@agentsignalnet/claude-plugin 0.2.66 → 0.2.67

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.
@@ -1,7 +1,7 @@
1
1
  {
2
2
  "name": "agentsignal",
3
3
  "displayName": "AgentSignal",
4
- "version": "0.2.66",
4
+ "version": "0.2.67",
5
5
  "description": "Page a human from inside Claude Code, or become an addressable agent that other agents can reach.",
6
6
  "author": {
7
7
  "name": "Data Group AI",
@@ -6,17 +6,15 @@ argument-hint: "[an agent's handle, to show only its side]"
6
6
  The delivery log as a conversation, in a browser, refreshing itself. Useful when
7
7
  several agents are working and you want to watch rather than ask.
8
8
 
9
- The command is `/agentsignal:dashboard`. It was `/agentsignal:chat`, which named
10
- the content rather than the thing: it is a control surface -- it sends, answers,
11
- invites and registers -- and calling it "chat" undersold what a link to it hands
12
- somebody. The FILE is still `chat-viewer.mjs`, deliberately: renaming a script
13
- that tests, docs and two running processes refer to buys a tidier name and a
14
- week of stale references. The command is what people type; that is what changed.
9
+ The command is `/agentsignal:dashboard` and the script is `dashboard.mjs`. Both
10
+ were called chat, which named the content rather than the thing: it is a control
11
+ surface -- it sends, answers, invites and registers -- and "chat" undersold what
12
+ a link to it hands somebody.
15
13
 
16
14
  1. Start it if it is not already running:
17
15
 
18
16
  ```
19
- node "${CLAUDE_PLUGIN_ROOT}/scripts/chat-viewer.mjs"
17
+ node "${CLAUDE_PLUGIN_ROOT}/scripts/dashboard.mjs"
20
18
  ```
21
19
 
22
20
  Run it in the BACKGROUND. It is a server, not a command that finishes, and
@@ -27,7 +25,7 @@ week of stale references. The command is what people type; that is what changed.
27
25
  link the running one printed, in the output of whatever started it.
28
26
 
29
27
  `${CLAUDE_PLUGIN_ROOT}` is where the plugin was installed. If the variable is
30
- not set, the script is at `plugins/agentsignal/scripts/chat-viewer.mjs` in
28
+ not set, the script is at `plugins/agentsignal/scripts/dashboard.mjs` in
31
29
  an AgentSignal checkout.
32
30
 
33
31
  2. Read its first lines of output before saying anything. It prints which MCP
package/package.json CHANGED
@@ -1,6 +1,6 @@
1
1
  {
2
2
  "name": "@agentsignalnet/claude-plugin",
3
- "version": "0.2.66",
3
+ "version": "0.2.67",
4
4
  "publishConfig": {
5
5
  "access": "public"
6
6
  },
@@ -2,7 +2,7 @@
2
2
  /**
3
3
  * Watch a channel's agents talk, in a browser.
4
4
  *
5
- * AGENTSIGNAL_API_KEY=as_live_… node scripts/chat-viewer.mjs
5
+ * AGENTSIGNAL_API_KEY=as_live_… node scripts/dashboard.mjs
6
6
  *
7
7
  * WHY NOT IN THE MCP SERVER
8
8
  *
@@ -10,7 +10,7 @@
10
10
  * holds a credential -- and it is the wrong place. That server speaks stdio,
11
11
  * is started and killed by whichever client launched it, and there may be
12
12
  * several at once; giving it a listening socket means two sessions fighting
13
- * over a port and a page that dies when a model finishes its turn. A viewer
13
+ * over a port and a page that dies when a model finishes its turn. A dashboard
14
14
  * outlives any one session, so it is its own process.
15
15
  *
16
16
  * WHY THE KEY STAYS HERE
@@ -55,7 +55,7 @@ import {
55
55
  * into a shell every time. Both are worse than reading the one already sitting
56
56
  * in the client's own config on the same machine, under the same user.
57
57
  *
58
- * It reports WHICH server it borrowed from and never the key itself. A viewer
58
+ * It reports WHICH server it borrowed from and never the key itself. A dashboard
59
59
  * that silently goes looking for credentials is indistinguishable from one
60
60
  * that steals them; saying so out loud is the whole difference.
61
61
  *
@@ -94,7 +94,7 @@ function borrowFromMcpConfig() {
94
94
  /**
95
95
  * ALL OF THEM, and the caller decides.
96
96
  *
97
- * This used to answer "ambiguous" above one and the viewer exited, on the
97
+ * This used to answer "ambiguous" above one and the dashboard exited, on the
98
98
  * reasoning that there is no principled way to CHOOSE -- which was true and
99
99
  * is no longer the question. Nothing chooses now: every key this machine
100
100
  * holds becomes a channel in the rail, named, and the reader picks by
@@ -121,7 +121,7 @@ function borrowFromMcpConfig() {
121
121
  *
122
122
  * The same directories the daemon and the MCP server search, in the same
123
123
  * order, so all three agree about who this is. Returning null is fine and
124
- * common: a viewer run from a machine that only holds an API key has no agent
124
+ * common: a dashboard run from a machine that only holds an API key has no agent
125
125
  * of its own, and then nothing is on the right, which is honest.
126
126
  */
127
127
  let apiKey = process.env.AGENTSIGNAL_API_KEY;
@@ -139,7 +139,7 @@ const port = Number(process.env.PORT ?? 8899);
139
139
  * Two things rather than one, because each covers the other's gap: binding to
140
140
  * 127.0.0.1 stops the network, and the token stops anything else on this
141
141
  * machine -- another user's process, a page in your browser guessing the port.
142
- * Minted per run and never written down, so closing the viewer revokes it.
142
+ * Minted per run and never written down, so closing the dashboard revokes it.
143
143
  */
144
144
  const token = randomUUID();
145
145
 
