tickmarkr 1.70.0 → 1.71.0

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.
@@ -1,9 +1,10 @@
1
- import { spawnSync } from "node:child_process";
2
1
  import { readdirSync, readFileSync, realpathSync, statSync } from "node:fs";
3
2
  import { homedir } from "node:os";
4
3
  import { join } from "node:path";
5
4
  import { probeVersion } from "./claude-code.js";
6
5
  import { parseWorkerResult } from "./prompt.js";
6
+ import { runInteractiveSeed } from "../run/interactive-seed.js";
7
+ import { sh } from "../run/git.js";
7
8
  import { channelsFromConfig, MODEL_ID_RE, shq, TokenUsageSchema } from "./types.js";
8
9
  // KIMI-03 → v1.58 T5: the "no harness-readable counter" block (research F-6, 2026-07-17) is
9
10
  // LIFTED for collectUsage — kimi 0.27.0 writes a wire journal per agent at
@@ -117,40 +118,18 @@ export function confirmKimiSeedBanner(banner, assignedModel) {
117
118
  }
118
119
  return { ok: true, sessionId: kimiBannerSessionId(banner) };
119
120
  }
120
- // Shared launch-then-seed surface (T6) + banner confirm helpers (T7). One definition so the
121
- // adapter property and runKimiInteractiveSeed cannot drift.
121
+ // Shared launch-then-seed surface (T6) + banner confirm (T7/T2). One definition so the adapter
122
+ // property and the daemon's generic runInteractiveSeed path cannot drift.
122
123
  const KIMI_SEED = {
123
124
  launch: (model) => `kimi -y -m ${shq(kimiAlias(model))}`,
124
125
  readinessMatch: "Send /help for help information.",
125
126
  seedLine: (promptFile) => `Read ${promptFile} and do exactly what it says.`,
127
+ confirmBanner: confirmKimiSeedBanner,
126
128
  };
127
- // v1.69 T7: kimi-local launch-then-seed that confirms the banner model and captures the session id
128
- // from the SAME pane text already present for the readiness match. No extra probe, no extra
129
- // dispatch — one read of the launch banner, then seed or fail closed.
129
+ // v1.69 T7 / v1.71 T2: thin wrapper over the daemon's generic dispatch path for kimi-only tests.
130
130
  export async function runKimiInteractiveSeed(opts) {
131
- await opts.driver.run(opts.slot, KIMI_SEED.launch(opts.assignment.model));
132
- const ready = await opts.driver.waitOutput(opts.slot, KIMI_SEED.readinessMatch, opts.taskTimeoutMinutes * 60_000);
133
- // Banner text already on the pane for the readiness match — the only text model/session use.
134
- const banner = await opts.driver.read(opts.slot, 1000);
135
- if (!ready) {
136
- return { output: banner, seedFailed: true, seedError: `readiness pattern not seen: ${KIMI_SEED.readinessMatch}` };
137
- }
138
- const confirm = confirmKimiSeedBanner(banner, opts.assignment.model);
139
- if (!confirm.ok) {
140
- return { output: banner, seedFailed: true, seedError: confirm.error };
141
- }
142
- const seedText = KIMI_SEED.seedLine(opts.promptFile);
143
- await opts.driver.run(opts.slot, seedText);
144
- let output = "";
145
- for (let attempt = 0; attempt < 5; attempt++) {
146
- await new Promise((r) => setTimeout(r, 200));
147
- output = await opts.driver.read(opts.slot, 1000);
148
- const bottom = output.trimEnd().split("\n").pop() ?? "";
149
- if (!bottom.includes(seedText)) {
150
- return { output, seedFailed: false, sessionId: confirm.sessionId };
151
- }
152
- }
153
- return { output, seedFailed: true, seedError: "seed line never left the input box", sessionId: confirm.sessionId };
131
+ const r = await runInteractiveSeed({ ...opts, adapter: kimi });
132
+ return { output: r.output, seedFailed: r.seedFailed, seedError: r.seedError, sessionId: r.sessionId };
154
133
  }
155
134
  export const kimi = {
156
135
  id: "kimi",
@@ -183,10 +162,8 @@ export const kimi = {
183
162
  // null → the daemon's print fallback (types.ts:101) keeps kimi workers visible without a TUI.
184
163
  interactiveCommand: () => null,
185
164
  // v1.69 T6/T7: launch the real TUI, wait for the cold-start readiness marker, then submit the
186
- // prompt as one user turn. Banner model/session confirmation for seed mode lives in
187
- // runKimiInteractiveSeed / confirmKimiSeedBanner (same pane text as the readiness match — no
188
- // extra probe). Live-probed 2026-07-22 (kimi 0.28.1): `kimi -y` opens the TUI with yolo active
189
- // and prints the model/session lines plus "Send /help for help information." before the input box.
165
+ // prompt as one user turn. Banner model/session confirmation runs on the daemon's generic
166
+ // runInteractiveSeed path via confirmBanner on KIMI_SEED (not a separate dispatch helper).
190
167
  interactiveSeed: KIMI_SEED,
191
168
  // v1.53 T3 resume — live-probed 2026-07-18: `-p` + `-S <id>` compose cleanly (no OBS-67-class
192
169
  // flag rejection) and the resumed session carries prior conversation state. `-S <id>` is the
@@ -298,7 +275,7 @@ export const kimi = {
298
275
  }
299
276
  },
300
277
  listModels: async () => {
301
- const r = spawnSync("kimi", ["provider", "list", "--json"], { encoding: "utf8", timeout: 15000 });
302
- return r.error || r.status !== 0 ? [] : parseKimiModels(r.stdout || "");
278
+ const r = await sh("kimi provider list --json", process.cwd(), 15000);
279
+ return r.code !== 0 ? [] : parseKimiModels(r.stdout || "");
303
280
  },
304
281
  };
@@ -3,6 +3,28 @@ import { type RoutingProfile } from "../route/profile.js";
3
3
  import { type AuthHealth, type BillingChannel, type WorkerAdapter } from "./types.js";
4
4
  export declare const CANDIDATE_CLI_CATALOG: readonly ["kimi", "gemini", "qwen", "aider", "goose", "amp", "droid", "auggie", "crush"];
5
5
  export declare const REGISTERED_ADAPTER_BINARIES: readonly ["claude", "codex", "cursor-agent", "opencode", "pi", "grok"];
6
+ export declare function resolveShellBinary(bin: string, cwd?: string): {
7
+ resolved?: string;
8
+ all: string[];
9
+ };
10
+ export declare function probeVersionShell(bin: string, cwd?: string): AuthHealth & {
11
+ binaryPaths?: string[];
12
+ };
13
+ export declare function binaryShadowWarnings(adapters: WorkerAdapter[], health: Record<string, AuthHealth>, cwd?: string, resolve?: (bin: string, cwd: string) => {
14
+ resolved?: string;
15
+ all: string[];
16
+ }): string[];
17
+ export declare function invalidConfiguredModels(cfg: TickmarkrConfig, adapterId: string, h: AuthHealth | undefined): string[];
18
+ export declare function modelAliasExclusions(cfg: TickmarkrConfig, adapters: WorkerAdapter[], health: Record<string, AuthHealth>): {
19
+ key: string;
20
+ adapter: string;
21
+ model: string;
22
+ }[];
23
+ export declare function modelAliasLine(excluded: {
24
+ key: string;
25
+ adapter: string;
26
+ model: string;
27
+ }[]): string;
6
28
  export type CandidateCliDetection = {
7
29
  binary: string;
8
30
  version: string | undefined;
@@ -19,7 +41,9 @@ export declare function allAdapters(opts?: {
19
41
  fakeScriptPath?: string;
20
42
  }): WorkerAdapter[];
21
43
  export declare function getAdapter(id: string, adapters: WorkerAdapter[]): WorkerAdapter;
22
- export declare function probeAll(adapters: WorkerAdapter[]): Promise<Record<string, AuthHealth>>;
44
+ export declare function probeAll(adapters: WorkerAdapter[], opts?: {
45
+ cwd?: string;
46
+ }): Promise<Record<string, AuthHealth>>;
23
47
  export type ProbeModelStatus = "ok" | "timeout" | "failed";
24
48
  export type ProbeModelProgress = (adapter: string, model: string, status: ProbeModelStatus, durationMs: number) => void;
25
49
  export interface AutoPreferDoc {
@@ -18,11 +18,89 @@ import { grok } from "./grok.js";
18
18
  import { kimi } from "./kimi.js";
19
19
  import { opencode } from "./opencode.js";
20
20
  import { pi } from "./pi.js";
21
- import { channelKey, modelAuthed, QUOTA_RE } from "./types.js";
21
+ import { sealHerdrEnv } from "../drivers/subprocess.js";
22
+ import { channelKey, modelAuthed, QUOTA_RE, shq } from "./types.js";
22
23
  // Agent-CLI binary names with no tickmarkr adapter — doctor sweeps these advisory-only (v1.48 T1).
