@phnx-labs/agents-cli 1.22.107 → 1.22.110

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.
Files changed (95) hide show
  1. package/CHANGELOG.md +96 -0
  2. package/dist/bootstrap.js +3 -2
  3. package/dist/commands/browser-sessions-picker.js +2 -1
  4. package/dist/commands/computer-sessions-picker.js +2 -1
  5. package/dist/commands/cost.js +2 -1
  6. package/dist/commands/daemon.js +45 -40
  7. package/dist/commands/doctor.js +7 -0
  8. package/dist/commands/exec.js +4 -2
  9. package/dist/commands/fork.js +6 -4
  10. package/dist/commands/logs.js +2 -1
  11. package/dist/commands/monitors.js +15 -2
  12. package/dist/commands/routines.js +46 -6
  13. package/dist/commands/sessions-backfill.d.ts +30 -0
  14. package/dist/commands/sessions-backfill.js +72 -0
  15. package/dist/commands/sessions-backup-setup.d.ts +12 -0
  16. package/dist/commands/sessions-backup-setup.js +65 -0
  17. package/dist/commands/sessions-inject.d.ts +3 -3
  18. package/dist/commands/sessions-inject.js +5 -11
  19. package/dist/commands/sessions-picker.js +10 -7
  20. package/dist/commands/sessions-resume.d.ts +17 -2
  21. package/dist/commands/sessions-resume.js +88 -12
  22. package/dist/commands/sessions.js +46 -24
  23. package/dist/commands/share.js +15 -2
  24. package/dist/commands/sync.js +30 -1
  25. package/dist/commands/utils.d.ts +9 -0
  26. package/dist/commands/utils.js +18 -0
  27. package/dist/commands/watchdog.js +4 -1
  28. package/dist/lib/accounting/rotate.d.ts +9 -0
  29. package/dist/lib/accounting/rotate.js +17 -1
  30. package/dist/lib/accounting/usage-sync.d.ts +22 -0
  31. package/dist/lib/accounting/usage-sync.js +70 -6
  32. package/dist/lib/accounting/usage.d.ts +32 -0
  33. package/dist/lib/accounting/usage.js +37 -2
  34. package/dist/lib/accounts/slots.js +7 -0
  35. package/dist/lib/agent-spec/agents.d.ts +60 -8
  36. package/dist/lib/agent-spec/agents.js +118 -45
  37. package/dist/lib/auth-health.js +19 -0
  38. package/dist/lib/claude-statusline.js +5 -0
  39. package/dist/lib/computer/sessions-list.js +2 -1
  40. package/dist/lib/daemon/auth-sync-service.d.ts +2 -0
  41. package/dist/lib/daemon/auth-sync-service.js +2 -2
  42. package/dist/lib/daemon/daemon.d.ts +34 -1
  43. package/dist/lib/daemon/daemon.js +70 -9
  44. package/dist/lib/daemon/leaked-daemons.d.ts +60 -0
  45. package/dist/lib/daemon/leaked-daemons.js +180 -0
  46. package/dist/lib/daemon/session-title-service.d.ts +41 -0
  47. package/dist/lib/daemon/session-title-service.js +76 -0
  48. package/dist/lib/daemon/usage-sync-service.d.ts +9 -1
  49. package/dist/lib/daemon/usage-sync-service.js +10 -2
  50. package/dist/lib/daemon-services.d.ts +1 -1
  51. package/dist/lib/daemon-services.js +5 -0
  52. package/dist/lib/daemon-ticks.js +6 -0
  53. package/dist/lib/devices/doctor-findings.d.ts +7 -1
  54. package/dist/lib/devices/doctor-findings.js +32 -1
  55. package/dist/lib/fleet-shared-state.d.ts +2 -0
  56. package/dist/lib/hooks/install.js +78 -64
  57. package/dist/lib/mailbox-target.js +2 -1
  58. package/dist/lib/models.js +76 -5
  59. package/dist/lib/session/active.d.ts +93 -16
  60. package/dist/lib/session/active.js +80 -14
  61. package/dist/lib/session/cloud.js +3 -1
  62. package/dist/lib/session/db.d.ts +44 -1
  63. package/dist/lib/session/db.js +122 -17
  64. package/dist/lib/session/fork.d.ts +10 -2
  65. package/dist/lib/session/fork.js +11 -2
  66. package/dist/lib/session/live-metadata.js +6 -0
  67. package/dist/lib/session/mirror.d.ts +3 -2
  68. package/dist/lib/session/mirror.js +5 -2
  69. package/dist/lib/session/parse.d.ts +1 -0
  70. package/dist/lib/session/parse.js +136 -2
  71. package/dist/lib/session/recovery.d.ts +9 -1
  72. package/dist/lib/session/recovery.js +14 -4
  73. package/dist/lib/session/remote/watch.js +18 -8
  74. package/dist/lib/session/title.d.ts +256 -0
  75. package/dist/lib/session/title.js +312 -0
  76. package/dist/lib/session/tool-calls.d.ts +1 -1
  77. package/dist/lib/session/tool-calls.js +40 -5
  78. package/dist/lib/session/tool-index.d.ts +5 -0
  79. package/dist/lib/session/tool-index.js +1 -0
  80. package/dist/lib/session/types.d.ts +11 -0
  81. package/dist/lib/share/backend.d.ts +23 -0
  82. package/dist/lib/share/backend.js +24 -0
  83. package/dist/lib/share/provision.d.ts +12 -0
  84. package/dist/lib/share/provision.js +30 -0
  85. package/dist/lib/share/worker-template.js +273 -0
  86. package/dist/lib/startup/root-command.d.ts +2 -0
  87. package/dist/lib/startup/root-command.js +22 -0
  88. package/dist/lib/terminal/engine.js +13 -1
  89. package/dist/lib/traces/sync.js +1 -0
  90. package/dist/lib/usage-refresh.d.ts +58 -7
  91. package/dist/lib/usage-refresh.js +145 -23
  92. package/dist/lib/watchdog/runner.d.ts +3 -0
  93. package/dist/lib/watchdog/runner.js +1 -0
  94. package/dist/session-tracker/dist/install-hook.js +4 -4
  95. package/package.json +1 -1
@@ -0,0 +1,312 @@
1
+ /**
2
+ * Daemon-generated session TITLES (PHNX-3797).
