tickmarkr 1.52.0 → 1.53.0
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.
- package/dist/adapters/kimi.d.ts +1 -0
- package/dist/adapters/kimi.js +24 -0
- package/dist/adapters/types.d.ts +2 -0
- package/dist/cli/commands/fleet.js +3 -0
- package/dist/cli/commands/plan.js +3 -3
- package/dist/cli/commands/report.js +14 -0
- package/dist/cli/commands/run.js +2 -0
- package/dist/cli/commands/status.js +5 -2
- package/dist/compile/collateral.d.ts +7 -1
- package/dist/compile/collateral.js +76 -19
- package/dist/config/config.d.ts +1 -0
- package/dist/config/config.js +6 -2
- package/dist/gates/review.d.ts +2 -1
- package/dist/gates/review.js +13 -4
- package/dist/run/daemon.d.ts +1 -0
- package/dist/run/daemon.js +24 -5
- package/package.json +1 -1
package/dist/adapters/kimi.d.ts
CHANGED
|
@@ -2,4 +2,5 @@ import { type WorkerAdapter, type WorkerResult } from "./types.js";
|
|
|
2
2
|
export declare function kimiAuthed(credentialsText: string, nowMs: number): boolean;
|
|
3
3
|
export declare function parseKimiModels(raw: string): string[];
|
|
4
4
|
export declare function parseKimiResult(raw: string, nonce: string): WorkerResult;
|
|
5
|
+
export declare function kimiSessionId(output: string): string | undefined;
|
|
5
6
|
export declare const kimi: WorkerAdapter;
|
package/dist/adapters/kimi.js
CHANGED
|
@@ -46,6 +46,21 @@ export function parseKimiResult(raw, nonce) {
|
|
|
46
46
|
const stripped = raw.split("\n").map((l) => l.replace(/^[\s]*[•*-]\s+/, "")).join("\n");
|
|
47
47
|
return parseWorkerResult(stripped, nonce);
|
|
48
48
|
}
|
|
49
|
+
// v1.53 T3: session-id capture from the run-output trailer — every `kimi -p` run (fresh or resumed)
|
|
50
|
+
// ends with `To resume this session: kimi -r session_<uuid>` (live probe 2026-07-18). Anchored full
|
|
51
|
+
// line only: prompt/model prose can contain lookalike text, and the anchored charset keeps a
|
|
52
|
+
// captured id shell-safe by construction (shq in resumeCommand is the second layer). Last valid
|
|
53
|
+
// line wins — a run may echo stale resume lines mid-transcript.
|
|
54
|
+
const RESUME_TRAILER_RE = /^\s*To resume this session: kimi -r (session_[0-9a-f-]+)\s*$/;
|
|
55
|
+
export function kimiSessionId(output) {
|
|
56
|
+
let id;
|
|
57
|
+
for (const line of output.split("\n")) {
|
|
58
|
+
const m = RESUME_TRAILER_RE.exec(line);
|
|
59
|
+
if (m)
|
|
60
|
+
id = m[1];
|
|
61
|
+
}
|
|
62
|
+
return id;
|
|
63
|
+
}
|
|
49
64
|
export const kimi = {
|
|
50
65
|
id: "kimi",
|
|
51
66
|
vendor: "moonshot",
|
|
@@ -73,6 +88,15 @@ export const kimi = {
|
|
|
73
88
|
// ("unknown command '…'", live-verified 2026-07-17, OBS-67) and -p is non-interactive-only.
|
|
74
89
|
// null → the daemon's print fallback (types.ts:101) keeps kimi workers visible without a TUI.
|
|
75
90
|
interactiveCommand: () => null,
|
|
91
|
+
// v1.53 T3 resume — live-probed 2026-07-18: `-p` + `-S <id>` compose cleanly (no OBS-67-class
|
|
92
|
+
// flag rejection) and the resumed session carries prior conversation state. `-S <id>` is the
|
|
93
|
+
// deterministic form; `-c` rejected as primary — cwd-keyed, nondeterministic under worktree
|
|
94
|
+
// recreation (probe finding 4). Never bare `-S` (it launches the interactive session picker).
|
|
95
|
+
resumeCommand: (sessionId, promptFile, model) => `kimi -S ${shq(sessionId)} -p "$(cat ${shq(promptFile)})" --model ${shq(model)} --output-format text`,
|
|
96
|
+
sessionIdFrom: kimiSessionId,
|
|
97
|
+
// KIMI-03: no contextUsage surface exists, so a known-context requirement would leave resume
|
|
98
|
+
// dead code — declare the unknown-context opt-in the daemon retry seam enforces.