23
24
  export const CANDIDATE_CLI_CATALOG = ["kimi", "gemini", "qwen", "aider", "goose", "amp", "droid", "auggie", "crush"];
24
25
  // Binaries probed by registered adapters — kept beside the catalog so a test can forbid overlap.
25
26
  export const REGISTERED_ADAPTER_BINARIES = ["claude", "codex", "cursor-agent", "opencode", "pi", "grok"];
27
+ // OBS-117: worker panes run commands through `bash -lc` (SubprocessDriver + sh()), not bare spawnSync(bin).
28
+ // Doctor must resolve adapter binaries the same way so a shadow install on PATH cannot disagree with dispatch.
29
+ function shellSpawnSync(cmd, cwd, timeoutMs = 10000) {
30
+ const r = spawnSync("bash", ["-lc", cmd], {
31
+ encoding: "utf8",
32
+ cwd,
33
+ timeout: timeoutMs,
34
+ env: sealHerdrEnv(process.env),
35
+ stdio: ["ignore", "pipe", "pipe"],
36
+ });
37
+ return { code: r.status ?? 1, stdout: (r.stdout || "").trim(), stderr: (r.stderr || "").trim() };
38
+ }
39
+ export function resolveShellBinary(bin, cwd = process.cwd()) {
40
+ const r = shellSpawnSync(`type -a ${shq(bin)} 2>/dev/null | sed -n 's/^.* is \\(.*\\)$/\\1/p'`, cwd);
41
+ const all = r.stdout.split("\n").map((s) => s.trim()).filter(Boolean);
42
+ return { resolved: all[0], all };
43
+ }
44
+ export function probeVersionShell(bin, cwd = process.cwd()) {
45
+ const { resolved, all } = resolveShellBinary(bin, cwd);
46
+ if (!resolved)
47
+ return { installed: false, authed: false, models: [] };
48
+ const r = shellSpawnSync(`${shq(resolved)} --version`, cwd);
49
+ if (r.code !== 0)
50
+ return { installed: false, authed: false, models: [], binaryPaths: all };
51
+ return {
52
+ installed: true,
53
+ authed: true,
54
+ version: (r.stdout || r.stderr).trim().split("\n")[0],
55
+ models: [],
56
+ note: "auth assumed; verified at dispatch (failover on auth/quota errors)",
57
+ binaryPaths: all,
58
+ };
59
+ }
60
+ export function binaryShadowWarnings(adapters, health, cwd = process.cwd(), resolve) {
61
+ const resolveFn = resolve ?? resolveShellBinary;
62
+ const out = [];
63
+ for (const a of adapters) {
64
+ const bin = a.hardcodedFlags?.binary;
65
+ if (!bin || !health[a.id]?.installed)
66
+ continue;
67
+ const { resolved, all } = resolveFn(bin, cwd);
68
+ if (all.length <= 1)
69
+ continue;
70
+ out.push(`binary shadow: ${a.id} resolves to ${resolved ?? bin} but ${all.length} installs on PATH — ${all.join(", ")}`);
71
+ }
72
+ return out;
73
+ }
74
+ function modelListValidated(h) {
75
+ return !!h?.modelsDetectedAt && Array.isArray(h.models);
76
+ }
77
+ export function invalidConfiguredModels(cfg, adapterId, h) {
78
+ if (!modelListValidated(h))
79
+ return [];
80
+ const listed = new Set(h.models);
81
+ const configured = Object.keys(cfg.tiers[adapterId]?.models ?? {});
82
+ // Fail open until the CLI list intersects at least one configured model — proves the list is authoritative
83
+ // for this adapter's tier namespace (MODEL-04: unrelated detected ids must not invalidate routing).
84
+ if (!configured.some((m) => listed.has(m)))
85
+ return [];
86
+ return configured.filter((m) => !listed.has(m));
87
+ }
88
+ export function modelAliasExclusions(cfg, adapters, health) {
89
+ const out = [];
90
+ for (const a of adapters) {
91
+ const h = health[a.id];
92
+ if (!h?.installed || !h?.authed)
93
+ continue;
94
+ for (const model of invalidConfiguredModels(cfg, a.id, h)) {
95
+ out.push({ key: channelKey({ adapter: a.id, model }), adapter: a.id, model });
96
+ }
97
+ }
98
+ return out;
99
+ }
100
+ export function modelAliasLine(excluded) {
101
+ const parts = excluded.map(({ key, adapter }) => `${key} (not in ${adapter}'s reported model list)`);
102
+ return `model alias: ${excluded.length} channel(s) invalid — ${parts.join(", ")}`;
103
+ }
26
104
  function candidateOnPath(bin, pathEnv) {
27
105
  const r = spawnSync("which", [bin], { encoding: "utf8", env: { ...process.env, PATH: pathEnv } });
28
106
  return r.status === 0 && r.stdout.trim().length > 0;
@@ -105,9 +183,20 @@ export function getAdapter(id, adapters) {
105
183
  throw new Error(`unknown adapter ${id} (have: ${adapters.map((a) => a.id).join(", ")})`);
106
184
  return a;
107
185
  }
108
- export async function probeAll(adapters) {
186
+ export async function probeAll(adapters, opts = {}) {
187
+ const cwd = opts.cwd ?? process.cwd();
109
188
  const out = {};
110
- await Promise.all(adapters.map(async (a) => { out[a.id] = await a.probe(); }));
189
+ await Promise.all(adapters.map(async (a) => {
190
+ let h = await a.probe();
191
+ const bin = a.hardcodedFlags?.binary;
192
+ if (bin) {
193
+ const shell = probeVersionShell(bin, cwd);
194
+ h = { ...h, installed: shell.installed, version: shell.version ?? h.version };
195
+ if (!shell.installed)
196
+ h = { ...h, installed: false, authed: false };
197
+ }
198
+ out[a.id] = h;
199
+ }));
111
200
  return out;
112
201
  }
113
202
  const MODEL_PROBE_PROMPT = "Reply with exactly OK and nothing else.";
@@ -317,10 +406,12 @@ export function discoverChannels(cfg, adapters, health) {
317
406
  .flatMap((a) => {
318
407
  const h = health[a.id];
319
408
  const s = h?.servable;
409
+ const invalid = new Set(invalidConfiguredModels(cfg, a.id, h));
320
410
  // T2 (2026-07-13): only a model doctor marked authed (modelAuth[model].authed===true) advertises a
321
411
  // channel — routing an unknown or 403 into dispatch is fail-closed. Operators with pre-v1.21 doctor.json
322
412
  // can opt into the prior unknown-is-routable behavior with routing.allowUnverifiedModels.
323
- return a.channels(cfg).filter((c) => modelAuthed(h, c.model, cfg.routing.allowUnverifiedModels) && (!s || s.includes(c.model)));
413
+ // OBS-117: configured aliases absent from the adapter CLI's own model list are not dispatchable.