3
+ *
4
+ * The headline on a session row answers "what is this session about?", so it has
5
+ * to be anchored in what the USER asked for. It used to be the agent's latest
6
+ * transcript line — verbose, rolling, and unrecognizable to the person who
7
+ * started the run. The fix is a two-rung answer:
8
+ *
9
+ * 1. INSTANT, free: the user's own first message (already in the index) is the
10
+ * honest fallback shown from the moment the session appears.
11
+ * 2. UPGRADED, once: this module asks a CHEAP model (via a swappable
12
+ * {@link SessionTitleProvider}; the default {@link CloudSessionTitleProvider}
13
+ * uses the `cheap` tier — haiku on Claude — through the same
14
+ * `agents run --model <tier>` resolution every other call uses) for a short
15
+ * ACTION + OBJECT headline of what the session is doing, and persists it in
16
+ * the session index.
17
+ *
18
+ * Generation happens ONCE per session, in the daemon, and is keyed by
19
+ * {@link sessionTitleSourceKey} — a hash of the user text the title was derived
20
+ * from — so a sweep that sees a stored key matching the row's current text skips
21
+ * it (the cache hit) and regenerates only when that first user message actually
22
+ * changed, or when an operator asks explicitly
23
+ * (`agents sessions backfill titles --refresh`). There is no per-tick model call
24
+ * and no per-client generation: the daemon writes one value, every consumer
25
+ * (local list, picker, `sessions watch --json`, the fleet mirror, AGI EXT) reads
26
+ * it off the row.
27
+ *
28
+ * Everything here except {@link runSessionTitleTick} is pure, and the tick takes
29
+ * an injectable runner seam, so the whole path is testable without spawning a
30
+ * harness.
31
+ */
32
+ import { createHash } from 'node:crypto';
33
+ /**
34
+ * The phrase every generated-title prompt carries. Load-bearing twice over:
35
+ * `traces/sync.ts` classifies a session whose topic matches it as internal
36
+ * `utility` plumbing rather than agent work, and {@link isSessionTitlePrompt}
37
+ * uses it to keep the titler from titling its OWN spawned sessions — which
38
+ * would otherwise be a runaway loop, since each generation creates one more
39
+ * untitled session.
40
+ *
41
+ * It is a STABLE sentinel, not the human-readable instruction: keep it exact and
42
+ * keep `traces/sync.ts`'s matching regex in sync when it changes. The surrounding
43
+ * prompt (see {@link renderSessionTitlePrompt}) asks for an action+object headline
44
+ * of 4–8 words; the sentinel deliberately carries no word budget so the two never
45
+ * drift.