|
|
99
|
+
resumeUnknownContext: true,
|
|
76
100
|
invoke(task, _cwd, a, ctx) {
|
|
77
101
|
return { command: this.headlessCommand(ctx.promptFile, a.model) };
|
|
78
102
|
},
|
package/dist/adapters/types.d.ts
CHANGED
|
@@ -79,6 +79,8 @@ export interface WorkerAdapter {
|
|
|
79
79
|
headlessCommand(promptFile: string, model: string): string;
|
|
80
80
|
interactiveCommand(promptFile: string, model: string): string | null;
|
|
81
81
|
resumeCommand?(sessionId: string, promptFile: string, model: string): string;
|
|
82
|
+
sessionIdFrom?(output: string): string | undefined;
|
|
83
|
+
resumeUnknownContext?: boolean;
|
|
82
84
|
invoke(task: Task, cwd: string, a: Assignment, ctx: {
|
|
83
85
|
promptFile: string;
|
|
84
86
|
}): Invocation;
|
|
@@ -140,6 +140,9 @@ function openTerm(input, output) {
|
|
|
140
140
|
finally {
|
|
141
141
|
rl.close();
|
|
142
142
|
setRaw(true);
|
|
143
|
+
// rl.close() pauses the input stream — without a resume the next key() await has
|
|
144
|
+
// nothing keeping the event loop alive and the process silently exits 0 (OBS-77)
|
|
145
|
+
input.resume();
|
|
143
146
|
queue.splice(typedFrom);
|
|
144
147
|
prevLines = 0;
|
|
145
148
|
}
|
|
@@ -2,7 +2,7 @@ import { allAdapters, discoverChannels, doctorAgeMs, modelAuthExclusions, probeA
|
|
|
2
2
|
import { formatModelAuthLine, contextWindowLints, modelLints, ttyVisual } from "../../adapters/model-lints.js";
|
|
3
3
|
import { GLYPHS, dim, rule, title, warn } from "../../brand.js";
|
|
4
4
|
import { parseArgs } from "node:util";
|
|
5
|
-
import { collateralLints } from "../../compile/collateral.js";
|
|
5
|
+
import { collateralLints, sourceScopeLints } from "../../compile/collateral.js";
|
|
6
6
|
import { DEFAULT_CONFIG, ROUTING_MODES, TIER_RANK } from "../../config/config.js";
|
|
7
7
|
import { loadGraph } from "../../graph/graph.js";
|
|
8
8
|
import { resolveRunMode } from "../../run/daemon.js";
|
|
@@ -170,8 +170,8 @@ export async function plan(argv, cwd = process.cwd(), adapters = allAdapters())
|
|
|
170
170
|
const windowLints = contextWindowLints(g.tasks, routed, cfg, cwd);
|
|
171
171
|
if (windowLints.length)
|
|
172
172
|
lines.push("", "context window lints:", ...windowLints.map((l) => ` ! ${l}`));
|
|
173
|
-
// OBS-12/13/14/21: advisory only — never blocks --route-strict (run refuses on routing lints only)
|
|
174
|
-
const scopeLints = collateralLints(g.tasks, cwd);
|
|
173
|
+
// OBS-12/13/14/21 + OBS-76: advisory only — never blocks --route-strict (run refuses on routing lints only)
|
|
174
|
+
const scopeLints = [...collateralLints(g.tasks, cwd), ...sourceScopeLints(g.tasks, cwd)];
|
|
175
175
|
if (scopeLints.length)
|
|
176
176
|
lines.push("", "scope lints:", ...scopeLints.map((l) => ` ! ${l}`));
|
|
177
177
|
return stylizePlan(lines.join("\n"));
|
|
@@ -94,6 +94,14 @@ const wallClock = (start, end) => {
|
|
|
94
94
|
return minutes ? `${minutes}m ${seconds % 60}s` : `${seconds}s`;
|
|
95
95
|
};
|
|
96
96
|
const detail = (value) => typeof value === "string" || typeof value === "number" ? String(value) : EM;
|
|
97
|
+
// v1.53 T5: supersession is derived from the run's OWN journal only — `superseded by` from the
|
|
98
|
+
// appended superseded event (last wins), `supersedes` from the run-start stamp. No cross-run scan.