414
+ return a.channels(cfg).filter((c) => !invalid.has(c.model) && modelAuthed(h, c.model, cfg.routing.allowUnverifiedModels) && (!s || s.includes(c.model)));
324
415
  });
325
416
  if (!cfg.routing.allow && !cfg.routing.deny)
326
417
  return base;
@@ -48,10 +48,18 @@ export interface WorkerResult {
48
48
  deviations: string[];
49
49
  raw: string;
50
50
  }
51
+ export type SeedBannerConfirmResult = {
52
+ ok: true;
53
+ sessionId?: string;
54
+ } | {
55
+ ok: false;
56
+ error: string;
57
+ };
51
58
  export interface InteractiveSeed {
52
59
  launch(model: string): string;
53
60
  readinessMatch: string;
54
61
  seedLine(promptFile: string): string;
62
+ confirmBanner?(banner: string, assignedModel: string): SeedBannerConfirmResult;
55
63
  }
56
64
  export interface ContextUsage {
57
65
  tokens: number;
@@ -1,12 +1,12 @@
1
1
  import { writeFileSync } from "node:fs";
2
2
  import { join } from "node:path";
3
- import { allAdapters, detectCandidateClis, flagDriftWarnings, probeAll, probeModels, readAutoPrefer, servableExclusions, servabilityLine, writeDoctor } from "../../adapters/registry.js";
3
+ import { allAdapters, binaryShadowWarnings, detectCandidateClis, flagDriftWarnings, modelAliasExclusions, modelAliasLine, probeAll, probeModels, readAutoPrefer, servableExclusions, servabilityLine, writeDoctor } from "../../adapters/registry.js";
4
4
  import { BANNER, dim, fail, kvRow, legend, ok, rule, statusRow, title } from "../../brand.js";
5
5
  import { tickmarkrDir, stateDirName } from "../../graph/graph.js";
6
6
  import { declaredModelWindow, hasWindowsConfig, modelLints, suggestOverlay, ttyVisual } from "../../adapters/model-lints.js";
7
7
  import { DEFAULT_CONFIG, loadConfig, overlayPreferShapes } from "../../config/config.js";
8
8
  import { HerdrDriver } from "../../drivers/herdr.js";
9
- import { disallowedBy, excludedChannels, exclusionLine, preferRanks } from "../../route/preference.js";
9
+ import { denyPreferCollisionLine, denyPreferCollisions, disallowedBy, excludedChannels, exclusionLine, preferRanks } from "../../route/preference.js";
10
10
  const visual = () => process.stdout.isTTY === true && process.env.NO_COLOR === undefined;
11
11
  const alignedStatusRow = (verdict, key, value) => ` ${statusRow(verdict, kvRow(key, value).slice(2))}`;
12
12
  const attentionRow = (text) => ` ${statusRow("warn", text)}`;
@@ -19,7 +19,7 @@ export async function doctor(_argv, cwd = process.cwd(), adapters = allAdapters(
19
19
  // otherwise prints nothing until the end — silence here reads as a hang (v1.33.1)
20
20
  console.error("probing installed agent CLIs — one short LLM call per configured model, may take a minute...");
21
21
  const probeProgressTTY = process.stderr.isTTY === true;
22
- const health = await probeAll(adapters);
22
+ const health = await probeAll(adapters, { cwd });
23
23
  // MODEL-02: detect models where the adapter exposes a list surface, BEFORE writing doctor.json (write once, below).
24
24
  // Fail OPEN — the inverse of gates' fail-closed: detection is advisory, so a broken list surface NEVER fails doctor.
25
25
  for (const a of adapters) {
@@ -81,6 +81,11 @@ export async function doctor(_argv, cwd = process.cwd(), adapters = allAdapters(
81
81
  const excluded = excludedChannels(cfg, adapters, health);
82
82
  if (excluded.length)
83
83
  rows.push(attentionRow(exclusionLine(excluded)));
84
+ rows.push(...denyPreferCollisions(cfg).map((c) => attentionRow(denyPreferCollisionLine(c))));
85
+ const aliasExcluded = modelAliasExclusions(cfg, adapters, health);
86
+ if (aliasExcluded.length)
87
+ rows.push(attentionRow(modelAliasLine(aliasExcluded)));
88
+ rows.push(...binaryShadowWarnings(adapters, health, cwd).map(attentionRow));
84
89
  // HYG-07(a): doctor just probed fresh (probeAll above), so servability attribution is current by construction.
85
90
  const servable = servableExclusions(cfg, adapters, health);
86
91
  if (servable.length)
@@ -2,6 +2,7 @@ import { loadConfig } from "../../config/config.js";
2
2
  import { pickDriver } from "../../drivers/index.js";
3
3
  import { formatSummary, runDaemon } from "../../run/daemon.js";
4
4
  import { formatJournalNarration } from "../../run/journal.js";
5
+ import { denyPreferCollisionLine, denyPreferCollisions } from "../../route/preference.js";
5
6
  const summaryGreen = (s) => s.failed.length === 0 && s.human.length === 0 && s.blocked.length === 0 && s.pending.length === 0
6
7
  && s.tipVerify !== "failed";
7
8
  export async function resume(argv, cwd = process.cwd()) {
@@ -12,7 +13,12 @@ export async function resume(argv, cwd = process.cwd()) {
12
13
  // Fable F2) — the daemon refuses a mismatched/unbound journal unless this is set, then journals a
13
14
  // graph-rehash event naming both hashes. Strip the flag before runId resolution so a bare id still wins.
14
15
  const graphChanged = argv.includes("--graph-changed");
15
- const s = await runDaemon(cwd, { runId, resume: true, graphChanged, driver: pickDriver(loadConfig(cwd)), narrate: (event) => console.log(formatJournalNarration(event)) });
16
+ const cfg = loadConfig(cwd);
17
+ const collisions = denyPreferCollisions(cfg);
18
+ if (collisions.length) {
19
+ throw new Error(collisions.map(denyPreferCollisionLine).join("; "));
20
+ }
21
+ const s = await runDaemon(cwd, { runId, resume: true, graphChanged, driver: pickDriver(cfg), narrate: (event) => console.log(formatJournalNarration(event)) });
16
22
  const out = `resumed ${s.runId} — ${formatSummary(s)}`;
17
23
  return { out, code: summaryGreen(s) ? 0 : 2 };
18
24
  }
@@ -2,7 +2,7 @@ import { BANNER, GLYPHS, dim, fail, legend, ok, rule, statusRow, title, warn } f
2
2
  import { blockedTasks, graphDefinitionHash, loadGraph } from "../../graph/graph.js";
3
3
  import { GATE_NAMES } from "../../graph/schema.js";
4
4
  import { foldActivity } from "../../run/activity.js";
5
- import { Journal, engagementComparable } from "../../run/journal.js";
5
+ import { Journal, engagementComparable, isQualityFailureParkKind, recordedTaskFailureKind, runHasEnded, } from "../../run/journal.js";
6
6
  // ponytail: fixed 2s refresh; promote to config.visibility.* only when an operator asks.
7
7
  const REFRESH_MS = 2000;
8
8
  const NOT_COMPARABLE_NOTICE = "graph recompiled since this run — task states not comparable; run `tickmarkr run` to execute";
@@ -178,15 +178,23 @@ const renderFrame = (cwd) => {
178
178
  const divider = unicode ? " · " : " / ";
179
179
  const width = process.stdout.columns ?? 120;
180
180
  const done = effective.tasks.filter((t) => t.status === "done").length;
181
+ const ended = comparable && runHasEnded(events);
181
182
  const cells = g.tasks.map((t) => {
182
183
  const st = replayed?.get(t.id) ?? t.status;
184
+ const failureKind = comparable ? recordedTaskFailureKind(events, t.id) : undefined;
185
+ const terminal = st === "failed" || st === "human";
186
+ // Unknown legacy task-failed events stay red. Typed availability noise is warn-tier only while
187
+ // the run is live; verified quality parks and every unresolved task in an ended run are red.