46
+ */
47
+ export const SESSION_TITLE_PROMPT_MARKER = 'Generate a concise session headline';
48
+ /** Hard ceiling on a stored title; the prompt asks for far less. */
49
+ export const SESSION_TITLE_MAX_CHARS = 60;
50
+ /** Hard ceiling on words kept from a model reply that ignored the word budget. */
51
+ export const SESSION_TITLE_MAX_WORDS = 8;
52
+ /** How much user text the prompt carries — enough to be specific, bounded for cost. */
53
+ export const SESSION_TITLE_INPUT_MAX_CHARS = 2000;
54
+ /** How many sessions one periodic sweep may generate for. */
55
+ export const SESSION_TITLE_MAX_PER_TICK = 2;
56
+ /** How many recent rows a sweep inspects before picking that batch. */
57
+ export const SESSION_TITLE_CANDIDATE_SCAN = 200;
58
+ /** Only sessions active within this window are titled — older rows are not shown. */
59
+ export const SESSION_TITLE_MAX_AGE_MS = 14 * 24 * 60 * 60_000;
60
+ /** Per-generation subprocess ceiling. A title is not worth waiting on. */
61
+ export const SESSION_TITLE_TIMEOUT_MS = 45_000;
62
+ /** The harnesses the titler will run as, best first. The first installed one wins. */
63
+ export const SESSION_TITLE_AGENTS = ['claude', 'codex', 'grok', 'kimi', 'opencode'];
64
+ /** The user text itself — the ONLY thing the source key is computed over. */
65
+ export function sessionTitleSourceText(input) {
66
+ const raw = (input.firstUserMessage || input.topic || '').replace(/\s+/g, ' ').trim();
67
+ return raw.length > SESSION_TITLE_INPUT_MAX_CHARS ? raw.slice(0, SESSION_TITLE_INPUT_MAX_CHARS) : raw;
68
+ }
69
+ /**
70
+ * Stable identity of the text a title was generated from. Storing this beside
71
+ * the title is what makes "already titled" distinguishable from "the user's
72
+ * first message changed" without keeping a second copy of that text.
73
+ * Empty text yields `null`: there is nothing to title.
74
+ */
75
+ export function sessionTitleSourceKey(input) {
76
+ const text = sessionTitleSourceText(input);
77
+ if (!text)
78
+ return null;
79
+ return createHash('sha256').update(text).digest('hex').slice(0, 16);
80
+ }
81
+ /**
82
+ * True when this text is one of the titler's OWN prompts. The titler runs a real
83
+ * harness, which writes a real transcript, which lands in the index as another
84
+ * untitled session — so without this guard every title generated would create
85
+ * work for the next sweep, forever.
86
+ */
87
+ export function isSessionTitlePrompt(...values) {
88
+ return values.some((v) => typeof v === 'string' && v.includes(SESSION_TITLE_PROMPT_MARKER));
89
+ }
90
+ /**
91
+ * The one-shot prompt handed to the cheap model.
92
+ *
93
+ * It asks for a descriptive ACTION + OBJECT headline (a verb phrase naming the
94
+ * concrete task, "Triage the AGI board", "Rename browser profile — default
95
+ * confusion"), not a single terse noun ("Triage") and not a full sentence — the
96
+ * headline slot has to tell the person who started the run what the session is
97
+ * doing at a glance. The word budget is soft in the prompt and hard-enforced by
98
+ * {@link sanitizeGeneratedTitle}'s {@link SESSION_TITLE_MAX_WORDS} ceiling.
99
+ */
100
+ export function renderSessionTitlePrompt(input) {
101
+ const text = sessionTitleSourceText(input);
102
+ const context = [
103
+ input.project ? `Repository: ${input.project}` : null,
104
+ input.ticketId ? `Ticket: ${input.ticketId}` : null,
105
+ input.gitBranch ? `Branch: ${input.gitBranch}` : null,
106
+ ].filter(Boolean);
107
+ return [
108
+ `${SESSION_TITLE_PROMPT_MARKER} naming what this coding session is working on.`,
109
+ 'Write an ACTION + OBJECT headline of 4 to 8 words — a verb phrase naming the concrete task,',
110
+ 'e.g. "Triage the AGI board" or "Rename browser profile — default confusion".',
111
+ 'NOT a single noun, NOT a full sentence. Name the concrete feature, component, or fix —',
112
+ 'not the person, not the pleasantries.',
113
+ 'Respond IMMEDIATELY with only the headline. Do NOT investigate, do NOT read files, do NOT use any tools.',
114
+ 'No quotes, no trailing punctuation, no explanation.',
115
+ ...(context.length ? ['', ...context] : []),
116
+ '',
117
+ "The user's request:",
118
+ '---',
119
+ text,
120
+ '---',
121
+ ].join('\n');
122
+ }
123
+ /**
124
+ * Reduce a model reply to a storable title, or `undefined` when it produced
125
+ * nothing usable. Takes the first non-empty line (a chatty model puts the title
126
+ * first), strips wrapping quotes/backticks and trailing punctuation, and applies
127
+ * the word + character ceilings. A reply that is really a refusal or an
128
+ * explanation fails the ceilings and is dropped rather than shown.
129
+ */
130
+ export function sanitizeGeneratedTitle(raw) {
131
+ if (!raw)
132
+ return undefined;
133
+ const firstLine = raw.split('\n').map((l) => l.trim()).find((l) => l.length > 0);
134
+ if (!firstLine)
135
+ return undefined;
136
+ let title = firstLine
137
+ .replace(/^["'`*_\s]+/, '')
138
+ .replace(/["'`*_\s]+$/, '')
139
+ .replace(/[.!?,;:]+$/, '')
140
+ .replace(/\s+/g, ' ')
141
+ .trim();
142
+ if (!title)
143
+ return undefined;
144
+ // A model that ignored the budget gets truncated, not shown in full — a
145
+ // paragraph in the headline slot is the very failure this feature removes.