|
|
99
|
+
const supersession = (events) => {
|
|
100
|
+
const by = [...events].reverse().find((e) => e.event === "superseded" && typeof e.data.by === "string")?.data.by;
|
|
101
|
+
const start = events.find((e) => e.event === "run-start");
|
|
102
|
+
const supersedes = typeof start?.data.supersedes === "string" ? start.data.supersedes : undefined;
|
|
103
|
+
return { ...(typeof by === "string" ? { supersededBy: by } : {}), ...(supersedes ? { supersedes } : {}) };
|
|
104
|
+
};
|
|
97
105
|
const taskIds = (events) => {
|
|
98
106
|
const seen = new Set();
|
|
99
107
|
const out = [];
|
|
@@ -152,10 +160,13 @@ export function renderMarkdownRecord(runId, events, prices = [], rows = []) {
|
|
|
152
160
|
usageLines.push(...journalUsage(events, new Set(prices.map((row) => `${row.channel}:${row.adapter}:${row.model}`))));
|
|
153
161
|
if (!usageLines.length)
|
|
154
162
|
usageLines.push("- **not recorded:** attempts/windows: not measurable; tokens: not measurable; price: not measurable");
|
|
163
|
+
const sup = supersession(events);
|
|
155
164
|
const lines = [
|
|
156
165
|
`# tickmarkr engagement`,
|
|
157
166
|
"",
|
|
158
167
|
`- **runId:** ${runId}`,
|
|
168
|
+
...(sup.supersededBy ? [`- **superseded by:** ${sup.supersededBy}`] : []),
|
|
169
|
+
...(sup.supersedes ? [`- **supersedes:** ${sup.supersedes}`] : []),
|
|
159
170
|
`- **base ref:** ${baseRef}`,
|
|
160
171
|
`- **branch:** ${branch}`,
|
|
161
172
|
`- **done:** ${count("done")}`,
|
|
@@ -269,8 +280,11 @@ function textReport(runId, events, rows, cwd) {
|
|
|
269
280
|
const escalations = events.filter((e) => e.event === "escalation").length;
|
|
270
281
|
const consults = events.filter((e) => e.event === "consult-verdict").length;
|
|
271
282
|
const failovers = events.filter((e) => e.event === "quota-failover").length;
|
|
283
|
+
const sup = supersession(events);
|
|
272
284
|
return [
|
|
273
285
|
`tickmarkr engagement — ${runId}`,
|
|
286
|
+
...(sup.supersededBy ? [`superseded by ${sup.supersededBy}`] : []),
|
|
287
|
+
...(sup.supersedes ? [`supersedes ${sup.supersedes}`] : []),
|
|
274
288
|
"",
|
|
275
289
|
"engagement summary — audit trail:",
|
|
276
290
|
...[...groups.entries()].map(([k, g]) => ` ${k.padEnd(30)} tasks ${g.rows.length}, attempts ${g.rows.reduce((s, r) => s + r.attempts, 0)}, done ${g.rows.filter((r) => r.outcome === "done").length}`),
|
package/dist/cli/commands/run.js
CHANGED
|
@@ -39,6 +39,7 @@ export async function run(argv, cwd = process.cwd()) {
|
|
|
39
39
|
"no-explore": { type: "boolean" },
|
|
40
40
|
mode: { type: "string" },
|
|
41
41
|
quality: { type: "boolean" },
|
|
42
|
+
supersedes: { type: "string" },
|
|
42
43
|
},
|
|
43
44
|
});
|
|
44
45
|
if (values.concurrency !== undefined) {
|
|
@@ -83,6 +84,7 @@ export async function run(argv, cwd = process.cwd()) {
|
|
|
83
84
|
concurrency: values.concurrency ? Number(values.concurrency) : undefined,
|
|
84
85
|
driver: pickDriver(cfg, values.driver),
|
|
85
86
|
mode: flagMode,
|
|
87
|
+
supersedes: values.supersedes,
|
|
86
88
|
narrate: (event) => console.log(narrationLine(event)),
|
|
87
89
|
});
|
|
88
90
|
const out = `run ${s.runId} finished — ${formatSummary(s)} (merge to main is a human decision)`;
|
|
@@ -120,9 +120,12 @@ const renderFrame = (cwd) => {
|
|
|
120
120
|
let events = [];
|
|
121
121
|
const contexts = new Map();
|
|
122
122
|
let comparable = false;
|
|
123
|
+
let supersededBy; // v1.53 T5: this run is dead — a newer run replaced it
|
|
123
124
|
if (runId) {
|
|
124
125
|
const j = Journal.open(cwd, runId);
|
|
125
126
|
events = j.read();
|
|
127
|
+
const sup = [...events].reverse().find((e) => e.event === "superseded" && typeof e.data.by === "string");
|
|
128
|
+
supersededBy = sup?.data.by;
|
|
126
129
|
// T3 (Sol #2 / Fable F2): the SAME comparator resume uses (engagementComparable) — one decision,
|
|
127
130
|
// two consumers. graphDefinitionHash (compiled task definitions only) survives status/evidence
|
|
128
131
|
// mutation but changes when a task definition changes, so a recompiled graph is detected here too.
