur` CLI.
22
+
23
+ ![Terminal recording: a group, one send, two replies from simulated peers, the thread and the status](https://raw.githubusercontent.com/rennf93/murmur/master/docs/assets/demo.gif)
24
+
25
+ This is a CLI demo with simulated peers: `alice` and `bob` are registered by a hidden setup line and reply with `--as`, against an isolated broker. No agent CLI is involved. The tape is [`docs/assets/demo.tape`](https://github.com/rennf93/murmur/blob/master/docs/assets/demo.tape).
26
+
27
+ ## Quickstart
12
28
 
13
- Requires Node 20 or newer.
29
+ Requires Node 22 or newer.
14
30
 
15
31
  ```
16
- npm install -g murmur-it # the package is murmur-it; the command it installs is murmur
17
- murmur init # ~/.murmur (0700), token (0600), murmur.db, owner peer
18
- murmur install claude # and/or: codex | gemini | zcode | opencode | hummin
32
+ npm install -g murmur-it # the package is murmur-it; the command it installs is murmur
33
+ murmur init # ~/.murmur (0700), token (0600), murmur.db, owner peer
34
+ murmur install claude # or: codex | gemini | zcode | opencode | hummin
19
35
  ```
20
36
 
21
- From source instead: `npm install && npm run build && npm link` in a clone (puts the same `murmur` on PATH; every hook, MCP entry and command runs plain `murmur`).
37
+ Start two sessions, then from inside one of them send to both with a single message (handles are `<cli>-<folder>`; `murmur status` lists them):
38
+
39
+ ```
40
+ /murmur claude-api,codex-web Which of you owns the migration?
41
+ ```
42
+
43
+ Each session receives the message, and each reply goes to whoever is waiting for it. Everyone in the thread can read it with `murmur thread <id>`. [Getting started](https://rennf93.github.io/murmur/getting-started/) has the full walk-through.
44
+
45
+ ## Why
46
+
47
+ The bus removes two costs and must never remove a third thing. The costs: duplicated composition, where every member of a group reformats the same fact, and sender-side relaying, where one session carries another's words to a third and garbles them. One message fans out; replies go to whoever is waiting and stay readable by the whole thread; the broker owns storage and delivery. Delivery modes are the token dial.
48
+
49
+ The third thing is independent verification: a message is a claim, not evidence. Peers verify against the artifacts themselves, which is why messages carry `references`, and why every session is told so by the MCP server's instructions.
50
+
51
+ Measured with `make bench-consult`: three real Claude Code sessions on haiku, the Implementer asked to consult Peer A and Peer B and report both answers, 3 runs per setup, medians, 2026-10-20.
52
+
53
+ | Setup | Finished | Tokens, all sessions | API calls | Wall time |
54
+ | --- | --- | --- | --- | --- |
55
+ | Native `SendMessage` | 0 of 3 (peers answer in plain text, nothing comes back) | 268,272 | 7 | 5.7 s |
56
+ | Native, peers told to reply with `SendMessage` | 3 of 3 | 510,931 | 13 | 13.8 s |
57
+ | murmur | 3 of 3 | 362,879 | 9 | 10.1 s |
58
+
59
+ Against the native setup that finishes, murmur used 29 percent fewer tokens and 4 fewer API calls. Each call costs about 38k tokens there, mostly Claude Code re-reading its own context, so the saving comes from fewer calls: one send reaches both peers, the tools are loaded up front, and the replies come back to the asker, nobody else is woken.
60
+
61
+ Method and numbers: [consult benchmark](https://rennf93.github.io/murmur/bench/consult/).
62
+
63
+ ## Install
64
+
65
+ From source instead of npm: `npm install && npm run build && npm link` in a clone (puts the same `murmur` on PATH; every hook, MCP entry and command runs plain `murmur`).
22
66
 
23
67
  The broker starts on demand: the first hook, MCP server or CLI command that finds it down starts one in the background (`murmur serve` runs it in the foreground; `murmur serve --daemon` detaches). `MURMUR_HOME` relocates the home for every component.
24
68
 
25
- What each installer writes (idempotent merges; every touched file gets a one-time `.bak-murmur` copy first):
69
+ What each installer writes (idempotent merges; every touched file gets a one-time `.bak-murmur` copy first; install refuses to touch a config file that is not valid JSON):
26
70
 
27
71
  | CLI | Files | Notes |
28
72
  | --- | --- | --- |
29
73
  | claude | `~/.claude/settings.json` hooks (SessionStart, UserPromptSubmit, Stop, SessionEnd), `mcpServers.murmur` in `~/.claude.json`, `~/.claude/commands/murmur.md` | Removes the old `CHANGE-ME-per-session` handle placeholder. |
30
74
  | codex | `~/.codex/hooks.json` (SessionStart, UserPromptSubmit, Stop, SessionEnd), `[mcp_servers.murmur]` in `~/.codex/config.toml`, `~/.codex/prompts/murmur.md` | Codex asks you to trust new hooks once: open `/hooks`. The command is `/prompts:murmur`. |
31
75
  | gemini | `~/.gemini/settings.json` hooks (SessionStart, BeforeAgent, AfterAgent, SessionEnd) and `mcpServers.murmur`, `~/.gemini/commands/murmur.toml` | |
32
- | `zcode` | Hooks fire under workspace trust; the session announces; one message is handed over by a hook and answered | BLOCKED: the provider answers 429 "Insufficient balance or no resource package. Please recharge." and the session never announced in headless mode, so whether hooks fire under `-p` is unverified |
76
+ | zcode | `~/.zcode/cli/config.json` (`hooks.enabled`, SessionStart, UserPromptSubmit, Stop; `mcp.servers.murmur`), `~/.zcode/commands/murmur.md`, `~/.zcode/skills/murmur/SKILL.md` | zcode has no SessionEnd hook. |
33
77
  | opencode | `~/.config/opencode/plugins/murmur.ts`, `~/.config/opencode/commands/murmur.md` | The plugin is the whole integration (tools, push, presence). |
34
78
  | hummin | `~/.hummin/agent/extensions/murmur.ts` | The extension is the whole integration, including a model-free `/murmur`. |
35
79
 
@@ -50,29 +94,29 @@ Delivery states, per recipient: `queued` (stored, nothing left the broker), `sen
50
94
 