146
+ const words = title.split(' ');
147
+ if (words.length > SESSION_TITLE_MAX_WORDS)
148
+ title = words.slice(0, SESSION_TITLE_MAX_WORDS).join(' ');
149
+ if (title.length > SESSION_TITLE_MAX_CHARS)
150
+ title = title.slice(0, SESSION_TITLE_MAX_CHARS).trimEnd();
151
+ // Never store the prompt back as a title (a harness that echoed its input).
152
+ if (isSessionTitlePrompt(title))
153
+ return undefined;
154
+ return title || undefined;
155
+ }
156
+ /**
157
+ * The headline for an INDEXED session row, on the same ladder the live path uses
158
+ * (`deriveSessionRecap`, active.ts): `/rename` label → daemon-generated title →
159
+ * first-prompt topic. Every `agents sessions` surface that shows "what this
160
+ * session is" reads it from here, so the CLI and the watch stream can never
161
+ * disagree about a session's name (PHNX-3797).
162
+ *
163
+ * A caller whose row type does not carry `generatedTitle` fails to compile
164
+ * (see {@link CarriesTitleRung}) — fix the projection to carry the field rather
165
+ * than casting past this.
166
+ */
167
+ export function sessionHeadline(row) {
168
+ return row.label || row.generatedTitle || row.topic || undefined;
169
+ }
170
+ /** The harness the titler runs as on this box, or null when none is installed. */
171
+ export async function resolveSessionTitleAgent() {
172
+ // Lazy import: this module is also imported by render paths that must not pull
173
+ // the installation store (and its filesystem probes) into a hot listing.
174
+ const { listInstalledVersions } = await import('../installations/store.js');
175
+ for (const agent of SESSION_TITLE_AGENTS) {
176
+ try {
177
+ if (listInstalledVersions(agent).length > 0)
178
+ return agent;
179
+ }
180
+ catch {
181
+ // An unreadable store for one agent must not hide the others.
182
+ }
183
+ }
184
+ return null;
185
+ }
186
+ /**
187
+ * The real runner: ONE `agents run <agent> --mode plan --model cheap <prompt>`
188
+ * subprocess. `--mode plan` keeps it read-only (it must never touch a repo), and
189
+ * `--model cheap` goes through the same tier resolution every other run uses, so
190
+ * the box's own catalog picks haiku (or its per-harness equivalent) rather than
191
+ * this module hardcoding a model id that ages out.
192
+ */
193
+ export async function defaultSessionTitleRunner(prompt, signal) {
194
+ const [{ getAgentsInvocation }, { execFile }, { promisify }] = await Promise.all([
195
+ import('../daemon/daemon.js'),
196
+ import('node:child_process'),
197
+ import('node:util'),
198
+ ]);
199
+ const agent = await resolveSessionTitleAgent();
200
+ if (!agent)
201
+ throw new Error('no installed harness can generate a session title');
202
+ const inv = getAgentsInvocation(['run', agent, '--mode', 'plan', '--model', 'cheap', prompt]);
203
+ const { stdout } = await promisify(execFile)(inv.command, inv.args, {
204
+ encoding: 'utf8',
205
+ maxBuffer: 1024 * 1024,
206
+ timeout: SESSION_TITLE_TIMEOUT_MS,
207
+ signal,
208
+ });
209
+ return stdout;
210
+ }
211
+ /**
212
+ * The default, shipped provider: render the shared prompt and run it through a
213
+ * {@link SessionTitleRunner} — by default {@link defaultSessionTitleRunner}, the
214
+ * one cheap `agents run --model cheap` subprocess. Tests inject a fake runner (or
215
+ * a whole fake provider) to exercise the tick without spawning a harness.
216
+ */
217
+ export class CloudSessionTitleProvider {
218
+ runner;
219
+ name = 'cloud';
220
+ constructor(runner = defaultSessionTitleRunner) {
221
+ this.runner = runner;
222
+ }
223
+ generate(input, signal) {
224
+ return this.runner(renderSessionTitlePrompt(input), signal);
225
+ }
226
+ }
227
+ /** The provider used when a caller injects neither a `provider` nor a `run`. */
228
+ export function defaultSessionTitleProvider() {
229
+ return new CloudSessionTitleProvider();
230
+ }
231
+ /**
232
+ * Decide what one sweep should do, without touching the model. Pure over its
233
+ * inputs so the "generate once, then cache-hit" property is directly testable:
234
+ * a candidate is work only when its user text yields a key AND that key differs
235
+ * from the one stored beside its title (or `force`).
236
+ */
237
+ export function selectSessionsNeedingTitle(rows, opts = { limit: SESSION_TITLE_MAX_PER_TICK }) {
238
+ const pending = [];
239
+ let cached = 0;
240
+ for (const row of rows) {
241
+ // The titler's own runs are sessions too; titling them would spawn another.