|
|
@@ -164,7 +167,7 @@ const renderFrame = (cwd) => {
|
|
|
164
167
|
return `${prefix}${shortGoal(t.goal, Math.max(0, width - prefix.length - suffix.length))}${suffix}`;
|
|
165
168
|
});
|
|
166
169
|
const header = runId
|
|
167
|
-
? `tickmarkr status${divider}run ${runId}${!comparable ? `${divider}${NOT_COMPARABLE_NOTICE}` : ""}${divider}${liveness(events).replaceAll(" · ", divider)}${divider}${done}/${g.tasks.length} done`
|
|
170
|
+
? `tickmarkr status${divider}run ${runId}${supersededBy ? `${divider}superseded by ${supersededBy}` : ""}${!comparable ? `${divider}${NOT_COMPARABLE_NOTICE}` : ""}${divider}${liveness(events).replaceAll(" · ", divider)}${divider}${done}/${g.tasks.length} done`
|
|
168
171
|
: `tickmarkr status${divider}no runs yet${divider}${done}/${g.tasks.length} done`;
|
|
169
172
|
const legendLine = ` gates: ${GATE_NAMES.map((gate) => `${GATE_KEYS[gate]} ${gate}`).join(divider)}`;
|
|
170
173
|
return [header, legendLine, ...rows].join("\n");
|
|
@@ -186,7 +189,7 @@ const renderFrame = (cwd) => {
|
|
|
186
189
|
const tally = `${done}/${g.tasks.length} done`;
|
|
187
190
|
const header = ` ${title(runId ? `run ${runId}` : "tickmarkr")}${dot}` +
|
|
188
191
|
(runId
|
|
189
|
-
? `${!comparable ? `${warn(NOT_COMPARABLE_NOTICE)}${dot}` : ""}${live}${dot}`
|
|
192
|
+
? `${supersededBy ? `${warn(`superseded by ${supersededBy}`)}${dot}` : ""}${!comparable ? `${warn(NOT_COMPARABLE_NOTICE)}${dot}` : ""}${live}${dot}`
|
|
190
193
|
: `no runs yet${dot}`) +
|
|
191
194
|
`${gauge} ${done === g.tasks.length && g.tasks.length > 0 ? ok(tally) : tally}`;
|
|
192
195
|
const hr = rule(Math.min(width, 100));
|
|
@@ -1,6 +1,12 @@
|
|
|
1
|
-
import type
|
|
1
|
+
import { type Task } from "../graph/schema.js";
|
|
2
2
|
/**
|
|
3
3
|
* Return human-readable scope-lint lines for plan output (no `!` prefix — plan owns that).
|
|
4
4
|
* Each line names the task id and at least one missing collateral test path.
|
|
5
5
|
*/
|
|
6
6
|
export declare function collateralLints(tasks: ReadonlyArray<Pick<Task, "id" | "files">>, repoRoot: string): string[];
|
|
7
|
+
/**
|
|
8
|
+
* OBS-76 class: sweep src/ for out-of-scope source files that reference a symbol the acceptance
|
|
9
|
+
* criteria name — the v1.52 router.ts omission, named at plan time instead of one judge round in.
|
|
10
|
+
* Advisory plan output only, same contract as collateralLints.
|
|
11
|
+
*/
|
|
12
|
+
export declare function sourceScopeLints(tasks: ReadonlyArray<Pick<Task, "id" | "files" | "acceptance">>, repoRoot: string): string[];
|
|
@@ -1,10 +1,11 @@
|
|
|
1
1
|
import { readdirSync, readFileSync, statSync } from "node:fs";
|
|
2
2
|
import { extname, join, relative } from "node:path";
|
|
3
3
|
import picomatch from "picomatch";
|
|
4
|
-
|
|
5
|
-
//
|
|
6
|
-
|
|
7
|
-
|
|
4
|
+
import { renderAcceptanceItem } from "../graph/schema.js";
|
|
5
|
+
// Advisory plan-time scan only (OBS-12/13/14/21, OBS-76). NEVER expands files[], fails compile,
|
|
6
|
+
// or feeds the scope gate — a warning the author acts on. Plain-text (no AST), capped + sorted.
|
|
7
|
+
/** Max files walked per root (sorted walk; rest ignored). */
|
|
8
|
+
const MAX_WALK_FILES = 400;
|
|
8
9
|
/** Max collateral test paths listed per task. */
|
|
9
10
|
const MAX_HITS_PER_TASK = 20;
|
|
10
11
|
/** Skip giant fixtures / snapshots. */
|
|
@@ -22,23 +23,23 @@ function needlesFor(srcPath) {
|
|
|
22
23
|
// repo-relative import forms that appear in tests (ESM often uses .js)
|
|
23
24
|
return [...new Set([noExt, `${noExt}.js`, `${noExt}.ts`, n])].filter((s) => s.length > 0);
|
|
24
25
|
}
|
|
25
|
-
function
|
|
26
|
-
const root = join(repoRoot,
|
|
26
|
+
function walkCode(repoRoot, subdir) {
|
|
27
|
+
const root = join(repoRoot, subdir);
|
|
27
28
|
const out = [];
|
|
28
29
|
const walk = (dir) => {
|
|
29
|
-
if (out.length >=
|
|
30
|
+
if (out.length >= MAX_WALK_FILES)
|
|
30
31
|
return;
|
|
31
32
|
let entries;
|
|
32
33
|
try {
|
|
33
34
|
entries = readdirSync(dir, { withFileTypes: true });
|
|
34
35
|
}
|
|
35
36
|
catch {
|
|
36
|
-
return; // no
|
|
37
|
+
return; // no such root or unreadable — zero lints
|
|
37
38
|
}
|
|
38
39
|
// deterministic order
|
|
39
40
|
entries.sort((a, b) => a.name.localeCompare(b.name));
|
|
40
41
|
for (const e of entries) {
|
|
41
|
-
if (out.length >=
|
|
42
|
+
if (out.length >= MAX_WALK_FILES)
|
|
42
43
|
return;
|
|
43
44
|
if (e.name === "node_modules" || e.name === ".git" || e.name.startsWith("."))