51
95
  | CLI | Live push into a running session | Waiting messages also arrive | Evidence for `delivered` |
52
96
  | --- | --- | --- | --- |
53
- | Claude Code | Native messaging socket (`CLAUDE_CODE_MESSAGING_SOCKET`): the session gets the message as a peer turn; an idle session starts working on it | At SessionStart, each prompt, and at turn end (Stop continues the turn) | The message tag found in the receiver's own transcript |
54
- | Hummin | Extension calls `sendUserMessage` (follow-up when idle, steer mid-turn) | When the extension (re)connects | The extension confirms its session accepted it; the tag in the session file too |
55
- | opencode | Plugin posts the message into the session with `promptAsync` | When the plugin (re)connects | The plugin confirms the post succeeded |
56
- | Codex | `codex queue --thread <id>`; a running window takes it as a user turn | At SessionStart, each prompt, and at turn end (Stop continues the turn) | The tag in the receiver's rollout transcript |
57
- | Gemini CLI | None: Gemini has no inbound channel | At SessionStart, each prompt (BeforeAgent), and at turn end (AfterAgent continues the turn) | Handed over by the hook |
58
- | `zcode` | Hooks fire under workspace trust; the session announces; one message is handed over by a hook and answered | BLOCKED: the provider answers 429 "Insufficient balance or no resource package. Please recharge." and the session never announced in headless mode, so whether hooks fire under `-p` is unverified |
97
+ | [Claude Code](https://claude.com/product/claude-code) | Native messaging socket (`CLAUDE_CODE_MESSAGING_SOCKET`): the session gets the message as a peer turn; an idle session starts working on it | At SessionStart, each prompt, and at turn end (Stop continues the turn) | The message tag found in the receiver's own transcript |
98
+ | [Hummin](https://github.com/rennf93/hummin) | Extension calls `sendUserMessage` (follow-up when idle, steer mid-turn) | When the extension (re)connects | The extension confirms its session accepted it; the tag in the session file too |
99
+ | [opencode](https://github.com/anomalyco/opencode) | Plugin posts the message into the session with `promptAsync` | When the plugin (re)connects | The plugin confirms the post succeeded |
100
+ | [Codex](https://github.com/openai/codex) | `codex queue --thread <id>`; a running window takes it as a user turn | At SessionStart, each prompt, and at turn end (Stop continues the turn) | The tag in the receiver's rollout transcript |
101
+ | [Gemini CLI](https://github.com/google-gemini/gemini-cli) | None: Gemini has no inbound channel | At SessionStart, each prompt (BeforeAgent), and at turn end (AfterAgent continues the turn) | Handed over by the hook |
102
+ | [zcode](https://zcode.z.ai) | None found | At SessionStart, each prompt, and at turn end (Stop continues the turn) | Handed over by the hook (zcode's hook `transcript_path` is a temporary copy, so it cannot serve as proof) |
59
103
 
60
- Delivery modes: `push` (default) uses every path above; `notify` stores the message and sends a small batched note; `silent` only stores. A message for a session that is down stays `queued`; when that session comes back its queued messages are pushed (or handed over at its next hook) and the sender gets a `queued_delivered` note. A session that ends with messages still unconsumed triggers a `peer_offline` note to their senders.
104
+ Delivery modes: `push` (default) uses every path above; `notify` stores the message and sends a small batched note; `silent` only stores. A message for a session that is down stays `queued`; when that session comes back its queued messages are pushed (or handed over at its next hook) and the sender gets a `queued_delivered` note (not for a reply held for the asker's open `murmur_wait`: the wait result is its delivery, and the replier is not woken for it). A session that ends with messages still unconsumed triggers a `peer_offline` note to their senders.
61
105
 
62
106
  What a session sees for every message, on every path:
63
107
 
64
108
  ```
65
109
  [murmur m42 t7 s0 from ledger-3 to archify-2, prs-5]
66
110
  <body, cut at 4000 characters with a pointer to the full text>
67
- (reply: murmur_reply messageId 42, or from a shell: murmur reply 42 <text>; whole thread: murmur_thread 7)
111
+ (answer with murmur_reply messageId 42, plain text never reaches the sender; from a shell: murmur reply 42 <text>; its replies: murmur_thread 7 messageId 42)
68
112
  ```
69
113
 
70
114
  ## Threads, groups, verdicts
71
115
 
72
116
  - One message, one id, fanned out. Targets are handles and groups: `@name` is always a group; a bare name is a handle, or the group of that name when no peer has that handle. Overlapping groups dedupe.
73
- - Replies go to everyone in the thread (every sender, every recipient and every session an owner message was typed in, so far), in the parent's mode, so peers see each other's answers without relaying. To consult several sessions, send once to all of them.
117
+ - A reply is stored once in the thread and goes only to whoever is waiting for it (the author of the message it answers and the session an owner message was typed in), in the parent's mode; everyone in the thread can read it with `murmur_thread` / `murmur thread`, and `murmur_wait` returns the replies' text. `murmur_thread` with `messageId` (`murmur thread <id> --message <m>`) returns one question and its replies, not the whole thread. To consult several sessions, send once to all of them.
74
118
  - Every message has a per-thread monotonic `seq`. `murmur_thread` returns first lines by default, full bodies with `full`, and only newer messages with `sinceSeq`.
75
- - Verdicts: `agree`, `disagree`, `info`, with the body starting `AGREE`, `DISAGREE` or `INFO`. `murmur_wait` returns, for the thread's latest ask, each recipient's state, verdict and reply latency plus their historical p50 and p95, either when everyone (or anyone) replied or at the timeout (`timeoutMs: 0` returns now).
119
+ - Verdicts: `agree`, `disagree`, `info`, with the body starting `AGREE`, `DISAGREE` or `INFO`. `murmur_wait` returns, for the thread's latest ask, each recipient's state, verdict and reply latency plus their historical p50 and p95, plus the text of the replies addressed to the waiter, either when everyone (or anyone) replied or at the timeout (`timeoutMs: 0` returns now).
76
120
  - `references` carries pointers (paths, commits, URLs) so artifacts stay in their repos.
77
121
  - Groups: create, list, add, remove, delete. Deleting a group never touches sessions, threads or history.
78
122
 
@@ -85,16 +129,12 @@ The human surface, as the owner:
85
129
  - `/murmur <group | @group | handle[,more]> <message>`; bare `/murmur` lists peers and groups. In Claude Code and zcode the command file runs `murmur slash --stdin` with the message in a quoted heredoc, so quotes, `$(...)`, backticks and `;` arrive verbatim (Claude Code turns a word-initial `!` into `\!` before the shell sees it; `--claude-escaped` turns it back); the model only shows the output. Hummin's `/murmur` runs with no model turn at all.
86
130
  - `murmur send <targets> <body...> [--mode push|notify|silent] [--refs a,b] [--thread <id>] [--as <handle>]`
87
131
  - `murmur reply <id> <body...> [--verdict agree|disagree|info] [--refs a,b] [--as <handle>]`
88
- - `murmur status`, `murmur thread <id> [--since <seq>] [--full]`, `murmur groups [list|create|add|remove|delete] ...`, `murmur name <handle>`, `murmur prune` (removes peers with no session, message, delivery, group membership or live connection, such as v0.1 leftovers, and prints their handles)
89
-
90
- ## Doctrine
91
-
92
- The bus removes two costs and must never remove a third thing. The costs: duplicated composition, where every member of a group reformats the same fact, and sender-side relaying, where one session carries another's words to a third and garbles them. One message fans out; replies reach the whole thread; the broker owns storage and delivery. Delivery modes are the token dial. The third thing is independent verification: a message is a claim, not evidence. Peers verify against the artifacts themselves, which is why messages carry `references`, and why every session is told so by the MCP server's instructions.
132
+ - `murmur status`, `murmur thread <id> [--since <seq>] [--full] [--message <id>]`, `murmur groups [list|create|add|remove|delete] ...`, `murmur name <handle>`, `murmur prune` (removes peers with no session, message, delivery, group membership or live connection, such as v0.1 leftovers, and prints their handles)
93
133
 