242
+ if (isSessionTitlePrompt(row.firstUserMessage, row.topic))
243
+ continue;
244
+ const sourceKey = sessionTitleSourceKey(row);
245
+ if (!sourceKey)
246
+ continue;
247
+ if (!opts.force && row.generatedTitle && row.generatedTitleKey === sourceKey) {
248
+ cached++;
249
+ continue;
250
+ }
251
+ if (pending.length < opts.limit)
252
+ pending.push({ row, sourceKey });
253
+ }
254
+ return { pending, cached };
255
+ }
256
+ /**
257
+ * One titling sweep: pick the sessions whose headline is still the raw user
258
+ * message, generate a title for at most {@link SessionTitleTickOptions.limit} of
259
+ * them, and persist each. Best-effort by contract — a failed generation leaves
260
+ * the row untitled, which the ladder renders as the user's own first message.
261
+ *
262
+ * Throws only for an explicit {@link SessionTitleTickOptions.id} that matches no
263
+ * indexed session (a caller error); the periodic sweep never throws.
264
+ */
265
+ export async function runSessionTitleTick(options = {}) {
266
+ const limit = options.limit ?? SESSION_TITLE_MAX_PER_TICK;
267
+ const now = options.nowMs ?? Date.now();
268
+ const result = { scanned: 0, cached: 0, generated: 0, failed: 0, titles: [] };
269
+ const { querySessionTitleCandidates, setSessionGeneratedTitle } = await import('./db.js');
270
+ const rows = querySessionTitleCandidates(options.id ? 1 : SESSION_TITLE_CANDIDATE_SCAN, options.id
271
+ ? { id: options.id }
272
+ : { sinceMs: now - (options.maxAgeMs ?? SESSION_TITLE_MAX_AGE_MS) });
273
+ result.scanned = rows.length;
274
+ // An explicitly requested session that has no indexed row is a caller error,
275
+ // not a quiet sweep outcome: reporting "0 generated" would read as "already
276
+ // current" for an id that was simply mistyped or not scanned yet.
277
+ if (options.id && rows.length === 0) {
278
+ throw new Error(`no indexed session matches "${options.id}" on this machine — ` +
279
+ `a session is titled after it is indexed (the daemon indexes within seconds), ` +
280
+ `and a peer's sessions are titled on the box that owns them`);
281
+ }
282
+ const { pending, cached } = selectSessionsNeedingTitle(rows, { limit, force: options.force });
283
+ result.cached = cached;
284
+ if (pending.length === 0)
285
+ return result;
286
+ // The provider is the ONLY thing that varies by model host; everything above
287
+ // (candidate selection, the source-key cache) and below (sanitize, persist) is
288
+ // shared. A local backend is dropped in here, not woven through the tick.
289
+ const provider = options.provider ?? new CloudSessionTitleProvider(options.run);
290
+ for (const { row, sourceKey } of pending) {
291
+ if (options.signal?.aborted)
292
+ break;
293
+ let title;
294
+ try {
295
+ title = sanitizeGeneratedTitle(await provider.generate(row, options.signal));
296
+ }
297
+ catch {
298
+ // Harness unavailable, signed out, or over its deadline. Best-effort: the
299
+ // row keeps showing the user's own words, and the caller backs off.
300
+ title = undefined;
301
+ }
302
+ if (!title) {
303
+ result.failed++;
304
+ continue;
305
+ }
306
+ if (setSessionGeneratedTitle(row.id, title, sourceKey, now)) {
307
+ result.generated++;
308
+ result.titles.push({ id: row.id, title });
309
+ }
310
+ }
311
+ return result;
312
+ }
@@ -10,7 +10,7 @@ export declare const TOOL_CHANGED_MAX_CALLS = 10000;
10
10
  export declare const TOOL_INDEX_LIMIT_ORDINAL: number;
11
11
  export declare const TOOL_TEXT_PROCESSING_MAX_BYTES: number;
12
12
  export declare const TOOL_SHELL_PARSE_MAX_BYTES: number;
13
- export declare const TOOL_INDEX_VERSION = 8;
13
+ export declare const TOOL_INDEX_VERSION = 9;
14
14
  export type ToolCallOutcome = 'ok' | 'error' | 'unknown';
15
15
  export interface IndexedToolCall {
16
16
  ordinal: number;
@@ -12,10 +12,12 @@ export const TOOL_CHANGED_MAX_CALLS = 10_000;
12
12
  export const TOOL_INDEX_LIMIT_ORDINAL = Number.MAX_SAFE_INTEGER;
13
13
  export const TOOL_TEXT_PROCESSING_MAX_BYTES = 64 * 1024;
14
14
  export const TOOL_SHELL_PARSE_MAX_BYTES = 64 * 1024;
15
- // Bumped to 8 for the per-call end timestamp (PHNX-3437): the extractor now
16
- // records when a call's result arrived, so a re-index re-derives it for rows
17
- // stored by an older extractor.