|
|
44
45
|
continue;
|
|
@@ -70,17 +71,10 @@ function mentions(content, needles) {
|
|
|
70
71
|
}
|
|
71
72
|
return false;
|
|
72
73
|
}
|
|
73
|
-
|
|
74
|
-
|
|
75
|
-
* Each line names the task id and at least one missing collateral test path.
|
|
76
|
-
*/
|
|
77
|
-
export function collateralLints(tasks, repoRoot) {
|
|
78
|
-
const testFiles = walkTests(repoRoot);
|
|
79
|
-
if (!testFiles.length)
|
|
80
|
-
return [];
|
|
81
|
-
// cache file bodies (one read per test path)
|
|
74
|
+
// cache file bodies (one read per path)
|
|
75
|
+
function makeReader(repoRoot) {
|
|
82
76
|
const body = new Map();
|
|
83
|
-
|
|
77
|
+
return (rel) => {
|
|
84
78
|
if (body.has(rel))
|
|
85
79
|
return body.get(rel);
|
|
86
80
|
try {
|
|
@@ -98,6 +92,16 @@ export function collateralLints(tasks, repoRoot) {
|
|
|
98
92
|
return null;
|
|
99
93
|
}
|
|
100
94
|
};
|
|
95
|
+
}
|
|
96
|
+
/**
|
|
97
|
+
* Return human-readable scope-lint lines for plan output (no `!` prefix — plan owns that).
|
|
98
|
+
* Each line names the task id and at least one missing collateral test path.
|
|
99
|
+
*/
|
|
100
|
+
export function collateralLints(tasks, repoRoot) {
|
|
101
|
+
const testFiles = walkCode(repoRoot, "tests");
|
|
102
|
+
if (!testFiles.length)
|
|
103
|
+
return [];
|
|
104
|
+
const read = makeReader(repoRoot);
|
|
101
105
|
const lines = [];
|
|
102
106
|
for (const t of tasks) {
|
|
103
107
|
// OBS-22: scopeGate accepts picomatch globs; advisory collateral warnings must agree.
|
|
@@ -128,3 +132,56 @@ export function collateralLints(tasks, repoRoot) {
|
|
|
128
132
|
}
|
|
129
133
|
return lines;
|
|
130
134
|
}
|
|
135
|
+
// v1.53 T4 (OBS-76): needles are code-shaped tokens only (camelCase / snake_case) — plain prose
|
|
136
|
+
// words never match, so prose-only criteria yield zero needles instead of alarm-fatigue noise.
|
|
137
|
+
// ponytail: token heuristic, not AST symbol resolution — promote after a version of precision data.
|
|
138
|
+
function criteriaSymbols(acceptance) {
|
|
139
|
+
const out = new Set();
|
|
140
|
+
for (const item of acceptance) {
|
|
141
|
+
for (const tok of renderAcceptanceItem(item).match(/\b[A-Za-z_][A-Za-z0-9_]*\b/g) ?? []) {
|
|
142
|
+
if (tok.includes("_") || /[a-z][A-Z]/.test(tok))
|
|
143
|
+
out.add(tok);
|
|
144
|
+
}
|
|
145
|
+
}
|
|
146
|
+
return [...out].sort();
|
|
147
|
+
}
|
|
148
|
+
/**
|
|
149
|
+
* OBS-76 class: sweep src/ for out-of-scope source files that reference a symbol the acceptance
|
|
150
|
+
* criteria name — the v1.52 router.ts omission, named at plan time instead of one judge round in.
|
|
151
|
+
* Advisory plan output only, same contract as collateralLints.
|
|
152
|
+
*/
|
|
153
|
+
export function sourceScopeLints(tasks, repoRoot) {
|
|
154
|
+
const perTask = tasks
|
|
155
|
+
.map((t) => ({ t, needles: criteriaSymbols(t.acceptance) }))
|
|
156
|
+
.filter((x) => x.needles.length);
|
|
157
|
+
if (!perTask.length)
|
|
158
|
+
return [];
|
|
159
|
+
const srcFiles = walkCode(repoRoot, "src");
|
|
160
|
+
if (!srcFiles.length)
|
|
161
|
+
return [];
|
|
162
|
+
const read = makeReader(repoRoot);
|
|
163
|
+
const lines = [];
|
|
164
|
+
for (const { t, needles } of perTask) {
|
|
165
|
+
// OBS-22: scopeGate accepts picomatch globs; advisory warnings must agree.