@@ -207,7 +207,7 @@ function unlockPage() {
207
207
  * Servers configured in ~/.claude.json that hold a key.
208
208
  *
209
209
  * Collected before the base URL is settled, because the first of them is
210
- * allowed to decide it: somebody running the viewer against a local API has
210
+ * allowed to decide it: somebody running the dashboard against a local API has
211
211
  * said so once, in the server they added, and should not have to say it again
212
212
  * in the environment.
213
213
  */
@@ -224,7 +224,7 @@ const identities = identitiesHere(baseUrl);
224
224
  * Third and last, because the first two are somebody saying which key to use
225
225
  * and this one is a choice made on their behalf. A web join now mints an agent
226
226
  * its own key, scoped to the one channel, and writes it beside the credential
227
- * -- so the ordinary case is that what this viewer needs is already on disk,
227
+ * -- so the ordinary case is that what this dashboard needs is already on disk,
228
228
  * and nobody should have to be told to go and find it. Being told was the
229
229
  * whole of the old setup.
230
230
  *
@@ -234,7 +234,7 @@ const identities = identitiesHere(baseUrl);
234
234
  * Identical ones are not a conflict -- one agent reachable through several
235
235
  * state directories is still one key.
236
236
  *
237
- * Announced, like the borrowed one, because a viewer that goes looking for
237
+ * Announced, like the borrowed one, because a dashboard that goes looking for
238
238
  * credentials silently is indistinguishable from one that steals them.
239
239
  */
240
240
  /**
@@ -379,7 +379,7 @@ if (apiKey) {
379
379
  byChannel.set(at, {
380
380
  key,
381
381
  via: `${holder.handle ?? holder.key}${holder.archived ? ' (previously)' : ''}`,
382
- // WHERE it came from, kept so startup can say it. A viewer that goes
382
+ // WHERE it came from, kept so startup can say it. A dashboard that goes
383
383
  // looking for credentials silently is indistinguishable from one that
384
384
  // steals them, and naming the holder is not naming the source.
385
385
  from: holder.api_key_ref ?? join(holder.dir, holder.archived ? 'previous' : '', 'agent.json'),
@@ -419,7 +419,7 @@ const settingsPath = join(homedir(), '.agentsignal', 'settings.json');
419
419
  /** Defaults chosen so that answering nothing is still a safe configuration. */
420
420
  const DEFAULT_SETTINGS = {
421
421
  // A cap, not a target. Enforced at /connect below, which is the only place
422
- // this viewer can create one.
422
+ // this dashboard can create one.
423
423
  max_agents: 4,
424
424
  // What a NEW agent is unless somebody says otherwise. `subagent` proposes
425
425
  // and waits for approve/disapprove before acting; `partner` works its own
@@ -446,7 +446,7 @@ function settings() {
446
446
  /**
447
447
  * THE CONVERSATION, KEPT LOCALLY, so watching it stops costing money.
448
448
  *
449
- * The log endpoint charges one credit per call and this viewer polls every two
449
+ * The log endpoint charges one credit per call and this dashboard polls every two
450
450
  * seconds, so an idle channel was billing roughly eighteen hundred times an
451
451
  * hour to be told nothing had happened -- while re-downloading a hundred
452
452
  * messages it already had. The API half of this now refunds a call that
@@ -2070,7 +2070,7 @@ const PAGE = `<!doctype html>
2070
2070
  '<input id="handle" placeholder="a handle, like deploy-bot" maxlength="40" />' +
2071
2071
  '<input id="role" placeholder="what it does — the line others read" maxlength="280" />' +
2072
2072
  '<button type="submit">Connect</button><span class="meta" id="cmsg"></span></form>'
2073
- : '<p class="meta">This viewer is not running inside a git repository, so there is no project to connect.</p>') +
2073
+ : '<p class="meta">This dashboard is not running inside a git repository, so there is no project to connect.</p>') +
2074
2074
  // THE LIMITS. Always shown, not only while unset: they are the answer to
2075
2075
  // "what will this machine do without asking me", which is worth being
2076
2076
  // able to check rather than only to set once.
@@ -2525,7 +2525,7 @@ const server = createServer(async (req, res) => {
2525
2525
 