18
- export const TOOL_INDEX_VERSION = 8;
15
+ // Bumped to 9 for Codex tool-outcome classification (PHNX-3761): Codex results
16
+ // arrive as a string or input_text[] rather than an object, so every one used to
17
+ // store as 'unknown' and error rates read 0%. A re-index re-derives outcome for
18
+ // rows stored by an older extractor.
19
+ // (8 added the per-call end timestamp, PHNX-3437.)
20
+ export const TOOL_INDEX_VERSION = 9;
19
21
  const BASE64_BLOCK = /(?:[A-Za-z0-9+/]{256,}={0,2})/g;
20
22
  const SECRET_FIELD = /(?:token|secret|password|authorization|cookie|api[_-]?key|private[_-]?key)$/i;
21
23
  const KNOWN_SECRET_VALUES = knownSecretValuesFromEnv();
@@ -258,9 +260,37 @@ function structuredString(source, name) {
258
260
  const value = structuredField(source, name);
259
261
  return typeof value === 'string' ? value : undefined;
260
262
  }
263
+ // Codex tool results arrive as a plain string or an array of {type:'input_text',
264
+ // text} blocks (never a plain object), so the text lives across the blocks and
265
+ // the only outcome signal is the leading "Script completed"/"Script failed"
266
+ // marker the exec harness prefixes onto the first block.
267
+ function textSegments(value) {
268
+ if (typeof value === 'string')
269
+ return [value];
270
+ if (Array.isArray(value)) {
271
+ return value
272
+ .map((block) => block && typeof block === 'object' && typeof block.text === 'string'
273
+ ? block.text
274
+ : '')
275
+ .filter((text) => text.length > 0);
276
+ }
277
+ return [];
278
+ }
279
+ function codexScriptOutcome(value) {
280
+ const lead = textSegments(value)[0]?.trimStart();
281
+ if (!lead)
282
+ return undefined;
283
+ if (lead.startsWith('Script failed'))
284
+ return 'error';
285
+ if (lead.startsWith('Script completed'))
286
+ return 'ok';
287
+ return undefined;
288
+ }
261
289
  function resultText(value) {
262
290
  if (typeof value === 'string')
263
291
  return value;
292
+ if (Array.isArray(value))
293
+ return textSegments(value).join('\n');
264
294
  const object = resultObject(value);
265
295
  if (object) {
266
296
  for (const key of ['output', 'stdout', 'content', 'text', 'message']) {
@@ -280,9 +310,14 @@ export function structuredToolResult(value) {
280
310
  const structured = resultObject(value);
281
311
  const success = structuredField(structured, 'success');
282
312
  const isError = success === false || structuredField(structured, 'is_error') === true;
313
+ const outcome = isError
314
+ ? 'error'
315
+ : success === true
316
+ ? 'ok'
317
+ : codexScriptOutcome(value) ?? 'unknown';
283
318
  return {
284
319
  text: resultText(value),
285
- outcome: isError ? 'error' : success === true ? 'ok' : 'unknown',
320
+ outcome,
286
321
  exitCode: structuredNumber(structured, 'exit_code'),
287
322
  statusCode: structuredNumber(structured, 'status_code'),
288
323
  errorCode: structuredString(structured, 'error_code'),
@@ -31,6 +31,11 @@ export interface ToolSessionEvidence {
31
31
  cwd?: string;
32
32
  topic?: string;
33
33
  label?: string;
34
+ /** The daemon-generated headline (PHNX-3797). Carried so a consumer of this
35
+ * envelope names a session the same way every other surface does; a
36
+ * projection that drops it silently degrades `sessionHeadline` to
37
+ * `label || topic`. */
38
+ generatedTitle?: string;
34
39
  filePath?: string;
35
40
  calls: ToolCallEvidence[];
36
41
  }
@@ -869,6 +869,7 @@ export function searchToolCalls(sessions, clauseSources, coverage, resultLimit =
869
869
  cwd: session.cwd,
870
870
  topic: session.topic,
871
871
  label: session.label,
872
+ generatedTitle: session.generatedTitle,
872
873
  calls: selected,
873
874
  });
874
875
  if (matched.length >= resultLimit)
@@ -412,6 +412,17 @@ export interface SessionMeta {
412
412
  * single-turn session; absent for rows indexed before this field shipped.
413
413
  */
414
414
  lastUserMessage?: string;
415
+ /**
416
+ * The daemon-generated session TITLE (PHNX-3797) — a short, descriptive
417
+ * ACTION + OBJECT headline ("Triage the AGI board") for what the session is
418
+ * doing, produced ONCE per session by the `session-title` daemon service from
419
+ * the user's own first message (never from the agent's latest turn) and
420
+ * persisted in the index. It is the second rung of
421
+ * the headline ladder (`deriveSessionRecap`): an explicit `/rename` {@link label}
422
+ * still wins, and a session the titler has not reached yet falls back to
423
+ * `firstUserMessage`/`topic`, never to the agent's last line.