|
|
166
|
+
const scoped = picomatch(t.files.map((f) => f.replace(/^\.\//, "")), { dot: true });
|
|
167
|
+
// needles are word-chars by construction — no regex escaping needed; whole-word match only
|
|
168
|
+
const res = needles.map((n) => new RegExp(`\\b${n}\\b`));
|
|
169
|
+
const hits = [];
|
|
170
|
+
for (const sf of srcFiles) {
|
|
171
|
+
if (scoped(sf))
|
|
172
|
+
continue;
|
|
173
|
+
const text = read(sf);
|
|
174
|
+
if (text === null)
|
|
175
|
+
continue;
|
|
176
|
+
if (res.some((re) => re.test(text)))
|
|
177
|
+
hits.push(sf);
|
|
178
|
+
}
|
|
179
|
+
if (!hits.length)
|
|
180
|
+
continue;
|
|
181
|
+
hits.sort();
|
|
182
|
+
const listed = hits.slice(0, MAX_HITS_PER_TASK).join(", ");
|
|
183
|
+
const tail = hits.length > MAX_HITS_PER_TASK ? " (capped)" : "";
|
|
184
|
+
lines.push(`${t.id}: criteria implicate out-of-scope source not in files[]: ${listed}${tail}`);
|
|
185
|
+
}
|
|
186
|
+
return lines;
|
|
187
|
+
}
|
package/dist/config/config.d.ts
CHANGED
|
@@ -168,6 +168,7 @@ export declare const TickmarkrConfigSchema: z.ZodObject<{
|
|
|
168
168
|
review: z.ZodObject<{
|
|
169
169
|
complexityThreshold: z.ZodNumber;
|
|
170
170
|
required: z.ZodBoolean;
|
|
171
|
+
prefer: z.ZodOptional<z.ZodArray<z.ZodString>>;
|
|
171
172
|
}, z.core.$strip>;
|
|
172
173
|
consult: z.ZodObject<{
|
|
173
174
|
adapter: z.ZodString;
|
package/dist/config/config.js
CHANGED
|
@@ -152,7 +152,9 @@ export const TickmarkrConfigSchema = z.object({
|
|
|
152
152
|
allowDeviations: z.array(z.string()),
|
|
153
153
|
}).partial().optional(),
|
|
154
154
|
judge: z.object({ adapter: z.string(), model: z.string() }),
|
|
155
|
-
|
|
155
|
+
// v1.53 T2: prefer — ordered reviewer preference (entry grammar: adapter | adapter:model, same as
|
|
156
|
+
// routing.map.prefer). Reorders diversity-eligible channels only; never widens or narrows eligibility.
|
|
157
|
+
review: z.object({ complexityThreshold: z.number(), required: z.boolean(), prefer: z.array(z.string()).optional() }),
|
|
156
158
|
consult: z.object({ adapter: z.string(), model: z.string(), stallMinutes: z.number().positive() }),
|
|
157
159
|
visibility: z.object({
|
|
158
160
|
llm: z.enum(["pane", "headless"]),
|
|
@@ -474,7 +476,9 @@ export function configTemplate(overlay) {
|
|
|
474
476
|
# test: npm test
|
|
475
477
|
# byShape:
|
|
476
478
|
# docs: { acceptance: false, review: false } # baseline, evidence, and scope are mandatory
|
|
477
|
-
# review: { complexityThreshold: 7, required: true }
|
|
479
|
+
# review: { complexityThreshold: 7, required: true, prefer: [codex:gpt-5.6-sol, kimi] }
|
|
480
|
+
# # prefer: ordered reviewer seat preference (adapter | adapter:model); ranks
|
|
481
|
+
# # diversity-eligible channels only — never admits a same-vendor/same-model reviewer
|
|
478
482
|
# consult: { adapter: claude-code, model: fable, stallMinutes: 15 }
|
|
479
483
|
# cost: # v1.20 REC-02 — OPTIONAL price table for the usage/cost report. Absent ⇒ every
|
|
480
484
|
# # channel reports "not measurable" (never a crash or a fake $0). Two economics,
|
package/dist/gates/review.d.ts
CHANGED
|
@@ -15,5 +15,6 @@ export declare function checkDiffCap(gate: string, measured: number, cap: number
|
|
|
15
15
|
export declare function isDiffCapPark(result: GateResult): boolean;
|
|
16
16
|
export declare function diffCapParkReason(results: GateResult[]): string | null;
|
|
17
17
|
export declare function modelId(model: string): string;
|
|
18
|
-
export declare function pickReviewer(author: Assignment, channels: BillingChannel[], exclude?: string[]
|
|
18
|
+
export declare function pickReviewer(author: Assignment, channels: BillingChannel[], exclude?: string[], // v1.1 failover: reviewer channels that already produced garbage for this task
|
|
19
|
+
prefer?: string[]): BillingChannel | null;
|
|
19
20
|
export declare function reviewGate(task: Task, worktree: string, baseRef: string, author: Assignment, channels: BillingChannel[], adapters: WorkerAdapter[], cfg: TickmarkrConfig, via?: GateVia, excludeReviewers?: string[]): Promise<GateResult>;
|
package/dist/gates/review.js
CHANGED
|
@@ -39,7 +39,15 @@ export function diffCapParkReason(results) {
|
|
|
39
39
|
export function modelId(model) {
|
|
40
40
|
return model.slice(model.lastIndexOf("/") + 1);
|
|
41
41
|
}
|
|
42
|
-
|
|
42
|
+
// v1.53 T2: same entry grammar as routing.map.prefer (router.ts preferIndex — router is out of this
|
|
43
|
+
// module's dependency direction for a private fn, so the 3 lines live here too): `adapter` matches
|
|
44
|
+
// every channel of that adapter, `adapter:model` exactly one; unmatched channels sort after all entries.