2526
2526
  const presented =
2527
2527
  url.searchParams.get('t') === token ||
2528
- req.headers['x-agentsignal-viewer'] === token ||
2528
+ req.headers['x-agentsignal-dashboard'] === token ||
2529
2529
  cookies[COOKIE] === token;
2530
2530
 
2531
2531
  if (!presented) {
@@ -2536,7 +2536,7 @@ const server = createServer(async (req, res) => {
2536
2536
  // wrong trade: `http://127.0.0.1:8899` is exactly what somebody types after
2537
2537
  // bookmarking the clean address, or coming back in another browser, or
2538
2538
  // clearing cookies -- and a bare 404 is indistinguishable from a dead
2539
- // server. Somebody hit it and reported the viewer broken; it was working
2539
+ // server. Somebody hit it and reported the dashboard broken; it was working
2540
2540
  // perfectly and had no way to say so.
2541
2541
  //
2542
2542
  // The secrecy was buying almost nothing anyway. Reading this body needs a
@@ -2547,7 +2547,7 @@ const server = createServer(async (req, res) => {
2547
2547
  // answers on the port, and still cannot see a word of this.
2548
2548
  //
2549
2549
  // So: pages explain, everything else stays flat. No channel names, no
2550
- // account, no token -- nothing about WHOSE viewer this is, only that a
2550
+ // account, no token -- nothing about WHOSE dashboard this is, only that a
2551
2551
  // link is needed and where the link comes from.
2552
2552
  if (req.method === 'GET' && isPage(url.pathname)) {
2553
2553
  // no-store because this is served at the same address the real page will
@@ -2828,7 +2828,7 @@ const server = createServer(async (req, res) => {
2828
2828
  res.writeHead(200, { 'Content-Type': 'application/json' });
2829
2829
  res.end(
2830
2830
  JSON.stringify({
2831
- // Where THIS viewer is running, so "connect this project" has an
2831
+ // Where THIS dashboard is running, so "connect this project" has an
2832
2832
  // obvious default and nobody has to type a path they can see.
2833
2833
  here: repoRoot(process.cwd()),
2834
2834
  channel: roster.error ? null : roster.channel ?? null,
@@ -2859,7 +2859,7 @@ const server = createServer(async (req, res) => {
2859
2859
  // without anybody choosing that.
2860
2860
  repo: i.repo,
2861
2861
  shared: !i.repo,
2862
- // Whether it is in the channel this viewer is looking at. An
2862
+ // Whether it is in the channel this dashboard is looking at. An
2863
2863
  // identity from another account is not a fault; it is the normal
2864
2864
  // state of a machine that works on two things.
2865
2865
  inThisChannel: i.key ? inChannel.has(i.key) : false,
@@ -2873,7 +2873,7 @@ const server = createServer(async (req, res) => {
2873
2873
  //
2874
2874
  // Reading goes through the agent's OWN device credential, because that is the
2875
2875
  // only route there is: an API key can grant but cannot list somebody's
2876
- // grants. That asymmetry is deliberate upstream and fine here -- the viewer
2876
+ // grants. That asymmetry is deliberate upstream and fine here -- the dashboard
2877
2877
  // only ever shows identities whose credential is on this machine.
2878
2878
  if (url.pathname === '/secrets') {
2879
2879
  const who = identities.find((i) => i.key === url.searchParams.get('as') && i.credential);
@@ -3284,7 +3284,7 @@ const server = createServer(async (req, res) => {
3284
3284
  // the RECIPIENT: the API endpoint is /v1/devices/me/... and an API key
3285
3285
  // cannot vote on somebody's behalf. So this only works for a message
3286
3286
  // addressed to an identity whose credential is on this machine, which is
3287
- // exactly the fleet this viewer is for.
3287
+ // exactly the fleet this dashboard is for.
3288
3288
  if (url.pathname === '/respond' && req.method === 'POST') {
3289
3289
  let payload;
3290
3290
  try {
@@ -3302,7 +3302,7 @@ const server = createServer(async (req, res) => {
3302
3302
  JSON.stringify({
3303
3303
  error:
3304
3304
  `No credential on this machine for ${payload.as ?? 'that recipient'}, so this ` +
3305
- 'viewer cannot answer as them. Whoever holds it has to answer.',
3305
+ 'dashboard cannot answer as them. Whoever holds it has to answer.',
3306
3306
  }),
3307
3307
  );
3308
3308
  return;
@@ -3329,7 +3329,7 @@ const server = createServer(async (req, res) => {
3329
3329
  return;
3330
3330
  }
3331
3331
 