94
134
  ## Limits
95
135
 
96
136
  - 20 sends per sender per 60 seconds; an identical send (sender, body, recipients) within 60 seconds is rejected as a duplicate. Replies count. In memory; reset on broker restart.
97
- - 500 messages per thread; 200 unread deliveries per recipient; 1 MiB per request line.
137
+ - 500 messages per thread; 200 deliveries per recipient not yet proven delivered to its session (queued or sent); 1 MiB per request line.
98
138
  - `murmur_wait`: default 120 s, at most 300 s.
99
139
  - Hooks: 10 s hard deadline, up to 10 messages per pull, exit 0 on every failure path.
100
140
  - Transcript proof polls for 120 s after a push, then re-checks at that session's next hook event.
@@ -115,7 +155,7 @@ The bus removes two costs and must never remove a third thing. The costs: duplic
115
155
 
116
156
  | Check | What it proves | Last result (2026-10-20) |
117
157
  | --- | --- | --- |
118
- | `claude` | Two sessions announce with native transports; one message to both goes sent, delivered (tag in each transcript), replied; each reply fans out to the other session and the owner; `/murmur` stores `$(...)`, backticks, quotes, `!` and `$HOME` verbatim | PASS, haiku, replies in about 3.6 s |
158
+ | `claude` | Two sessions announce with native transports; one message to both goes sent, delivered (tag in each transcript), replied; each reply reaches the owner who asked and not the other session; `/murmur` stores `$(...)`, backticks, quotes, `!` and `$HOME` verbatim | PASS, haiku, replies in about 3.6 s |
119
159
  | `codex` | The session announces with its thread id; `codex queue` reaches the running window; transcript proof; reply through the MCP tool | PASS, gpt-5.6-luna, 6.2 s |
120
160
  | `hummin` | The extension injects, acks delivered, carries the reply; the model-free `/murmur` | PASS, zai/glm-5.3-flash, 6.0 s |
121
161
  | `opencode` | The plugin announces, injects with `promptAsync`, acks delivered, carries the reply | PASS, opencode 1.19.3, opencode/glm-5.3-flash-free, reply in 5.2 s (the readiness check can take up to 60 s on the first start after an upgrade) |
@@ -363,7 +363,7 @@ export default function murmur(pi: ExtensionAPI): void {
363
363
  name: "murmur_reply",
364
364
  label: "Murmur Reply",
365
365
  description:
366
- "Reply to a murmur message; the reply reaches everyone in the thread, so start the body with AGREE, DISAGREE or INFO.",
366
+ "Reply to a murmur message; it goes to whoever is waiting for it and anyone in the thread can read it, so start the body with AGREE, DISAGREE or INFO.",
367
367
  promptSnippet: "murmur_reply: answer a [murmur m<id>] message",
368
368
  parameters: Type.Object({
369
369
  messageId: Type.Number({ description: "Id from the [murmur m<id>] tag" }),
@@ -249,7 +249,7 @@ const plugin: Plugin = async ({ client, directory }) => {
249
249
  ctxCall({ type: 'send', threadId: a.threadId ?? null, to: a.to, body: a.body, references: a.references, mode: a.mode }, ctx),
250
250
  }),
251
251
  murmur_reply: tool({
252
- description: 'Reply to a murmur message; the reply reaches everyone in the thread, so start the body with AGREE, DISAGREE or INFO.',
252
+ description: 'Reply to a murmur message; it goes to whoever is waiting for it and anyone in the thread can read it, so start the body with AGREE, DISAGREE or INFO.',
253
253
  args: {
254
254
  messageId: s.number(),
255
255
  body: s.string(),
package/dist/broker.js CHANGED
@@ -130,6 +130,8 @@ export class Broker {
130
130
  notifyQueue = new Map();
131
131
  // Thread id to waiters registered against that thread.
132
132
  waiters = new Map();
133
+ // Deliveries a push is sending right now (messageId:recipient): no other path hands them over.
134
+ inFlight = new Set();
133
135
  stopped = false;
134
136
  stopPromise = null;
135
137
  constructor(opts, server) {
@@ -271,6 +273,7 @@ export class Broker {
271
273
  }
272
274
  onSocketClose(conn) {
273
275
  this.conns.delete(conn);
276
+ const unresolved = conn.waiters.size > 0;
274
277
  for (const waiter of [...conn.waiters]) {
275
278
  clearTimeout(waiter.timer);
276
279
  this.removeWaiter(waiter);
@@ -278,6 +281,10 @@ export class Broker {
278
281
  if (conn.handle !== null && this.presence.get(conn.handle) === conn) {
279
282
  this.presence.delete(conn.handle);
280
283
  }
284
+ // Replies held for the closed wait have no result to ride on: push them now.
285
+ if (unresolved && conn.handle !== null && !this.stopped) {
286
+ void this.pushQueued(conn.handle).catch(() => { });
287
+ }
281
288
  // Stamp nothing on close; only register refreshes last_seen_at. Only a
282
289
  // long-lived connection going away means the peer went away.
283
290
  if (conn.handle !== null && conn.live && !this.stopped) {
@@ -671,7 +678,8 @@ export class Broker {
671
678
  handleThread(rec) {
672
679
  const threadId = reqNumber(rec, 'threadId');
673
680
  const sinceSeq = reqNumberOrNull(rec, 'sinceSeq');
674
- return this.core.threadView(threadId, sinceSeq, reqBool(rec, 'full'));
681
+ const messageId = rec.messageId === undefined ? undefined : reqNumber(rec, 'messageId');
682
+ return this.core.threadView(threadId, sinceSeq, reqBool(rec, 'full'), messageId);
675
683
  }
676
684
  handleCollect(rec) {
677
685
  const threadId = reqNumber(rec, 'threadId');
@@ -711,7 +719,7 @@ export class Broker {
711
719
  }
712
720
  // Everything the pull hands out was queued, so each one tells its
713
721
  // sender it finally arrived.
714
- const result = this.core.pendingPush(handle, limit);
722
+ const result = this.core.pendingPush(handle, limit, (id) => this.inFlight.has(String(id) + ':' + handle));
715
723
  for (const message of result.messages) {
716
724
  this.enqueueNotification(message.sender, {
717
725
  kind: 'queued_delivered',
@@ -735,7 +743,7 @@ export class Broker {
735
743
  if (this.stopped) {
736
744
  return;
737
745
  }
738
- if (target.kind === 'queued') {
746
+ if (target.kind === 'queued' || this.heldForWait(target.message, target.recipient)) {
739
747
  continue;
740
748
  }
741