|
|
45
|
+
function reviewPreferIndex(c, prefer) {
|
|
46
|
+
const i = prefer.findIndex((p) => p === c.adapter || p === channelKey(c));
|
|
47
|
+
return i === -1 ? prefer.length : i;
|
|
48
|
+
}
|
|
49
|
+
export function pickReviewer(author, channels, exclude = [], // v1.1 failover: reviewer channels that already produced garbage for this task
|
|
50
|
+
prefer = []) {
|
|
43
51
|
// FLEET-05 success criterion 2: an author not resolvable in the channel list yields NO reviewer.
|
|
44
52
|
// The old `?? author.adapter` fallback compared an adapter id to vendor names, matched nothing, and
|
|
45
53
|
// admitted every reviewer — including the author's own channel (fail-OPEN). null lands on reviewGate's
|
|
@@ -49,15 +57,16 @@ export function pickReviewer(author, channels, exclude = []) {
|
|
|
49
57
|
return null;
|
|
50
58
|
return (channels
|
|
51
59
|
// two independent axes: different vendor AND different base-model identity (ADDED TO the vendor
|
|
52
|
-
// rule, never replacing it — a future edit can't silently drop either).
|
|
60
|
+
// rule, never replacing it — a future edit can't silently drop either). The diversity filter runs
|
|
61
|
+
// BEFORE preference ranking: prefer sorts survivors only, so no entry can resurrect an excluded channel.
|
|
53
62
|
.filter((c) => c.vendor !== authorChannel.vendor && modelId(c.model) !== modelId(author.model) && !exclude.includes(channelKey(c)))
|
|
54
|
-
.sort((a, b) => TIER_RANK[b.tier] - TIER_RANK[a.tier] || marginalCostRank(a) - marginalCostRank(b))[0] ?? null);
|
|
63
|
+
.sort((a, b) => reviewPreferIndex(a, prefer) - reviewPreferIndex(b, prefer) || TIER_RANK[b.tier] - TIER_RANK[a.tier] || marginalCostRank(a) - marginalCostRank(b))[0] ?? null);
|
|
55
64
|
}
|
|
56
65
|
export async function reviewGate(task, worktree, baseRef, author, channels, adapters, cfg, via, excludeReviewers) {
|
|
57
66
|
if (task.complexity < cfg.review.complexityThreshold) {
|
|
58
67
|
return { gate: "review", pass: true, details: `skipped — complexity ${task.complexity} < threshold ${cfg.review.complexityThreshold}`, meta: { skipped: true } };
|
|
59
68
|
}
|
|
60
|
-
const reviewer = pickReviewer(author, channels, excludeReviewers ?? []);
|
|
69
|
+
const reviewer = pickReviewer(author, channels, excludeReviewers ?? [], cfg.review.prefer ?? []);
|
|
61
70
|
if (!reviewer) {
|
|
62
71
|
return cfg.review.required
|
|
63
72
|
? { gate: "review", pass: false, details: "no cross-vendor reviewer available (diversity rule); set review.required:false to waive" }
|
package/dist/run/daemon.d.ts
CHANGED
package/dist/run/daemon.js
CHANGED
|
@@ -98,6 +98,10 @@ export async function runDaemon(repoRoot, opts = {}) {
|
|
|
98
98
|
// HARD-01/02: hold the run lock across the whole read-modify-write of graph.json. Acquire
|
|
99
99
|
// BEFORE loadGraph; release in the finally below (every exit path, incl. throws).
|
|
100
100
|
const runId = opts.runId ?? newRunId();
|
|
101
|
+
// v1.53 T5: an unknown --supersedes id must fail BEFORE any run starts — Journal.open throws and
|
|
102
|
+
// no lock, journal, or baseline for the new run has been created yet. Opened without a narrate
|
|
103
|
+
// sink: the prior journal append below is silent bookkeeping, not this run's narration.
|
|
104
|
+
const prior = opts.supersedes !== undefined ? Journal.open(repoRoot, opts.supersedes) : undefined;
|
|
101
105
|
// T6 narrator: one live status surface per run (herdr only — driver.narrator is undefined on
|
|
102
106
|
// subprocess, so the optional-chain open below is a no-op there). Cosmetic-only: any failure is
|
|
103
107
|
// swallowed (never affects the run); the operator closes a surviving watch pane.
|
|
@@ -140,6 +144,11 @@ export async function runDaemon(repoRoot, opts = {}) {
|
|
|
140
144
|
// equivalence to the router.ts:194 profile⇒undefined pattern: no map entry ⇒ today's literal.
|
|
141
145
|
const resume = opts.resume ? journal.replayResumeState() : new Map();
|
|
142
146
|
if (opts.resume) {
|
|
147
|
+
// v1.53 T5: a superseded run is dead — resuming it beside its successor is the exact
|
|
148
|
+
// two-concurrent-runs hazard supersession exists to prevent. Fail closed, naming the successor.