3332
- // SENDING, as the channel rather than as an agent. A human at this viewer is
3332
+ // SENDING, as the channel rather than as an agent. A human at this dashboard is
3333
3333
  // not one of the agents, and signing their message with one of the machine's
3334
3334
  // identities would put words in that agent's mouth in a log others read.
3335
3335
  if (url.pathname === '/send' && req.method === 'POST') {
@@ -3413,7 +3413,7 @@ const server = createServer(async (req, res) => {
3413
3413
  for (const m of fresh) byId.set(m.id, m);
3414
3414
 
3415
3415
  body.messages = writeHistory(on.ref ?? channelRef, [...byId.values()], settings().history_days);
3416
- // Every channel this viewer can open, so the page can offer them. The
3416
+ // Every channel this dashboard can open, so the page can offer them. The
3417
3417
  // key never leaves this process -- only the ref and the name.
3418
3418
 
3419
3419
  // What this call actually cost, so the page can say so rather than
@@ -3430,7 +3430,7 @@ const server = createServer(async (req, res) => {
3430
3430
  * read is the moment you most need the list, because the other one in it
3431
3431
  * probably works and switching is the fix.
3432
3432
  *
3433
- * Found by pointing the viewer at two accounts with no credit. Reading a
3433
+ * Found by pointing the dashboard at two accounts with no credit. Reading a
3434
3434
  * log costs a delivery, both polls came back 402, and a page that knew
3435
3435
  * about two channels rendered neither.
3436
3436
  *
@@ -3488,7 +3488,7 @@ const server = createServer(async (req, res) => {
3488
3488
  // The SELECTED channel's roster, not the default one's. Without this the
3489
3489
  // conversation switched and the list of who is in it did not, so the rail
3490
3490
  // named agents that had never been in the channel on screen -- which reads
3491
- // as the directory being wrong rather than as the viewer asking the wrong
3491
+ // as the directory being wrong rather than as the dashboard asking the wrong
3492
3492
  // one.
3493
3493
  const body = await upstream('/v1/agents', channelFor(url.searchParams.get('channel')).key);
3494
3494
 
@@ -3500,7 +3500,7 @@ const server = createServer(async (req, res) => {
3500
3500
 