188
+ const redTier = terminal && (ended
189
+ || (failureKind !== undefined && isQualityFailureParkKind(failureKind))
190
+ || (st === "failed" && failureKind === undefined));
183
191
  const isStarved = starved.has(t.id);
184
192
  const phrase = isStarved ? undefined : activity.cells.get(t.id);
185
193
  const label = isStarved ? " starved" : phrase ? ` ${phrase}` : "";
186
194
  const channel = assignments.get(t.id) ?? "-";
187
195
  const ctx = contexts.get(t.id);
188
196
  const assignCol = ctx !== undefined ? `${channel}${divider}ctx ${ctx}` : channel;
189
- return { t, st, label, assignCol, isStarved, phrase, channel, ctx, states: comparable ? gateStates(t, events) : defaultGateStates(t) };
197
+ return { t, st, failureKind, redTier, label, assignCol, isStarved, phrase, channel, ctx, states: comparable ? gateStates(t, events) : defaultGateStates(t) };
190
198
  });
191
199
  if (!unicode) {
192
200
  // machine/CI surface — layout unchanged (pipes, greps, and the golden pins depend on it)
@@ -207,7 +215,7 @@ const renderFrame = (cwd) => {
207
215
  // rows (pad plain text FIRST, colorize after: ANSI has zero display width and would corrupt
208
216
  // padEnd math)
209
217
  const dot = dim(" · ");
210
- const anyFailed = cells.some((c) => c.st === "failed");
218
+ const anyFailed = cells.some((c) => c.redTier);
211
219
  const gaugeCells = 10;
212
220
  const fill = g.tasks.length ? Math.round((done / g.tasks.length) * gaugeCells) : 0;
213
221
  const gauge = (fill ? (anyFailed ? fail : ok)("█".repeat(fill)) : "") + (fill < gaugeCells ? dim("░".repeat(gaugeCells - fill)) : "");
@@ -232,17 +240,20 @@ const renderFrame = (cwd) => {
232
240
  // id, goal at full width, and only SHORT status words. Line 2 carries the machinery, dim and
233
241
  // aligned under the goal: gate chain, live activity phrase (or channel), ctx. Long channel names
234
242
  // and activity phrases live on line 2 only, so they can never squeeze the goal or wrap line 1.
235
- const taskVerdict = (st) => st === "done" ? "pass" : st === "failed" ? "fail" : st === "human" ? "warn" : "neutral";
243
+ const taskVerdict = (c) => c.st === "done" ? "pass" : c.redTier ? "fail" : c.st === "failed" || c.st === "human" ? "warn" : "neutral";
244
+ const statusWord = (c) => c.redTier ? "failed" : c.st === "failed" ? "warn" : String(c.st);
236
245
  const idW = Math.max(...cells.map((c) => c.t.id.length), 2);
237
246
  // plain-text status suffixes for width math only — starved / failed gate names / approval hint
238
247
  const suffixPlain = (c) => (c.isStarved ? " · starved" : "") + failedSuffix(c.states) + humanGateSuffix(c.t, c.st, c.states);
239
- const stW = Math.max(...cells.map((c) => String(c.st).length + suffixPlain(c).length));
248
+ const stW = Math.max(...cells.map((c) => statusWord(c).length + suffixPlain(c).length));
240
249
  const avail = Math.max(8, width - (5 + idW) - 2 - stW);
241
250
  const goals = cells.map((c) => shortGoal(c.t.goal, avail));
242
251
  const goalW = Math.max(8, ...goals.map((s) => s.length));
243
252
  const indent = " ".repeat(idW + 5); // line 2 starts under the goal column
244
- const rows = cells.map(({ t, st, states, isStarved, phrase, channel, ctx }, i) => {
245
- const stWord = st === "done" ? ok(String(st)) : st === "failed" ? fail(String(st)) : st === "human" ? warn(String(st)) : String(st);
253
+ const rows = cells.map((c, i) => {
254
+ const { t, st, failureKind, redTier, states, isStarved, phrase, channel, ctx } = c;
255
+ const word = statusWord(c);
256
+ const stWord = st === "done" ? ok(word) : redTier ? fail(word) : st === "failed" || st === "human" ? warn(word) : word;
246
257
  // a fail names its gate in words right here — the one moment gate identity is needed on a row
247
258
  const f = failedGates(states);
248
259
  const human = humanGateSuffix(t, st, states);
@@ -250,10 +261,11 @@ const renderFrame = (cwd) => {
250
261
  (isStarved ? dot + fail("starved") : "") +
251
262
  (f.length ? dot + fail(f.join(", ")) : "") +
252
263
  (human ? dot + warn("awaiting approval") : "");
253
- const line1 = ` ${statusRow(taskVerdict(st), `${t.id.padEnd(idW)} ${goals[i].padEnd(goalW)} ${statusCell}`)}`;
264
+ const line1 = ` ${statusRow(taskVerdict(c), `${t.id.padEnd(idW)} ${goals[i].padEnd(goalW)} ${statusCell}`)}`;
254
265
  // activity already names its channel for in-flight attempts — never repeat it
255
266
  const detail = [
256
267
  ...(phrase ? [phrase, ...(phrase.includes(channel) || channel === "-" ? [] : [channel])] : [channel]),
268
+ ...(failureKind && !phrase?.includes(failureKind) ? [failureKind] : []),
257
269
  ...(ctx !== undefined ? [`ctx ${ctx}`] : []),
258
270
  ].join(" · ");
259
271
  return [line1, `${indent}${gateChain(states, true)} ${dim(detail)}`, ""];
@@ -11,11 +11,14 @@ export declare class HerdrDriver implements ExecutorDriver {
11
11
  interactive: boolean;
12
12
  private groups;
13
13
  private groupSerial;
14
+ private deliverySerial;
14
15
  private ws;
15
16
  private callerPane;
16
17
  private watches;
17
18
  constructor(bin?: string, workersPerTab?: number);
18
19
  private serial;
20
+ private deliveryQueue;
21
+ private deliveryMatches;
19
22
  static available(): boolean;
20
23
  private herdr;
21
24
  private namedPaneId;
@@ -29,6 +32,8 @@ export declare class HerdrDriver implements ExecutorDriver {
29
32
  private glyphFor;
30
33
  private joinGroup;
31
34
  run(slot: Slot, cmd: string): Promise<void>;
35
+ private deliver;
36
+ private deliveryReadMatches;
32
37
  private waitOk;
33
38
  waitOutput(slot: Slot, pattern: string, timeoutMs: number, opts?: {
34
39
  regex?: boolean;
@@ -25,6 +25,8 @@ export class HerdrDriver {
25
25
  // grouped slot()/close() mutate shared group state across awaits — serialize them so two
26
26
  // concurrent first members can never both create the group tab (mergeSerial idiom, daemon.ts)
27
27
  groupSerial = Promise.resolve();
28
+ // OBS-119: concurrent deliveries contend on herdr's send path — one chain per driver (mergeSerial idiom)
29
+ deliverySerial = Promise.resolve();
28
30
  // VIS-10: the run's workspace id, captured once at construction (the daemon inherits it from the
29
31
  // operator's env before the driver is built). Required at slot() time, never in the constructor —
30
32
  // pickDriver and its unit test construct HerdrDriver without env, so slot() is the trust gate.
@@ -40,6 +42,18 @@ export class HerdrDriver {
40
42
  this.groupSerial = p.catch(() => undefined);
41
43
  return p;
42
44
  }
45
+ deliveryQueue(fn) {
46
+ const p = this.deliverySerial.then(fn, fn);
47
+ this.deliverySerial = p.catch(() => undefined);
48
+ return p;
49
+ }
50
+ // ponytail: narrow panes hard-wrap the input line — collapse whitespace before comparing.