749
  await this.pushOne(target.message.id, target.message.threadId, target.recipient, target.text, {
@@ -746,6 +754,22 @@ export class Broker {
746
754
  });
747
755
  }
748
756
  }
757
+ // A reply answering the thread's latest ask, whose recipient has a
758
+ // murmur_wait open on that thread, stays queued; the wait result carries it
759
+ // (takeReplies applies the same rule). Any other reply is pushed as normal.
760
+ heldForWait(message, recipient) {
761
+ if (!message.isReply || message.inReplyTo === null || this.store.latestAsk(message.threadId)?.id !== message.inReplyTo) {
762
+ return false;
763
+ }
764
+ let held = false;
765
+ for (const w of this.waiters.get(message.threadId) ?? []) {
766
+ if (w.conn.handle === recipient) {
767
+ w.held.add(message.id);
768
+ held = true;
769
+ }
770
+ }
771
+ return held;
772
+ }
749
773
  async pushOne(messageId, threadId, recipient, text, payload) {
750
774
  const listener = this.presence.get(recipient);
751
775
  if (listener !== undefined && listener.listening && !listener.socket.destroyed) {
@@ -754,6 +778,16 @@ export class Broker {
754
778
  this.markSent(messageId, threadId, recipient);
755
779
  return;
756
780
  }
781
+ const key = String(messageId) + ':' + recipient;
782
+ this.inFlight.add(key);
783
+ try {
784
+ await this.pushVia(messageId, threadId, recipient, text);
785
+ }
786
+ finally {
787
+ this.inFlight.delete(key);
788
+ }
789
+ }
790
+ async pushVia(messageId, threadId, recipient, text) {
757
791
  const session = this.liveSessionOf(recipient);
758
792
  const transcript = session?.transcriptPath ?? null;
759
793
  const offset = transcript === null ? 0 : ProofWatcher.offsetOf(transcript);
@@ -783,6 +817,9 @@ export class Broker {
783
817
  async pushQueued(handle) {
784
818
  const waiting = this.store.pendingPushFor(handle, 50).reverse();
785
819
  for (const { message } of waiting) {
820
+ if (this.heldForWait(message, handle)) {
821
+ continue;
822
+ }
786
823
  const recipients = this.store.getDeliveries(message.id).map((d) => d.recipient);
787
824
  const full = fullOf(message);
788
825
  const text = renderDelivery(full, message.deliveredBody, recipients);
@@ -807,14 +844,14 @@ export class Broker {
807
844
  }
808
845
  // queued or sent to delivered, with the queued_delivered notice to the
809
846
  // sender when the message had been waiting. False when not applicable.
810
- markDeliveredHop(messageId, recipient) {
847
+ markDeliveredHop(messageId, recipient, notifySender = true) {
811
848
  const message = this.store.messageById(messageId);
812
849
  const delivery = this.store.getDeliveries(messageId).find((d) => d.recipient === recipient);
813
850
  if (message === null || delivery === undefined || (delivery.state !== 'queued' && delivery.state !== 'sent')) {
814
851
  return false;
815
852
  }
816
853
  this.store.markDelivered(messageId, recipient, Date.now());
817
- if (delivery.state === 'queued') {
854
+ if (delivery.state === 'queued' && notifySender) {
818
855
  this.enqueueNotification(message.sender, {
819
856
  kind: 'queued_delivered',
820
857
  threadId: message.threadId,
@@ -915,6 +952,7 @@ export class Broker {
915
952
  conditionMet: true,
916
953
  timedOut: false,
917
954
  matrix: this.matrixFor(threadId),
955
+ replies: this.takeReplies(threadId, conn.handle),
918
956
  });
919
957
  return;
920
958
  }
@@ -923,6 +961,7 @@ export class Broker {
923
961
  id,
924
962
  threadId,
925
963
  until,
964
+ held: new Set(),
926
965
  timer: setTimeout(() => this.resolveTimedOut(waiter), timeoutMs),
927
966
  };
928
967
  conn.waiters.add(waiter);
@@ -950,6 +989,22 @@ export class Broker {
950
989
  }
951
990
  return deliveries.every((d) => d.state === 'replied');
952
991
  }
992
+ // The replies still queued for this handle in the thread, handed over in the
993
+ // wait result and marked delivered (the wait result is the receiving-side proof).
994
+ takeReplies(threadId, handle, held) {
995
+ const replies = [];
996
+ const ask = this.store.latestAsk(threadId);
997
+ for (const row of this.store.threadMessages(threadId, null)) {
998
+ const delivery = this.store.getDeliveries(row.id).find((d) => d.recipient === handle);
999
+ if (((ask !== null && row.inReplyTo === ask.id) || held?.has(row.id) === true) && delivery?.state === 'queued' && !this.inFlight.has(String(row.id) + ':' + delivery.recipient)) {
1000
+ // A reply held for this wait never waited on a session that was down:
1001
+ // no queued_delivered notice, which would wake the replier for nothing.
1002
+ this.markDeliveredHop(row.id, delivery.recipient, held?.has(row.id) !== true);
1003
+ replies.push({ id: row.id, seq: row.seq, sender: row.sender, verdict: row.verdict, body: row.deliveredBody });
1004
+ }
1005
+ }
1006
+ return replies;
1007
+ }
953
1008
  // The collect matrix for a WaitResult. A thread without messages or without
954
1009
  // an ask collects nothing, which degrades to an empty matrix.
955
1010
  matrixFor(threadId) {
@@ -976,7 +1031,8 @@ export class Broker {
976
1031
  if (this.stopped) {
977
1032
  return;
978
1033
  }
979
- if (!this.conditionHolds(threadId, waiter.until)) {
1034
+ // A handover below re-enters checkWaiters and may already have answered this waiter.
1035
+ if (!set.has(waiter) || !this.conditionHolds(threadId, waiter.until)) {
980
1036
  continue;
981
1037
  }
982
1038
  this.removeWaiter(waiter);
@@ -986,6 +1042,7 @@ export class Broker {
986
1042
  conditionMet: true,
987
1043
  timedOut: false,
988
1044
  matrix: this.matrixFor(threadId),
1045
+ replies: this.takeReplies(threadId, waiter.conn.handle, waiter.held),
989
1046
  });
990
1047
  }
991
1048
  }
@@ -1000,19 +1057,20 @@ export class Broker {
1000
1057
  conditionMet: false,
1001
1058
  timedOut: true,
1002
1059
  matrix: this.matrixFor(waiter.threadId),
1060
+ replies: this.takeReplies(waiter.threadId, waiter.conn.handle, waiter.held),
1003
1061
  });
1004
1062
  }
1005
1063
  catch {
1006
1064
  // Never let a timer callback crash the process.
1007
1065
  }
1008
1066
  }
1067