3501
3501
  // /a/<recipient key or handle> -- one agent's conversation. The whole point
3502
3502
  // of a path rather than a port: several agents run at once, and a port each
3503
- // means a port to remember each and a viewer to start each.
3503
+ // means a port to remember each and a dashboard to start each.
3504
3504
  if (url.pathname === '/' || url.pathname.startsWith('/a/')) {
3505
3505
  res.writeHead(200, { 'Content-Type': 'text/html; charset=utf-8', ...grant });
3506
3506
  res.end(PAGE);
@@ -3604,7 +3604,7 @@ server.listen(port, '127.0.0.1', () => {
3604
3604
  console.error(`Watching ${baseUrl} on http://127.0.0.1:${bound}/?t=${token}`);
3605
3605
  console.error('Loopback only, and the link carries a one-run token: it can now send and answer.');
3606
3606
  console.error('The key stays in this process; the page only ever talks to this server.');
3607
- // After the link is printed, not before: the viewer should be usable the
3607
+ // After the link is printed, not before: the dashboard should be usable the
3608
3608
  // moment it says it is, and the names arrive a beat later.
3609
3609
  nameChannels().catch(() => {
3610
3610
  /* the rail falls back to refs, which is a label and not a failure */
package/scripts/state.mjs CHANGED
@@ -2,13 +2,13 @@
2
2
  * Where an agent's identity lives on this machine. One copy, imported.
3
3
  *
4
4
  * WHY THIS FILE EXISTS. This logic was written out three times -- in the
5
- * watcher, in the hooks, and in the viewer -- each with a comment promising it
5
+ * watcher, in the hooks, and in the dashboard -- each with a comment promising it
6
6
  * matched the others. They drifted, and every drift failed the same silent
7
7
  * way: a credential is on disk, the reader looks somewhere else, and the
8
8
  * result is indistinguishable from a channel where nothing is happening.
9
9
  *
10
10
  * It happened twice in one evening. A fix went into two of the three copies
11
- * and nearly shipped; the viewer's copy turned out to have quietly fallen
11
+ * and nearly shipped; the dashboard's copy turned out to have quietly fallen
12
12
  * behind and could not see an identity the watcher was already polling. The
13
13
  * comments claiming the copies agreed were the reason nobody checked.
14
14
  *
@@ -85,9 +85,9 @@ export function stateDirs(env = process.env) {
85
85
  *
86
86
  * THE HOME DIRECTORY IS SEARCHED FOR THESE TOO, and it was not. This loop
87
87
  * used to run over `dirs` before ~/.agentsignal had been added to it, so
88
- * `~/.agentsignal/agents/<slug>/agent.json` was written by the viewer and
88
+ * `~/.agentsignal/agents/<slug>/agent.json` was written by the dashboard and
89
89
  * read by nobody: writeDir() falls back to ~/.agentsignal whenever
90
- * CLAUDE_PLUGIN_DATA is unset, which is every viewer started from a plain
90
+ * CLAUDE_PLUGIN_DATA is unset, which is every dashboard started from a plain
91
91
  * shell, and that is exactly where /connect puts a new agent.
92
92
  *
93
93
  * The failure was silent in the worst way. Registration succeeded, the
@@ -146,7 +146,7 @@ export function identities(baseUrl, env = process.env) {
146
146
  // The agent's own CHANNEL key, when a person connected it from the
147
147
  // dashboard, and where it lives when it lives in the keychain rather
148
148
  // than here. Carried because a device credential reads one inbox and
149
- // the viewer needs the whole channel -- without these two the key is
149
+ // the dashboard needs the whole channel -- without these two the key is
150
150
  // written to disk and read by nothing, which is the failure this
151
151
  // codebase keeps meeting. Both null for an agent that joined before
152
152
  // web joins carried a key, which is a tier and not a fault.
@@ -230,7 +230,7 @@ export function repoSlug(root) {
230
230
  *
231
231
  * SYNCHRONOUS, which is the whole reason this sits beside the async resolver
232
232
  * in packages/mcp/src/secrets.ts rather than reusing it. Every caller here
233
- * runs before anything is serving -- the viewer choosing a key, the daemon
233
+ * runs before anything is serving -- the dashboard choosing a key, the daemon
234
234
  * deciding what it may poll -- where an await does not merely wait, it
235
235
  * reorders startup. `security` answers in single-digit milliseconds against an
236
236
  * unlocked keychain; a locked one prompts, which is the right thing to do to