51
+ deliveryMatches(transcript, cmd) {
52
+ const norm = (s) => s.replace(/\s+/g, "");
53
+ const hay = norm(transcript);
54
+ const needle = norm(cmd);
55
+ return needle.length > 0 && hay.includes(needle);
56
+ }
43
57
  static available() {
44
58
  return process.env.HERDR_ENV === "1";
45
59
  }
@@ -286,6 +300,9 @@ export class HerdrDriver {
286
300
  // cleared (C-u), and retyped, bounded; persistent corruption fails closed WITH the captured
287
301
  // transcript — the dispatch-time pincer the ledger asks for, not post-hoc `git:` archaeology.
288
302
  async run(slot, cmd) {
303
+ return this.deliveryQueue(() => this.deliver(slot, cmd));
304
+ }
305
+ async deliver(slot, cmd) {
289
306
  const pane = await this.paneId(slot);
290
307
  let transcript = "";
291
308
  for (let attempt = 0; attempt < DELIVERY_ATTEMPTS; attempt++) {
@@ -300,9 +317,7 @@ export class HerdrDriver {
300
317
  if (typed.code !== 0)
301
318
  throw new Error(`herdr pane send-text failed: ${typed.stderr || typed.stdout}`);
302
319
  const back = await this.herdr(`wait output ${shq(pane)} --match ${shq(cmd)} --timeout ${DELIVERY_VERIFY_TIMEOUT_MS}`, slot.cwd, DELIVERY_VERIFY_TIMEOUT_MS + 15_000);
303
- // ponytail: transcript containment (scrollback included); anchor to the prompt line if a
304
- // prior echo of the same command ever false-verifies. Dead pane → error envelope → unverified.
305
- if (this.waitOk(back.code, back.stdout)) {
320
+ if (this.waitOk(back.code, back.stdout) || await this.deliveryReadMatches(pane, cmd, slot.cwd)) {
306
321
  const enter = await this.herdr(`pane send-keys ${shq(pane)} Enter`, slot.cwd);
307
322
  if (enter.code !== 0)
308
323
  throw new Error(`herdr pane send-keys Enter failed: ${enter.stderr || enter.stdout}`);
@@ -313,6 +328,10 @@ export class HerdrDriver {
313
328
  }
314
329
  throw new Error(`herdr delivery corrupted after ${DELIVERY_ATTEMPTS} attempts — enter never pressed (OBS-85); pane transcript:\n${transcript}`);
315
330
  }
331
+ async deliveryReadMatches(pane, cmd, cwd) {
332
+ const read = await this.herdr(`pane read ${shq(pane)} --source recent-unwrapped --lines ${DELIVERY_READ_LINES}`, cwd);
333
+ return read.code === 0 && this.deliveryMatches(read.stdout, cmd);
334
+ }
316
335
  waitOk(code, stdout) {
317
336
  if (code !== 0 || !stdout.trim())
318
337
  return code === 0; // herdr's successful waits may be silent
@@ -25,3 +25,12 @@ export declare function disallowedBy(c: {
25
25
  adapter: string;
26
26
  model: string;
27
27
  }, routing: TickmarkrConfig["routing"]): Disallowed | null;
28
+ export interface DenyPreferCollision {
29
+ kind: "prefer" | "pin";
30
+ shape: string;
31
+ detail: string;
32
+ disallowed: Disallowed;
33
+ }
34
+ export declare function preferEntryDenied(p: string, cfg: TickmarkrConfig): Disallowed | null;
35
+ export declare function denyPreferCollisions(cfg: TickmarkrConfig): DenyPreferCollision[];
36
+ export declare function denyPreferCollisionLine({ kind, shape, detail, disallowed }: DenyPreferCollision): string;
@@ -1,4 +1,6 @@
1
1
  import { channelKey, channelsFromConfig } from "../adapters/types.js";
2
+ import { validateGraph } from "../graph/schema.js";
3
+ import { route, RoutingError } from "./router.js";
2
4
  const adapterIds = (adapters) => typeof adapters[0] === "string" ? adapters : adapters.map((a) => a.id);
3
5
  export function excludedChannels(cfg, adapters, health) {
4
6
  const { allow, deny } = cfg.routing;
@@ -47,3 +49,62 @@ export function disallowedBy(c, routing) {
47
49
  }
48
50
  return null;
49
51
  }
52
+ // Delegates to route()'s preflightPrefer — doctor/resume preflight reuses the router grammar,
53
+ // not a second colon/tier/model parser of the same config.
54
+ const PREFLIGHT_SHAPE = "chore";
55
+ const preflightTask = validateGraph({
56
+ version: 1, spec: { source: "prd", paths: ["p"], hash: "h" },
57
+ tasks: [{ id: "T0", title: "preflight", goal: "preflight", shape: PREFLIGHT_SHAPE, complexity: 1, acceptance: ["preflight"] }],
58
+ }).tasks[0];
59
+ const disallowedFromPreferError = (msg) => {
60
+ const m = msg.match(/prefer entry .+ is disallowed by routing\.(deny|allow) \(([^)]+)\)/);
61
+ return m ? { by: m[1], entry: m[2] } : null;
62
+ };
63
+ export function preferEntryDenied(p, cfg) {
64
+ if (!cfg.routing.allow && !cfg.routing.deny)
65
+ return null;
66
+ const probe = structuredClone(cfg);
67
+ probe.routing.map = { ...probe.routing.map, [PREFLIGHT_SHAPE]: { prefer: [p] } };
68
+ try {
69
+ route(preflightTask, probe, []);
70
+ return null;
71
+ }
72
+ catch (e) {
73
+ if (e instanceof RoutingError)
74
+ return disallowedFromPreferError(e.message);
75
+ throw e;
76
+ }
77
+ }
78
+ export function denyPreferCollisions(cfg) {
79
+ if (!cfg.routing.allow && !cfg.routing.deny)
80
+ return [];
81
+ const out = [];
82
+ for (const [shape, entry] of Object.entries(cfg.routing.map)) {
83
+ if (entry.pin) {
84
+ const d = disallowedBy({ adapter: entry.pin.via, model: entry.pin.model }, cfg.routing);
85
+ if (d) {
86
+ out.push({
87
+ kind: "pin",
88
+ shape,
89
+ detail: `${entry.pin.via}:${entry.pin.model}`,
90
+ disallowed: d,
91
+ });
92
+ }
93
+ }
94
+ const prefer = entry.prefer ?? [];
95
+ if (prefer.length && prefer.every((p) => preferEntryDenied(p, cfg) !== null)) {
96
+ out.push({
97
+ kind: "prefer",
98
+ shape,
99
+ detail: prefer.join(" > "),
100
+ disallowed: preferEntryDenied(prefer[0], cfg),
101
+ });
102
+ }
103
+ }
104
+ return out;
105
+ }
106
+ export function denyPreferCollisionLine({ kind, shape, detail, disallowed }) {
107
+ const surface = kind === "pin" ? `routing.map.${shape}.pin` : `routing.map.${shape}.prefer`;
108
+ const verb = kind === "pin" ? "is disallowed" : "fully disallowed";
109
+ return `deny∩prefer: ${surface} ${detail} ${verb} by routing.${disallowed.by} (${disallowed.entry}) — remove the ${disallowed.by} entry or adjust ${kind === "pin" ? "pin" : "prefer"}`;
110
+ }
@@ -41,4 +41,8 @@ export interface RunSummary {
41
41
  lastMergedTask?: string;
42
42
  }
43
43
  export declare function formatSummary(s: RunSummary): string;
44
+ export declare const EARLY_LAUNCH_LIVENESS_MS = 60000;
45
+ /** Test seam — lowers the empty-pane liveness window without sleeping 60s per case. */
46
+ export declare function setEarlyLaunchLivenessMsForTests(ms: number): void;
47
+ export declare function resetEarlyLaunchLivenessMsForTests(): void;
44
48
  export declare function runDaemon(repoRoot: string, opts?: RunOptions): Promise<RunSummary>;
@@ -74,6 +74,18 @@ const BLOCKED_POLL_MS = 30_000; // between trailer-wait slices, check whether th
74
74
  const PROVIDER_DEATH_REQUEUE_CAP = 2; // v1.46 T1: requeue same assignment twice, then fall through to the normal ladder
75
75
  const PROVIDER_DEATH_BACKOFF_MS = 500; // short backoff before provider-death requeue
76
76
  const NO_TRAILER_DEMOTION_STREAK = 2; // OBS-57: consecutive no-trailer windows demote a channel for the rest of the run
77
+ // OBS-117 (v1.71 T6): a worker pane that never prints a byte by T+60s after dispatch is a dead
78
+ // channel — don't burn the full stall window waiting for a silent launch failure. Checked on the
79
+ // existing stall-wait poll cadence only (no new timer). Spinner/ANSI repaints count as output.