424
+ */
425
+ generatedTitle?: string;
415
426
  /**
416
427
  * The session's human-readable name — one field, several sources with a plain
417
428
  * priority: an agent-generated title / Claude `/rename` wins; else the launch
@@ -89,6 +89,29 @@ export declare function phoenixIdBaseForDeploy(opts?: {
89
89
  baseUrl?: string;
90
90
  domain?: string;
91
91
  }): string | undefined;
92
+ /**
93
+ * Managed collaboration backend config to bind on a managed Worker deploy
94
+ * (PHNX-3835): `PRIX_ARTIFACT_COLLAB_BASE` (the Prix artifact-collaboration API
95
+ * base) and `ARTIFACT_COLLAB_SERVICE_TOKEN` (the service credential the Worker
96
+ * sends as `Authorization: Bearer`, NEVER surfaced to the browser). Both are read
97
+ * from the deploy environment.
98
+ *
99
+ * Deliberately dormant-by-default: the two secrets are applied ONLY when BOTH are
100
+ * present. Until the Prix API counterpart is live an operator leaves them unset,
101
+ * and the Worker fails `/__collab` closed (404) with the artifact page GET
102
+ * unaffected. A pure-BYO endpoint never gets them (it has no Phoenix identity).
103
+ * Unlike `phoenixIdBaseForDeploy` this never throws on absence — collaboration is
104
+ * additive, not required for a working managed deploy.
105
+ */
106
+ export declare function collabConfigForDeploy(opts?: {
107
+ managed?: boolean;
108
+ }, cfg?: {
109
+ baseUrl?: string;
110
+ domain?: string;
111
+ }): {
112
+ collabBase?: string;
113
+ collabServiceToken?: string;
114
+ };
92
115
  /**
93
116
  * Pick the principal via the shared selection policy. Signed-in AND no explicit
94
117
  * BYO override → managed; otherwise BYO.
@@ -84,6 +84,30 @@ export function phoenixIdBaseForDeploy(opts = {}, cfg) {
84
84
  }
85
85
  return base;
86
86
  }
87
+ /**
88
+ * Managed collaboration backend config to bind on a managed Worker deploy
89
+ * (PHNX-3835): `PRIX_ARTIFACT_COLLAB_BASE` (the Prix artifact-collaboration API
90
+ * base) and `ARTIFACT_COLLAB_SERVICE_TOKEN` (the service credential the Worker
91
+ * sends as `Authorization: Bearer`, NEVER surfaced to the browser). Both are read
92
+ * from the deploy environment.
93
+ *
94
+ * Deliberately dormant-by-default: the two secrets are applied ONLY when BOTH are
95
+ * present. Until the Prix API counterpart is live an operator leaves them unset,
96
+ * and the Worker fails `/__collab` closed (404) with the artifact page GET
97
+ * unaffected. A pure-BYO endpoint never gets them (it has no Phoenix identity).
98
+ * Unlike `phoenixIdBaseForDeploy` this never throws on absence — collaboration is
99
+ * additive, not required for a working managed deploy.
100
+ */
101
+ export function collabConfigForDeploy(opts = {}, cfg) {
102
+ const managed = opts.managed === true || (cfg != null && isManagedShareEndpoint(cfg));
103
+ if (!managed)
104
+ return {};
105
+ const base = (process.env.PRIX_ARTIFACT_COLLAB_BASE || '').replace(/\/+$/, '').trim();
106
+ const token = (process.env.ARTIFACT_COLLAB_SERVICE_TOKEN || '').trim();
107
+ if (!base || !token)
108
+ return {};
109
+ return { collabBase: base, collabServiceToken: token };
110
+ }
87
111
  /**
88
112
  * The share surface's BYO-override signals: an explicit `--byo`, a
89
113
  * caller-supplied static write token, or `AGENTS_SHARE_BACKEND=byo`. Detecting
@@ -64,6 +64,10 @@ export interface UpdateWorkerResult {
64
64
  }
65
65
  /** Cloudflare `secret_text` name the Worker reads as `env.PHOENIX_ID_BASE`. */
66
66
  export declare const WORKER_PHOENIX_ID_BASE_SECRET = "PHOENIX_ID_BASE";
67
+ /** Cloudflare `secret_text` name the Worker reads as `env.PRIX_ARTIFACT_COLLAB_BASE`. */
68
+ export declare const WORKER_COLLAB_BASE_SECRET = "PRIX_ARTIFACT_COLLAB_BASE";
69
+ /** Cloudflare `secret_text` name the Worker reads as `env.ARTIFACT_COLLAB_SERVICE_TOKEN`. */
70
+ export declare const WORKER_COLLAB_SERVICE_TOKEN_SECRET = "ARTIFACT_COLLAB_SERVICE_TOKEN";
67
71
  export type UpdateWorkerOpts = ProvisionOptions & {
68
72
  force?: boolean;
69
73
  /**
@@ -73,6 +77,14 @@ export type UpdateWorkerOpts = ProvisionOptions & {
73
77
  * fails loud — do not pass a blank to "skip".