|
|
149
|
+
const superseded = [...journal.read()].reverse().find((e) => e.event === "superseded" && typeof e.data.by === "string");
|
|
150
|
+
if (superseded)
|
|
151
|
+
throw new Error(`refusing to resume ${runId}: superseded by ${superseded.data.by}`);
|
|
143
152
|
const start = journal.read().find((e) => e.event === "run-start");
|
|
144
153
|
if (!start)
|
|
145
154
|
throw new Error(`journal for ${runId} has no run-start event`);
|
|
@@ -175,7 +184,10 @@ export async function runDaemon(repoRoot, opts = {}) {
|
|
|
175
184
|
baseRef = await gitHead(repoRoot);
|
|
176
185
|
baseline = await captureBaseline(repoRoot, commands);
|
|
177
186
|
writeFileSync(join(journal.dir, "baseline.json"), JSON.stringify(baseline, null, 2));
|
|
178
|
-
journal.append("run-start", undefined, { pid: process.pid, baseRef, commands, channels: channels.map(channelKey), branch, graphDefinitionHash: graphDefinitionHash(graph), mode: rm.mode.mode, modeSource: rm.source }); // graphDefinitionHash: T3 engagement identity (status+resume share it); pid: v1.13 (VIS-11) liveness; mode/modeSource: v1.51 T2
|
|
187
|
+
journal.append("run-start", undefined, { pid: process.pid, baseRef, commands, channels: channels.map(channelKey), branch, graphDefinitionHash: graphDefinitionHash(graph), mode: rm.mode.mode, modeSource: rm.source, ...(prior ? { supersedes: prior.runId } : {}) }); // graphDefinitionHash: T3 engagement identity (status+resume share it); pid: v1.13 (VIS-11) liveness; mode/modeSource: v1.51 T2; supersedes: v1.53 T5
|
|
188
|
+
// v1.53 T5: mark the prior run AFTER this run's run-start exists, so the prior journal never
|
|
189
|
+
// names a successor that has no journal. Append-only — the prior journal is never rewritten.
|
|
190
|
+
prior?.append("superseded", undefined, { by: runId });
|
|
179
191
|
}
|
|
180
192
|
// T6: open the narrator AFTER run-start/run-resume is journaled so the watch surface has a run to
|
|
181
193
|
// show. driver.narrator is undefined on subprocess → no-op (subprocess spawns nothing). Swallowed:
|
|
@@ -412,13 +424,18 @@ export async function runDaemon(repoRoot, opts = {}) {
|
|
|
412
424
|
}
|
|
413
425
|
// v1.29: consume the prior gate-failed session once. Same channel + known under-threshold context
|
|
414
426
|
// + adapter capability resumes; every other path is today's fresh dispatch.
|
|
427
|
+
// v1.53 T3: an adapter with no context surface at all (kimi, KIMI-03) may declare
|
|
428
|
+
// resumeUnknownContext to loosen ONLY the contextTokens-known requirement — a KNOWN
|
|
429
|
+
// over-threshold context still forces fresh, and the escalation ladder bounds the chain.
|
|
415
430
|
const priorSession = retrySession;
|
|
416
431
|
retrySession = undefined;
|
|
432
|
+
const retryAdapter = adapters.find((a) => a.id === assignment.adapter);
|
|
417
433
|
retryMode = priorSession
|
|
418
434
|
&& priorSession.channel === channelKey(assignment)
|
|
419
|
-
&& priorSession.contextTokens !== undefined
|
|
420
|
-
|
|
421
|
-
|
|
435
|
+
&& (priorSession.contextTokens !== undefined
|
|
436
|
+
? priorSession.contextTokens < cfg.contextWarnTokens
|
|
437
|
+
: retryAdapter?.resumeUnknownContext === true)
|
|
438
|
+
&& retryAdapter?.resumeCommand
|
|
422
439
|
? "resume"
|
|
423
440
|
: "fresh";
|
|
424
441
|
// v1.23 T3: over-threshold context still forces fresh at the retry boundary; never interrupt a
|
|
@@ -854,7 +871,9 @@ export async function runDaemon(repoRoot, opts = {}) {
|
|
|
854
871
|
break gateLoop;
|
|
855
872
|
}
|
|
856
873
|
gateFails++; // this attempt's gates failed — the one place quality degradation is verified (never inferred from attempts)
|
|
857
|
-
|
|
874
|
+
// v1.53 T3: prefer the CLI's own session id captured from this attempt's output (kimi's resume
|
|
875
|
+
// trailer) over the harness slot name; absent hook or no capture keeps today's slot-name id.
|
|
876
|
+
retrySession = { channel: channelKey(assignment), id: adapter.sessionIdFrom?.(output) ?? sessionId, contextTokens };
|
|
858
877
|
feedback = results.filter((g) => !g.pass).map((g) => `${g.gate}: ${g.details}`).join("\n\n");
|
|
859
878
|
const step = r.ladder[Math.min(ladderIdx++, r.ladder.length - 1)];
|
|
860
879
|
journal.append("escalation", t.id, { step, attempt: attempt + 1 });
|