+ // Only called for a registered waiter: checkWaiters checks set.has, a close
1068
+ // walks conn.waiters, and a timeout fires only while the timer was never cleared.
1009
1069
  removeWaiter(waiter) {
1010
1070
  const set = this.waiters.get(waiter.threadId);
1011
- if (set !== undefined) {
1012
- set.delete(waiter);
1013
- if (set.size === 0) {
1014
- this.waiters.delete(waiter.threadId);
1015
- }
1071
+ set.delete(waiter);
1072
+ if (set.size === 0) {
1073
+ this.waiters.delete(waiter.threadId);
1016
1074
  }
1017
1075
  waiter.conn.waiters.delete(waiter);
1018
1076
  }
package/dist/cli.js CHANGED
@@ -11,7 +11,7 @@ const OWNER_DISPLAY = 'renn (owner)';
11
11
  const SEND_USAGE = 'usage: murmur send <targets> <body...> [--as <handle>] [--mode <mode>] [--refs <a,b>] [--thread <id>]';
12
12
  const REPLY_USAGE = 'usage: murmur reply <message id> <body...> [--as <handle>] [--verdict agree|disagree|info] [--refs <a,b>]';
13
13
  const NAME_USAGE = 'usage: murmur name <handle> (renames the agent session this runs in)';
14
- const THREAD_USAGE = 'usage: murmur thread <id> [--since <n>] [--full]';
14
+ const THREAD_USAGE = 'usage: murmur thread <id> [--since <n>] [--full] [--message <id>]';
15
15
  const GROUPS_USAGE = 'usage: murmur groups [list] | groups create <name> | groups add <name> <members...> | groups remove <name> <members...> | groups delete <name>';
16
16
  const SLASH_USAGE = '/murmur <group | handle[,more]> <message> (sent as renn, via the session you type it in; run bare like this for peers and groups)';
17
17
  export async function runCli(argv, io) {
@@ -95,7 +95,7 @@ function printUsage(err) {
95
95
  ' slash [args...] the /murmur backing, sent as the owner via your session; bare prints peers and groups',
96
96
  ' prune remove peers with no session, message, group or connection',
97
97
  ' status list peers with presence and unread counts',
98
- ' thread <id> [--since <n>] [--full]',
98
+ ' thread <id> [--since <n>] [--full] [--message <id>]',
99
99
  ' groups [list] | groups create <name> | groups add <name> <members...>',
100
100
  ' | groups remove <name> <members...> | groups delete <name>',
101
101
  ' install <' + INSTALL_CLIENTS.join('|') + '>',
@@ -430,6 +430,7 @@ async function cmdPrune(out) {
430
430
  async function cmdThread(args, out, err) {
431
431
  let full = false;
432
432
  let sinceSeq = null;
433
+ let messageId;
433
434
  const positionals = [];
434
435
  for (let i = 0; i < args.length; i += 1) {
435
436
  const arg = args[i];
@@ -440,6 +441,16 @@ async function cmdThread(args, out, err) {
440
441
  full = true;
441
442
  continue;
442
443
  }
444
+ if (arg === '--message') {
445
+ const parsed = Number(args[i + 1]);
446
+ if (args[i + 1] === undefined || !Number.isInteger(parsed) || parsed < 0) {
447
+ err('murmur thread: --message needs a non-negative integer\n' + THREAD_USAGE + '\n');
448
+ return 1;
449
+ }
450
+ i += 1;
451
+ messageId = parsed;
452
+ continue;
453
+ }
443
454
  if (arg === '--since') {
444
455
  const value = args[i + 1];
445
456
  if (value === undefined) {
@@ -469,10 +480,10 @@ async function cmdThread(args, out, err) {
469
480
  }
470
481
  const client = await connectCli();
471
482
  try {
472
- const msg = { type: 'thread', threadId, sinceSeq, full };
483
+ const msg = { type: 'thread', threadId, sinceSeq, full, ...(messageId !== undefined ? { messageId } : {}) };
473
484
  const result = await client.request(msg);
474
485
  for (const message of result.messages) {
475
- const text = full ? message.body : message.firstLine;
486
+ const text = full || messageId !== undefined ? message.body : message.firstLine;
476
487
  out(String(message.seq) + ' ' + message.sender + ' ' + text + '\n');
477
488
  }
478
489
  return 0;
package/dist/client.js CHANGED
@@ -35,24 +35,22 @@ export class MurmurClient {
35
35
  const socket = createConnection(opts.socketPath);
36
36
  return new Promise((resolve, reject) => {
37
37
  let settled = false;
38
- let timer = null;
39
38
  const finish = (err, client) => {
40
39
  if (settled) {
41
40
  return;
42
41
  }
43
42
  settled = true;
44
- if (timer !== null) {
45
- clearTimeout(timer);
46
- }
43
+ clearTimeout(timer);
47
44
  if (err !== null) {
48
45
  socket.destroy();
49
46
  reject(err);
50
47
  }
51
- else if (client !== null) {
48
+ else {
49
+ // Every caller passes exactly one of err and client.
52
50
  resolve(client);
53
51
  }
54
52
  };
55
- timer = setTimeout(() => {
53
+ const timer = setTimeout(() => {
56
54
  finish(new MurmurError('broker_down', 'connect to ' + opts.socketPath + ' timed out after ' + String(timeoutMs) + 'ms'), null);
57
55
  }, timeoutMs);
58
56
  socket.once('error', (err) => {
package/dist/core.js CHANGED
@@ -5,6 +5,10 @@ export { HANDLE_RE };
5
5
  const GROUP_NAME_RE = /^@?[a-z0-9][a-z0-9-]{1,31}$/;
6
6
  const VERDICTS = ['agree', 'disagree', 'info'];
7
7
  const UNREAD_STATES = ['queued', 'sent', 'delivered'];
8
+ // The inbox cap guards the backlog not yet proven in the session: a
9
+ // delivered message is in the session, and counting it would let replies that
10
+ // are never acked fill the cap and block every later reply to the asker.
11
+ const BACKLOG_STATES = ['queued', 'sent'];
8
12
  const OFFLINE_AFTER_MS = 5 * 60 * 1000;
9
13
  // A session title as a handle: lowercase, runs of anything else become one
10
14
  // hyphen, trimmed, cut to the handle limit. Not necessarily a valid handle.
@@ -165,7 +169,7 @@ export class MurmurCore {
165
169
  const stripped = stripGroupPrefix(entry);
166
170
  const group = this.store.listGroups().find((g) => g.name === stripped);
167
171
  if (!entry.startsWith('@') && (group === undefined || this.store.getPeer(entry) !== null)) {
168
- add(entry);
172
+ add(this.store.getPeer(entry) === null ? this.handleForName(entry) : entry);
169
173
  continue;
170
174
  }
171
175
  if (group === undefined) {
@@ -180,6 +184,21 @@ export class MurmurCore {
180
184
  }
181
185
  return out;
182
186
  }
187
+ // A bare name that is no handle: a session's display name (case-insensitive)
188