80
+ export const EARLY_LAUNCH_LIVENESS_MS = 60_000;
81
+ let earlyLaunchLivenessMs = EARLY_LAUNCH_LIVENESS_MS;
82
+ /** Test seam — lowers the empty-pane liveness window without sleeping 60s per case. */
83
+ export function setEarlyLaunchLivenessMsForTests(ms) {
84
+ earlyLaunchLivenessMs = ms;
85
+ }
86
+ export function resetEarlyLaunchLivenessMsForTests() {
87
+ earlyLaunchLivenessMs = EARLY_LAUNCH_LIVENESS_MS;
88
+ }
77
89
  async function commitsAheadOf(base, wt) {
78
90
  const head = await gitHead(wt);
79
91
  if (head === base)
@@ -217,6 +229,7 @@ export async function runDaemon(repoRoot, opts = {}) {
217
229
  // GATE-08 lesson at the humanGate guard: condition on the data, not the code path). Dead-code
218
230
  // equivalence to the router.ts:194 profile⇒undefined pattern: no map entry ⇒ today's literal.
219
231
  const resume = opts.resume ? journal.replayResumeState() : new Map();
232
+ const replayedExclusions = opts.resume ? journal.replayExcludedChannels() : new Set();
220
233
  if (opts.resume) {
221
234
  // v1.53 T5: a superseded run is dead — resuming it beside its successor is the exact
222
235
  // two-concurrent-runs hazard supersession exists to prevent. Fail closed, naming the successor.
@@ -252,7 +265,10 @@ export async function runDaemon(repoRoot, opts = {}) {
252
265
  continue;
253
266
  graph = setStatus(graph, id, st);
254
267
  }
255
- journal.append("run-resume", undefined, { pid: process.pid }); // v1.13 (VIS-11): record the live daemon pid for status liveness
268
+ journal.append("run-resume", undefined, {
269
+ pid: process.pid, // v1.13 (VIS-11): record the live daemon pid for status liveness
270
+ ...(replayedExclusions.size > 0 ? { excludedChannels: [...replayedExclusions].sort() } : {}),
271
+ });
256
272
  }
257
273
  else {
258
274
  baseRef = await gitHead(repoRoot);
@@ -301,7 +317,8 @@ export async function runDaemon(repoRoot, opts = {}) {
301
317
  const keptSlots = [];
302
318
  const runTag = runId.replace(/^run-/, ""); // full date-time — cross-run unique even across days
303
319
  // OBS-57: per-run in-run demotion — channels that burn consecutive no-trailer windows route around for later attempts.
304
- const demotedChannels = new Set();
320
+ // v1.71 OBS-119: on resume, re-seed from the journal fold (replayedExclusions) before any dispatch.
321
+ const demotedChannels = new Set(replayedExclusions);
305
322
  const noTrailerStreak = new Map();
306
323
  // OBS-17 T2: reconcile at every safe point — run start/resume (just journaled above), each task
307
324
  // terminal event, and run-end. The desired set is the pure journal fold (reconcile.ts); the driver
@@ -394,7 +411,9 @@ export async function runDaemon(repoRoot, opts = {}) {
394
411
  // fresh-budget release, prefer nextChannel over the surviving tried-list so burned channels are
395
412
  // not re-tried first (consult bans / prior failovers survive the release).
396
413
  const rs = resume.get(t.id);
397
- if (rs?.lastAssignment && rs.attempts > 0 && channels.some((c) => channelKey(c) === channelKey(rs.lastAssignment))) {
414
+ if (rs?.lastAssignment && rs.attempts > 0
415
+ && channels.some((c) => channelKey(c) === channelKey(rs.lastAssignment))
416
+ && !demotedChannels.has(channelKey(rs.lastAssignment))) {
398
417
  assignment = rs.lastAssignment; // restore the consult-chosen assignment (bypasses route()'s static re-pick)
399
418
  }
400
419
  else if (rs && rs.tried.length) {
@@ -690,6 +709,7 @@ export async function runDaemon(repoRoot, opts = {}) {
690
709
  let output;
691
710
  let exitCode;
692
711
  let timedOut = false;
712
+ let earlyLaunchDead = false;
693
713
  let settleParsed;
694
714
  let seedResult;
695
715
  if (interactive) {
@@ -724,12 +744,18 @@ export async function runDaemon(repoRoot, opts = {}) {
724
744
  // OBS-82: the stall clock compares NORMALIZED snapshots so a spinner glyph/elapsed-time
725
745
  // repaint is silence, not activity. ONLY this inactivity compare sees normalized text —
726
746
  // trailer detection, harvest, paging, and quota checks all read the raw pane.
747
+ let everHadOutput = output.length > 0;
727
748
  let lastStallSnapshot = normalizeStallSnapshot(output);
728
749
  let lastOutputAt = Date.now();
729
750
  while (Date.now() - lastOutputAt < stallWindowMs) {
730
751
  const sliceStart = Date.now();
731
752
  const remaining = stallWindowMs - (sliceStart - lastOutputAt);
732
- const slice = Math.min(BLOCKED_POLL_MS, Math.max(100, Math.min(stallWindowMs / 2, remaining)));
753
+ let slice = Math.min(BLOCKED_POLL_MS, Math.max(100, Math.min(stallWindowMs / 2, remaining)));
754
+ if (!everHadOutput) {
755
+ const earlyLeft = earlyLaunchLivenessMs - (sliceStart - attemptStart);
756
+ if (earlyLeft > 0)
757
+ slice = Math.min(slice, earlyLeft);
758
+ }
733
759
  if (await driver.waitOutput(slot, `(${trailerPattern(nonce)})|TICKMARKR_EXIT_${nonce}:\\d`, slice, { regex: true })) {
734
760
  // verify before accepting: a worker that merely DISPLAYS a marker (e.g. editing tickmarkr's
735
761
  // own source, where "TICKMARKR_EXIT:" is a string literal) must not end the wait. Only a
@@ -743,11 +769,20 @@ export async function runDaemon(repoRoot, opts = {}) {
743
769
  break;
744
770
  }
745
771
  }
746
- const currentStallSnapshot = normalizeStallSnapshot(await driver.read(slot, 1000));
772
+ const paneText = await driver.read(slot, 1000);
773
+ if (paneText.length > 0)
774
+ everHadOutput = true;
775
+ const currentStallSnapshot = normalizeStallSnapshot(paneText);
747
776
  if (currentStallSnapshot !== lastStallSnapshot) {
748
777
  lastStallSnapshot = currentStallSnapshot;
749
778
  lastOutputAt = Date.now();
750
779
  }
780
+ // OBS-117 (v1.71 T6): zero raw output by the early-launch deadline is a dead channel now.
781
+ if (!everHadOutput && Date.now() - attemptStart >= earlyLaunchLivenessMs) {
782
+ earlyLaunchDead = true;
783
+ output = paneText;
784
+ break;
785
+ }
751
786
  // v1.23 T2: piggyback on this poll slice — same cadence as blocked/idle checks, no new timer.