74
78
  */
75
79
  phoenixIdBase?: string;
80
+ /**
81
+ * Managed collaboration backend (PHNX-3835). BOTH must be present to enable
82
+ * `/__collab`; pass neither to leave the surface disabled (the Worker fails it
83
+ * closed). The service token rides Cloudflare's Secrets API, never rendered
84
+ * HTML or browser JS. Applied like `phoenixIdBase` — only when defined.
85
+ */
86
+ collabBase?: string;
87
+ collabServiceToken?: string;
76
88
  };
77
89
  /**
78
90
  * Re-deploy the Worker script against an ALREADY-provisioned endpoint (same
@@ -107,6 +107,10 @@ export function hashWorkerScript(script) {
107
107
  }
108
108
  /** Cloudflare `secret_text` name the Worker reads as `env.PHOENIX_ID_BASE`. */
109
109
  export const WORKER_PHOENIX_ID_BASE_SECRET = 'PHOENIX_ID_BASE';
110
+ /** Cloudflare `secret_text` name the Worker reads as `env.PRIX_ARTIFACT_COLLAB_BASE`. */
111
+ export const WORKER_COLLAB_BASE_SECRET = 'PRIX_ARTIFACT_COLLAB_BASE';
112
+ /** Cloudflare `secret_text` name the Worker reads as `env.ARTIFACT_COLLAB_SERVICE_TOKEN`. */
113
+ export const WORKER_COLLAB_SERVICE_TOKEN_SECRET = 'ARTIFACT_COLLAB_SERVICE_TOKEN';
110
114
  /**
111
115
  * Re-deploy the Worker script against an ALREADY-provisioned endpoint (same
112
116
  * account/worker/bucket `deployWorker` was first called with) and idempotently
@@ -145,6 +149,7 @@ export async function updateWorker(apiToken, accountId, workerName, bucketName,
145
149
  if (opts.phoenixIdBase !== undefined) {
146
150
  await applyPhoenixIdBaseSecret(apiToken, accountId, workerName, opts.phoenixIdBase, opts);
147
151
  }
152
+ await applyCollabSecrets(apiToken, accountId, workerName, opts);
148
153
  return { templateHash, skipped: true };
149
154
  }
150
155
  await deployWorker(apiToken, accountId, workerName, worker, bucketName, opts);
@@ -168,8 +173,33 @@ export async function updateWorker(apiToken, accountId, workerName, bucketName,
168
173
  throw new Error(`Worker deployed but PHOENIX_ID_BASE failed to re-apply — Phoenix-bearer publishes will 401 until you re-run \`agents artifacts share update\`. (${detail})`);
169
174
  }
170
175
  }
176
+ // The script upload wiped bindings/secrets (see JSDoc above), so re-apply the
177
+ // collaboration secrets too — but only when the caller provided them. An
178
+ // ordinary deploy with collaboration still dormant passes neither and the
179
+ // Worker keeps failing `/__collab` closed.
180
+ try {
181
+ await applyCollabSecrets(apiToken, accountId, workerName, opts);
182
+ }
183
+ catch (e) {
184
+ const detail = e instanceof Error ? e.message : String(e);
185
+ throw new Error(`Worker deployed but the collaboration secrets failed to re-apply — \`/__collab\` stays disabled until you re-run \`agents artifacts share update\`. (${detail})`);
186
+ }
171
187
  return { templateHash, skipped: false };
172
188
  }
189
+ /**
190
+ * Bind the managed collaboration secrets (PHNX-3835) — both or neither. A no-op
191
+ * when the caller passed neither (collaboration dormant), so a plain managed
192
+ * deploy never touches them; fails loud on a half-config (one without the other).
193
+ */
194
+ async function applyCollabSecrets(apiToken, accountId, workerName, opts) {
195
+ if (opts.collabBase === undefined && opts.collabServiceToken === undefined)
196
+ return;
197
+ if (opts.collabBase === undefined || opts.collabServiceToken === undefined) {
198
+ throw new Error('Managed collaboration needs BOTH PRIX_ARTIFACT_COLLAB_BASE and ARTIFACT_COLLAB_SERVICE_TOKEN — set both or neither.');
199
+ }
200
+ await putWorkerSecret(apiToken, accountId, workerName, WORKER_COLLAB_BASE_SECRET, opts.collabBase, opts);
201
+ await putWorkerSecret(apiToken, accountId, workerName, WORKER_COLLAB_SERVICE_TOKEN_SECRET, opts.collabServiceToken, opts);
202
+ }
173
203
  /** Add/update a secret_text binding using Cloudflare's Workers Secrets API. */
174
204
  export async function putWorkerSecret(apiToken, accountId, workerName, name, text, opts = {}) {
175
205
  const request = opts.request ?? defaultCloudflareRequester;