+ // or a name whose slug is a handle. Several matches are ambiguous. No match
189
+ // keeps the plain unknown handle error.
190
+ handleForName(name) {
191
+ const lower = name.toLowerCase();
192
+ const slug = slugify(name);
193
+ const found = this.store
194
+ .listPeers()
195
+ .filter((p) => p.handle === slug || p.display?.toLowerCase() === lower)
196
+ .map((p) => p.handle);
197
+ if (found.length > 1) {
198
+ throw new MurmurError('bad_request', 'ambiguous name ' + name + ', use one of: ' + found.join(', '));
199
+ }
200
+ return found[0] ?? name;
201
+ }
183
202
  // Rate, duplicate and cap checks for one intended send. A send that passes
184
203
  // every check consumes one rate slot and one duplicate fingerprint; a send
185
204
  // rejected by any check consumes none.
@@ -200,9 +219,9 @@ export class MurmurCore {
200
219
  throw new MurmurError('cap_thread', 'thread ' + String(threadId) + ' already holds ' + String(THREAD_CAP) + ' messages');
201
220
  }
202
221
  for (const recipient of recipients) {
203
- const unread = this.store.deliveryStatesForRecipient(recipient, UNREAD_STATES);
204
- if (unread >= UNREAD_CAP) {
205
- throw new MurmurError('cap_inbox', 'recipient ' + recipient + ' already has ' + String(unread) + ' unread deliveries');
222
+ const backlog = this.store.deliveryStatesForRecipient(recipient, BACKLOG_STATES);
223
+ if (backlog >= UNREAD_CAP) {
224
+ throw new MurmurError('cap_inbox', 'recipient ' + recipient + ' already has ' + String(backlog) + ' deliveries waiting (queued or sent)');
206
225
  }
207
226
  }
208
227
  recent.push(ts);
@@ -277,9 +296,9 @@ export class MurmurCore {
277
296
  }
278
297
  return this.store.getNativeTransport(recipient) !== null ? 'native' : 'connection';
279
298
  }
280
- // A reply reaches everyone in the thread (every sender, every recipient and
281
- // every session an owner message was typed in, minus the replier) in the parent's delivery mode, so peers see
282
- // each other's answers without anyone relaying them.
299
+ // A reply is stored once in the thread; only the ones waiting for it are
300
+ // delivered it (in the parent's delivery mode). Everyone else in the thread
301
+ // reads it with the thread view.
283
302
  reply(sender, arg) {
284
303
  if (arg.verdict !== undefined && !VERDICTS.includes(arg.verdict)) {
285
304
  throw new MurmurError('bad_request', 'invalid verdict: ' + arg.verdict);
@@ -288,15 +307,14 @@ export class MurmurCore {
288
307
  if (parent === null) {
289
308
  throw new MurmurError('unknown_message', 'unknown message: ' + String(arg.messageId));
290
309
  }
291
- const participants = new Set(this.store.threadRecipientHandles(parent.threadId));
292
- for (const row of this.store.threadMessages(parent.threadId, null)) {
293
- participants.add(row.sender);
294
- if (row.via !== null) {
295
- participants.add(row.via);
296
- }
297
- }
298
- participants.delete(sender);
299
- const recipients = [...participants].filter((handle) => this.store.getPeer(handle) !== null);
310
+ // Only the ones waiting for this reply get a row: the author of the message
311
+ // it answers plus the session that message was typed in. A reply inside
312
+ // its own thread (the replier wrote the parent) goes to the parent's recipients.
313
+ const waiting = new Set(parent.sender === sender
314
+ ? this.store.getDeliveries(parent.id).map((d) => d.recipient)
315
+ : [parent.sender, ...(parent.via === null ? [] : [parent.via])]);
316
+ waiting.delete(sender);
317
+ const recipients = [...waiting].filter((handle) => this.store.getPeer(handle) !== null);
300
318
  this.checkLimits(sender, arg.body, recipients, parent.threadId);
301
319
  const mode = parent.mode;
302
320
  const ts = this.nowMs();
@@ -418,13 +436,20 @@ export class MurmurCore {
418
436
  threads.sort((a, b) => b.latestSeq - a.latestSeq);
419
437
  return { handle, unreadTotal, threads };
420
438
  }
421
- threadView(threadId, sinceSeq, full = true) {
439
+ threadView(threadId, sinceSeq, full = true, messageId) {
422
440
  const all = this.store.threadMessages(threadId, null);
423
441
  if (all.length === 0) {
424
442
  throw new MurmurError('unknown_thread', 'unknown thread: ' + String(threadId));
425
443
  }
426
444
  const last = all[all.length - 1];
427
445
  const latestSeq = last === undefined ? 0 : last.seq;
446
+ if (messageId !== undefined) {
447
+ if (!all.some((row) => row.id === messageId)) {
448
+ throw new MurmurError('bad_request', 'message ' + String(messageId) + ' is not in thread ' + String(threadId));
449
+ }
450
+ const asked = all.filter((row) => row.id === messageId || row.inReplyTo === messageId);
451
+ return { threadId, latestSeq, messages: asked.map((row) => ({ ...summaryOf(row), body: row.deliveredBody })) };
452
+ }
428
453
  const rows = sinceSeq === null ? all : this.store.threadMessages(threadId, sinceSeq);
429
454
  return { threadId, latestSeq, messages: rows.map(full ? fullOf : summaryOf) };
430
455
  }
@@ -810,11 +835,14 @@ export class MurmurCore {
810
835
  }
811
836
  // The hook injection pull: queued push-mode messages, newest first,
812
837
  // marked delivered as they are handed out, each with its rendered text.
813
- pendingPush(handle, limit) {
838
+ pendingPush(handle, limit, skip = () => false) {
814
839
  const ts = this.nowMs();
815
840
  const messages = [];
816
841
  const texts = [];
817
842
  for (const { message: row } of this.store.pendingPushFor(handle, limit)) {
843
+ if (skip(row.id)) {
844
+ continue;
845
+ }
818
846
  this.store.markDelivered(row.id, handle, ts);
819
847
  this.events.onStateChange({ threadId: row.threadId, recipient: handle, from: 'queued', to: 'delivered' });
820
848
  const message = fullOf(row);
package/dist/install.js CHANGED
@@ -12,6 +12,7 @@ const PRE_EXEC_COMMAND_MD = `---
12
12
  description: Send to a murmur group or peer as the owner; bare lists peers and groups
13
13
  argument-hint: <group | @group | handle[,more]> <message>
14
14
  allowed-tools: Bash(murmur slash:*)
15
+ disable-model-invocation: true
15
16
  ---
16
17
 