752
787
  await sampleContext();
753
788
  // page on "idle" too: herdr's blocked-scrape is strict and proved flaky for TUI dialogs
@@ -828,21 +863,35 @@ export async function runDaemon(repoRoot, opts = {}) {
828
863
  // OBS-82: same normalized-snapshot compare as the interactive site — spinner-only repaints
829
864
  // exhaust the budget here too; harvest below still reads the raw pane.
830
865
  const stallWindowMs = taskTimeoutMinutes * 60_000;
831
- let lastStallSnapshot = normalizeStallSnapshot(await driver.read(slot, 500));
866
+ const initialPane = await driver.read(slot, 500);
867
+ let everHadOutput = initialPane.length > 0;
868
+ let lastStallSnapshot = normalizeStallSnapshot(initialPane);
832
869
  let lastOutputAt = Date.now();
833
870
  finished = false;
834
871
  while (Date.now() - lastOutputAt < stallWindowMs) {
835
872
  const remaining = stallWindowMs - (Date.now() - lastOutputAt);
836
- const slice = Math.min(BLOCKED_POLL_MS, Math.max(100, Math.min(stallWindowMs / 2, remaining)));
873
+ let slice = Math.min(BLOCKED_POLL_MS, Math.max(100, Math.min(stallWindowMs / 2, remaining)));
874
+ if (!everHadOutput) {
875
+ const earlyLeft = earlyLaunchLivenessMs - (Date.now() - attemptStart);
876
+ if (earlyLeft > 0)
877
+ slice = Math.min(slice, earlyLeft);
878
+ }
837
879
  if (await driver.waitOutput(slot, `TICKMARKR_EXIT_${nonce}:\\d`, slice, { regex: true })) {
838
880
  finished = true;
839
881
  break;
840
882
  }
841
- const currentStallSnapshot = normalizeStallSnapshot(await driver.read(slot, 500));
883
+ const paneText = await driver.read(slot, 500);
884
+ if (paneText.length > 0)
885
+ everHadOutput = true;
886
+ const currentStallSnapshot = normalizeStallSnapshot(paneText);
842
887
  if (currentStallSnapshot !== lastStallSnapshot) {
843
888
  lastStallSnapshot = currentStallSnapshot;
844
889
  lastOutputAt = Date.now();
845
890
  }
891
+ if (!everHadOutput && Date.now() - attemptStart >= earlyLaunchLivenessMs) {
892
+ earlyLaunchDead = true;
893
+ break;
894
+ }
846
895
  }
847
896
  output = await driver.read(slot, 500);
848
897
  exitCode = Number(exitRe.exec(output)?.[1] ?? 1);
@@ -940,10 +989,13 @@ export async function runDaemon(repoRoot, opts = {}) {
940
989
  // AFTER the quota check so quota behavior stays byte-identical (a quota hit returns/continues
941
990
  // before reaching here); provider-outage lands here only once the v1.46 same-channel requeue
942
991
  // cap above is spent, so a transient blip still recovers in place.
943
- const dead = classifyDeadChannel(result);
992
+ // OBS-117 (v1.71 T6): a silent launch failure has no CLI signature to parse — the same
993
+ // setup-required typed dead-channel path a late-harvest "command not found" would take.
994
+ const dead = classifyDeadChannel(result) ?? (earlyLaunchDead ? "setup-required" : undefined);
944
995
  if (dead) {
945
996
  const from = channelKey(assignment);
946
997
  demotedChannels.add(from); // excluded for later attempts AND later tasks in this run
998
+ journal.append("channel-exclusion", t.id, { channel: from, reason: dead, kind: "dead-channel" });
947
999
  const next = failover("dead-channel");
948
1000
  journal.append("dead-channel-failover", t.id, { reason: dead, from, to: next ? channelKey(next) : null });
949
1001
  if (next) {
@@ -1130,9 +1182,14 @@ export async function runDaemon(repoRoot, opts = {}) {
1130
1182
  for (const t of ready) {
1131
1183
  const p = execTask(t)
1132
1184
  .catch(async (err) => {
1185
+ const taskEvents = journal.read().filter((e) => e.taskId === t.id);
1186
+ const dispatch = [...taskEvents].reverse().find((e) => e.event === "task-dispatch");
1187
+ const verifiedFailure = taskEvents.some((e) => e.event === "gate-result" && e.data.pass === false);
1188
+ const kind = verifiedFailure ? "gate-fail" : dispatch ? "dispatch" : "infra";
1189
+ const attempts = dispatch && Number.isInteger(dispatch.data.attempt) ? dispatch.data.attempt : 0;
1133
1190
  graph = setStatus(graph, t.id, "failed");
1134
1191
  saveGraph(repoRoot, graph);
1135
- journal.append("task-failed", t.id, { error: String(err) });
1192
+ journal.append("task-failed", t.id, { error: String(err), kind, attempts });
1136
1193
  journal.telemetry({ taskId: t.id, shape: t.shape, adapter: "-", model: "-", channel: "-", attempts: 0, outcome: "failed", durationMs: 0 });
1137
1194
  await reconcile({ spareLiveLlm: true }); // task-failed is a terminal event
1138
1195
  })
@@ -4,6 +4,7 @@ export interface InteractiveSeedResult {
4
4
  output: string;
5
5
  seedFailed: boolean;
6
6
  seedError?: string;
7
+ sessionId?: string;
7
8
  }
8
9
  export declare function runInteractiveSeed(opts: {
9
10
  driver: Pick<ExecutorDriver, "run" | "waitOutput" | "read">;
@@ -6,9 +6,17 @@ export async function runInteractiveSeed(opts) {
6
6
  const seed = opts.adapter.interactiveSeed;
7
7
  await opts.driver.run(opts.slot, seed.launch(opts.assignment.model));
8
8
  const ready = await opts.driver.waitOutput(opts.slot, seed.readinessMatch, opts.taskTimeoutMinutes * 60_000);
9
+ const banner = await opts.driver.read(opts.slot, 1000);
9
10
  if (!ready) {
10
- const output = await opts.driver.read(opts.slot, 1000);
11
- return { output, seedFailed: true, seedError: `readiness pattern not seen: ${seed.readinessMatch}` };
11
+ return { output: banner, seedFailed: true, seedError: `readiness pattern not seen: ${seed.readinessMatch}` };
12
+ }
13
+ let sessionId;
14
+ if (seed.confirmBanner) {
15
+ const confirm = seed.confirmBanner(banner, opts.assignment.model);
16
+ if (!confirm.ok) {
17
+ return { output: banner, seedFailed: true, seedError: confirm.error };
18
+ }
19
+ sessionId = confirm.sessionId;
12
20
  }
13
21
  const seedText = seed.seedLine(opts.promptFile);
14
22
  await opts.driver.run(opts.slot, seedText);
@@ -18,8 +26,8 @@ export async function runInteractiveSeed(opts) {
18
26
  output = await opts.driver.read(opts.slot, 1000);
19
27
  const bottom = output.trimEnd().split("\n").pop() ?? "";
20
28
  if (!bottom.includes(seedText)) {
21
- return { output, seedFailed: false };
29
+ return { output, seedFailed: false, ...(sessionId ? { sessionId } : {}) };
22
30
  }
23
31
  }
24
- return { output, seedFailed: true, seedError: "seed line never left the input box" };
32
+ return { output, seedFailed: true, seedError: "seed line never left the input box", ...(sessionId ? { sessionId } : {}) };
25
33
  }
@@ -15,13 +15,18 @@ export interface ResumeState {
15
15
  tried: string[];
16
16
  lastAssignment?: Assignment;
17
17
  }
18
+ export declare const CHANNEL_EXCLUSION_KINDS: readonly ["dead-channel"];
19
+ export type ChannelExclusionKind = (typeof CHANNEL_EXCLUSION_KINDS)[number];
18
20
  export declare const ATTEMPT_CAP_RELEASE: "attempt-cap";
19
- export declare const PARK_KINDS: readonly ["human-gate", "ladder-exhausted", "attempt-cap", "gate-fail", "quota", "reroute-exhausted", "setup", "stall", "merge-conflict", "tip-moved"];