17
18
  \`\`\`!
@@ -106,7 +107,8 @@ function installClaude(home) {
106
107
  writeTextFile(commandPath, PRE_EXEC_COMMAND_MD);
107
108
  const claudeJsonPath = join(home, '.claude.json');
108
109
  const claudeJson = readJsonObject(claudeJsonPath);
109
- mergeMcpServer(objectAt(claudeJson, 'mcpServers', claudeJsonPath));
110
+ // alwaysLoad keeps the tool schemas in the prompt instead of behind ToolSearch (one call less per use).
111
+ mergeMcpServer(objectAt(claudeJson, 'mcpServers', claudeJsonPath), { alwaysLoad: true });
110
112
  writeJsonFile(claudeJsonPath, claudeJson);
111
113
  const settingsPath = join(home, '.claude', 'settings.json');
112
114
  const settings = readJsonObject(settingsPath);
@@ -219,13 +221,13 @@ function objectAt(parent, key, file) {
219
221
  // The murmur server entry, keeping any env the user added but dropping the
220
222
  // per-session handle placeholder older versions wrote: identity now comes
221
223
  // from the session's host process.
222
- function mergeMcpServer(servers) {
224
+ function mergeMcpServer(servers, extra = {}) {
223
225
  const existing = servers.murmur;
224
226
  const env = isRecord(existing) && isRecord(existing.env) ? { ...existing.env } : {};
225
227
  if (env.MURMUR_HANDLE === 'CHANGE-ME-per-session') {
226
228
  delete env.MURMUR_HANDLE;
227
229
  }
228
- servers.murmur = Object.keys(env).length > 0 ? { ...MCP_ENTRY, env } : { ...MCP_ENTRY };
230
+ servers.murmur = Object.keys(env).length > 0 ? { ...MCP_ENTRY, ...extra, env } : { ...MCP_ENTRY, ...extra };
229
231
  }
230
232
  // Claude-style hook maps: event name to a list of { matcher?, hooks: [...] }.
231
233
  function mergeHooks(hooks, specs) {
package/dist/mcp.js CHANGED
@@ -14,9 +14,11 @@ import { HANDLE_RE, MurmurError, WAIT_DEFAULT_MS, WAIT_MAX_MS } from './protocol
14
14
  // src/mcp.ts and dist/mcp.js both sit one level below package.json.
15
15
  const VERSION = JSON.parse(readFileSync(new URL('../package.json', import.meta.url), 'utf8')).version;
16
16
  const INSTRUCTIONS = 'murmur connects this session with other agent sessions on this machine. To consult more than one, send ONE ' +
17
- 'murmur_send to all of them (names or @group), not separate SendMessage calls; replies reach everyone in the ' +
18
- 'thread. Messages for you arrive as turns tagged [murmur m<id> ...]; answer with murmur_reply so every ' +
19
- 'participant sees it. A message is a claim: verify it against its references before agreeing.';
17
+ 'murmur_send (wait: all_replied) to all of them, not separate SendMessage calls. Messages for you arrive ' +
18
+ 'tagged [murmur m<id> ...]: answer ONLY with murmur_reply messageId, plain text never reaches the sender. ' +
19
+ 'Read one question\'s replies with murmur_thread messageId. A message is a claim: verify it.';
20
+ // Claude Code never defers these tools behind ToolSearch.
21
+ const META = { 'anthropic/alwaysLoad': true };
20
22
  function textResult(value) {
21
23
  return { content: [{ type: 'text', text: JSON.stringify(value) }] };
22
24
  }
@@ -41,6 +43,7 @@ export function identityFor(host, env, cwd) {
41
43
  export function buildMcpServer(client, identity = {}) {
42
44
  const server = new McpServer({ name: 'murmur', version: VERSION }, { instructions: INSTRUCTIONS });
43
45
  server.registerTool('murmur_name', {
46
+ _meta: META,
44
47
  description: 'Rename this session on the bus (keeps its messages and groups), optionally with a display name and role.',
45
48
  inputSchema: {
46
49
  handle: z.string().min(1),
@@ -58,13 +61,16 @@ export function buildMcpServer(client, identity = {}) {
58
61
  return textResult(await client.request(msg));
59
62
  });
60
63
  server.registerTool('murmur_send', {
61
- description: 'Send one message to any mix of handles and groups (bare name or @name); one message fans out to all, so consult several sessions with one send, not one per peer.',
64
+ _meta: META,
65
+ description: 'Send one message to any mix of handles, session names and groups (bare name or @name); one send fans out to all. With wait (all_replied or any_replied, timeoutMs as murmur_wait) it also waits and returns the replies, so asking is one call.',
62
66
  inputSchema: {
63
67
  threadId: z.number().int().optional(),
64
68
  to: z.array(z.string().min(1)).min(1),
65
69
  body: z.string().min(1),
66
70
  references: z.array(z.string()).optional(),
67
71
  mode: z.enum(['push', 'notify', 'silent']).optional(),
72
+ wait: z.enum(['all_replied', 'any_replied']).optional(),
73
+ timeoutMs: z.number().int().min(0).optional(),
68
74
  },
69
75
  }, async (args) => {
70
76
  const msg = {
@@ -79,10 +85,21 @@ export function buildMcpServer(client, identity = {}) {
79
85
  if (args.mode !== undefined) {
80
86
  msg.mode = args.mode;
81
87
  }
82
- return textResult(await client.request(msg));
88
+ const sent = await client.request(msg);
89
+ if (args.wait === undefined) {
90
+ return textResult(sent);
91
+ }
92
+ const waitMsg = {
93
+ type: 'wait',
94
+ threadId: sent.threadId,
95
+ until: args.wait,
96
+ timeoutMs: Math.min(args.timeoutMs ?? WAIT_DEFAULT_MS, WAIT_MAX_MS),
97
+ };
98
+ return textResult({ sent, wait: await client.request(waitMsg) });
83
99
  });
84
100
  server.registerTool('murmur_reply', {
85
- description: 'Reply to a message by id; reaches everyone in its thread. Start the body with AGREE, DISAGREE or INFO and set verdict to match.',
101
+ _meta: META,
102
+ description: 'Reply to a message by id; delivered to whoever is waiting for it, and anyone in the thread can read it with murmur_thread. Start the body with AGREE, DISAGREE or INFO and set verdict to match.',
86
103
  inputSchema: {
87
104
  messageId: z.number().int(),
88
105
  body: z.string().min(1),
@@ -100,6 +117,7 @@ export function buildMcpServer(client, identity = {}) {
100
117
  return textResult(await client.request(msg));
101
118
  });
102
119
  server.registerTool('murmur_ack', {
120
+ _meta: META,
103
121
  description: 'Tell senders you saw these messages and will answer later.',
104
122
  inputSchema: {
105
123
  ids: z.array(z.number().int()).min(1),
@@ -109,11 +127,13 @@ export function buildMcpServer(client, identity = {}) {
109
127
  return textResult(await client.request(msg));
110
128
  });
111
129
  server.registerTool('murmur_thread', {
112
- description: 'Read a thread: first lines by default, full bodies with full, only newer than sinceSeq.',
130
+ _meta: META,
131
+ description: 'Read a thread: first lines by default, full bodies with full, only newer than sinceSeq. With messageId, only that message and the replies to it (the question you care about), full bodies.',
113
132
  inputSchema: {
114
133
  threadId: z.number().int(),
115
134
  sinceSeq: z.number().int().optional(),
116
135
  full: z.boolean().optional(),
136
+ messageId: z.number().int().optional(),
117
137
  },
118
138
  }, async (args) => {
119
139
  const msg = {
@@ -121,11 +141,13 @@ export function buildMcpServer(client, identity = {}) {
121
141
  threadId: args.threadId,
122
142
  sinceSeq: args.sinceSeq ?? null,
123
143
  full: args.full ?? false,
144
+ ...(args.messageId !== undefined ? { messageId: args.messageId } : {}),
124
145
  };
125
146
  return textResult(await client.request(msg));
126
147
  });
127
148
  server.registerTool('murmur_wait', {
128
- description: 'Who got, read and answered the latest ask, with verdicts and latencies. Waits until all_replied or any_replied, up to timeoutMs (default 120000, max 300000; 0 returns now).',
149
+ _meta: META,
150
+ description: 'Who got, read and answered the latest ask, with verdicts and latencies, plus the text of the replies addressed to you. Waits until all_replied or any_replied, up to timeoutMs (default 120000, max 300000; 0 returns now).',
129
151
  inputSchema: {
130
152
  threadId: z.number().int(),
131
153
  until: z.enum(['all_replied', 'any_replied']),
@@ -142,6 +164,7 @@ export function buildMcpServer(client, identity = {}) {
142
164
  return textResult(await client.request(msg));
143
165
  });
144
166
  server.registerTool('murmur_groups', {
167
+ _meta: META,
145
168
  description: 'Create, list, add to, remove from, or delete a group. Deleting never touches sessions or history.',
146
169
  inputSchema: {
147
170
  action: z.enum(['create', 'list', 'add', 'remove', 'delete']),
@@ -159,6 +182,7 @@ export function buildMcpServer(client, identity = {}) {
159
182
  return textResult(await client.request(msg));
160
183
  });
161
184
  server.registerTool('murmur_status', {
185
+ _meta: META,
162
186
  description: 'Every peer with presence (busy, idle, online, offline), its CLI and unread counts.',
163
187
  }, async () => {
164
188
  const msg = { type: 'status' };
package/dist/native.js CHANGED
@@ -31,19 +31,16 @@ export async function nativeSend(socketPath, token, text, timeoutMs) {
31
31
  const reader = new LineReader();
32
32
  let settled = false;
33
33
  let lastSocketError = null;
34
- let timer = null;
35
34
  const finish = (result) => {
36
35
  if (settled) {
37
36
  return;
38
37
  }
39
38
  settled = true;
40
- if (timer !== null) {
41
- clearTimeout(timer);
42
- }
39
+ clearTimeout(timer);
43
40
  socket.destroy();
44
41
  resolve(result);
45
42
  };
46
- timer = setTimeout(() => {
43
+ const timer = setTimeout(() => {
47
44
  finish({ ok: false, error: 'no reply or close from ' + socketPath + ' within ' + String(limit) + 'ms' });
48
45
  }, limit);
49
46
  socket.once('error', (err) => {
package/dist/protocol.js CHANGED
@@ -72,8 +72,8 @@ export function deliveryTag(messageId) {
72
72
  return '[murmur m' + String(messageId) + ' ';
73
73
  }
74
74
  export function replyFooter(m) {
75
- return ('(reply: murmur_reply messageId ' + String(m.id) + ', or from a shell: murmur reply ' + String(m.id) +
76
- ' <text>; whole thread: murmur_thread ' + String(m.threadId) + ')');
75
+ return ('(answer with murmur_reply messageId ' + String(m.id) + ', plain text never reaches the sender; from a shell: murmur reply ' +
76
+ String(m.id) + ' <text>; its replies: murmur_thread ' + String(m.threadId) + ' messageId ' + String(m.id) + ')');
77
77
  }
78
78
  // What a session sees for one message, on every transport: a tag line with
79
79
  // ids, sender and recipients, the (possibly truncated) body, references, and
package/package.json CHANGED
@@ -1,6 +1,6 @@
1
1
  {
2
2
  "name": "murmur-it",
3
- "version": "0.2.2",
3
+ "version": "0.2.4",
4
4
  "description": "Local, client-agnostic message bus for AI CLI sessions with group threads",
5
5
  "license": "MIT",
6
6
  "author": "Renn F",
@@ -37,7 +37,7 @@
37
37
  "CHANGELOG.md"
38
38
  ],
39
39
  "engines": {
40
- "node": ">=20"
40
+ "node": ">=22"
41
41
  },
42
42
  "publishConfig": {
43
43
  "access": "public"
@@ -51,14 +51,14 @@
51
51
  "dependencies": {
52
52
  "@modelcontextprotocol/sdk": "^1.12.0",
53
53
  "better-sqlite3": "^12.2.0",
54
- "zod": "^3.24.0"
54
+ "zod": "^4.6.7"
55
55
  },
56
56
  "devDependencies": {
57
57
  "@types/better-sqlite3": "^7.6.13",
58
- "@types/node": "^24.0.0",
59
- "@vitest/coverage-v8": "^3.2.0",
58
+ "@types/node": "^26.6.5",
59
+ "@vitest/coverage-v8": "^5.0.5",
60
60
  "tsx": "^4.20.0",
61
- "typescript": "^5.8.0",
62
- "vitest": "^3.2.0"
61
+ "typescript": "^7.0.2",
62
+ "vitest": "^5.0.5"
63
63
  }
64
64
  }