21
+ export declare const PARK_KINDS: readonly ["human-gate", "ladder-exhausted", "attempt-cap", "gate-fail", "quota", "reroute-exhausted", "setup", "stall", "merge-conflict", "tip-moved", "infra", "dispatch"];
20
22
  export type ParkKind = (typeof PARK_KINDS)[number];
21
23
  export declare const RETRY_MODES: readonly ["resume", "fresh"];
22
24
  export type RetryMode = (typeof RETRY_MODES)[number];
23
25
  export declare const WORKER_RESULT_CAUSES: readonly ["provider-death", "stall-timeout", "malformed-trailer", "clean-exit-no-trailer"];
24
26
  export type WorkerResultCause = (typeof WORKER_RESULT_CAUSES)[number];
27
+ export declare function isQualityFailureParkKind(kind: ParkKind): boolean;
28
+ export declare function recordedTaskFailureKind(events: JournalEvent[], taskId: string): ParkKind | undefined;
29
+ export declare function runHasEnded(events: JournalEvent[]): boolean;
25
30
  /** OBS-53: classify worker-result failures so retries and routing see the true signal, not one lumped bucket. */
26
31
  export declare function classifyWorkerResultCause(opts: {
27
32
  output: string;
@@ -58,6 +63,8 @@ export declare const TelemetryRowSchema: z.ZodObject<{
58
63
  stall: "stall";
59
64
  "merge-conflict": "merge-conflict";
60
65
  "tip-moved": "tip-moved";
66
+ infra: "infra";
67
+ dispatch: "dispatch";
61
68
  }>>;
62
69
  tokens: z.ZodCatch<z.ZodOptional<z.ZodObject<{
63
70
  input: z.ZodNumber;
@@ -139,6 +146,7 @@ export declare class Journal {
139
146
  read(): JournalEvent[];
140
147
  replayStatuses(): Map<string, TaskStatus>;
141
148
  replayResumeState(): Map<string, ResumeState>;
149
+ replayExcludedChannels(): Set<string>;
142
150
  telemetry(row: TelemetryRow): void;
143
151
  readTelemetry(): TelemetryRow[];
144
152
  }
@@ -4,7 +4,7 @@ import { z } from "zod";
4
4
  import { channelKey, TokenUsageSchema } from "../adapters/types.js";
5
5
  import { stateDirName, tickmarkrDir } from "../graph/graph.js";
6
6
  import { TIERS } from "../graph/schema.js";
7
- import { buildProfile } from "../route/profile.js";
7
+ import { buildProfile, classify } from "../route/profile.js";
8
8
  import { redactSecrets } from "./redact.js";
9
9
  export function formatJournalNarration({ event, taskId, data }) {
10
10
  const assignment = data.assignment;
@@ -26,6 +26,9 @@ export function formatJournalNarration({ event, taskId, data }) {
26
26
  : String(direct);
27
27
  return [event, taskId, detail?.replace(/\s+/g, " ").slice(0, 120)].filter(Boolean).join(" — ");
28
28
  }
29
+ // v1.71 OBS-119: run-wide channel exclusions derived from journal events — companion to
30
+ // replayResumeState(), consumed by the daemon on resume to re-seed demotedChannels.
31
+ export const CHANNEL_EXCLUSION_KINDS = ["dead-channel"];
29
32
  // v1.24 OBS-18: explicit data on task-approved when the operator releases an attempt-cap park.
30
33
  // Approve stamps this; replayResumeState zeros the attempt counter (fresh budget) while keeping
31
34
  // tried (consult bans / burned channels). Absent on pre-v1.24 events ⇒ inert (corpus outcome-identical).
@@ -40,9 +43,43 @@ const DispatchAssignmentSchema = z.object({
40
43
  tier: z.enum(TIERS),
41
44
  });
42
45
  export const PARK_KINDS = ["human-gate", "ladder-exhausted", "attempt-cap", "gate-fail", "quota",
43
- "reroute-exhausted", "setup", "stall", "merge-conflict", "tip-moved"];
46
+ "reroute-exhausted", "setup", "stall", "merge-conflict", "tip-moved", "infra", "dispatch"];
44
47
  export const RETRY_MODES = ["resume", "fresh"];
45
48
  export const WORKER_RESULT_CAUSES = ["provider-death", "stall-timeout", "malformed-trailer", "clean-exit-no-trailer"];
49
+ // Status consumes the routing profile's existing quality split directly: verified park kinds classify
50
+ // to 0, while availability/recovery noise classifies to null. Keep the synthetic row here at the
51
+ // run↔route seam so presentation code never grows a second list of "bad" park kinds.
52
+ export function isQualityFailureParkKind(kind) {
53
+ return classify({
54
+ shape: "", adapter: "-", model: "-", channel: "-", attempts: 0,
55
+ outcome: "human", durationMs: 0, parkKind: kind,
56
+ }) === 0;
57
+ }
58
+ // The newest terminal event owns the cause. Unknown/malformed kinds fail toward undefined so a
59
+ // task-failed row keeps the legacy red treatment instead of being mistaken for recoverable noise.
60
+ export function recordedTaskFailureKind(events, taskId) {
61
+ for (let i = events.length - 1; i >= 0; i--) {
62
+ const e = events[i];
63
+ if (e.taskId !== taskId || (e.event !== "task-human" && e.event !== "task-failed"))
64
+ continue;
65
+ return typeof e.data.kind === "string" && PARK_KINDS.includes(e.data.kind)
66
+ ? e.data.kind
67
+ : undefined;
68
+ }
69
+ return undefined;
70
+ }
71
+ // Runs can end and later resume in the same journal. The newest lifecycle marker decides whether
72
+ // an unresolved task is still recoverable by this live daemon or belongs to an ended run.
73
+ export function runHasEnded(events) {
74
+ for (let i = events.length - 1; i >= 0; i--) {
75
+ const event = events[i].event;
76
+ if (event === "run-end" || event === "superseded")
77
+ return true;
78
+ if (event === "run-start" || event === "run-resume")
79
+ return false;
80
+ }
81
+ return false;
82
+ }
46
83
  // OBS-53: provider-outage signatures in dead worker output ("Unable to reach the model provider" and kin).
47
84
  const PROVIDER_OUTAGE_RE = /Unable to reach the model provider|cannot reach the model provider|model provider.*(?:unavailable|unreachable)/i;
48
85
  /** OBS-53: classify worker-result failures so retries and routing see the true signal, not one lumped bucket. */
@@ -424,6 +461,20 @@ export class Journal {
424
461
  }
425
462
  return m;
426
463
  }
464
+ // v1.71 OBS-119: run-wide exclusion fold — same replay discipline as replayResumeState().
465
+ // channel-exclusion is the typed event; dead-channel-failover.from is the pre-v1.71 compat seam.
466
+ replayExcludedChannels() {
467
+ const excluded = new Set();
468
+ for (const e of this.read()) {
469
+ if (e.event === "channel-exclusion" && typeof e.data.channel === "string") {
470
+ excluded.add(e.data.channel);
471
+ }
472
+ else if (e.event === "dead-channel-failover" && typeof e.data.from === "string") {
473
+ excluded.add(e.data.from);
474
+ }
475
+ }
476
+ return excluded;
477
+ }
427
478
  telemetry(row) {
428
479
  // T3 secret redaction: same persistence seam as append — credential-free rows are byte-identical.
429
480
  appendFileSync(join(this.dir, "telemetry.jsonl"), redactSecrets(JSON.stringify(row)) + "\n");
package/package.json CHANGED
@@ -1,6 +1,6 @@
1
1
  {
2
2
  "name": "tickmarkr",
3
- "version": "1.70.0",
3
+ "version": "1.71.0",
4
4
  "description": "Spec in, verified work out.",
5
5
  "type": "module",
6
6
  "license": "MIT",