@feigi/fleet-ctl 3.110.2 → 3.110.4

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
@@ -5,7 +5,7 @@ model: "@slow:high"
5
5
  spawns: task
6
6
  ---
7
7
 
8
- You are a fleet fix-applier. Your brief names this dispatch's worktree, PR branch and test command, and the kind of work it is: a review's findings to apply or defer, a dispositions retry, a conflict hold to clear, or a red CI check to fix. Every brief names the scratch directory too, which your report file goes under; a conflict hold's and a red-check fix's name it for that file alone, so every other rule below that cites `<scratch>` is for a review and a dispositions retry. The rules about findings — scope and refuters, the apply/defer split, the dispositions record and the report of the split and the deferral issues — are for a review and a dispositions retry: a conflict hold and a red-check fix have no review file, and what each says to report is in its brief. Every other rule below holds for every kind. The review file and review path that those rules cite are the ones your brief names. Where the rules write `<abs-path>`, `<branch>`, `<testCmd>`, `<scratch>`, `<pr>` or `<run>`, they name the value your brief gives you; a prompt you write for a refuter carries the value itself, never the placeholder.
8
+ You are a fleet fix-applier. Your brief names this dispatch's worktree, PR branch and test command, and the kind of work it is: a review's findings to apply or defer, a dispositions retry, a conflict hold to clear, or a red CI check to fix. Every brief names the scratch directory too, which your report file goes under; a conflict hold's and a red-check fix's name it for that file alone, so every other rule below that cites `<scratch>` is for a review and a dispositions retry. The rules about findings — scope and refuters, the apply/defer split, the dispositions record and the report of the split and the deferral issues — are for a review and a dispositions retry: a conflict hold and a red-check fix have no review file and write no dispositions record, so their report's `dispositionsRecord` is `null`, and what else each says to report is in its brief. Every other rule below holds for every kind. The review file and review path that those rules cite are the ones your brief names. Where the rules write `<abs-path>`, `<branch>`, `<testCmd>`, `<scratch>`, `<pr>` or `<run>`, they name the value your brief gives you; a prompt you write for a refuter carries the value itself, never the placeholder.
9
9
 
10
10
  **`edit` and `read` resolve a bare relative path against the SESSION ROOT —
11
11
  the controller's own main checkout — not against that worktree and not
@@ -332,8 +332,10 @@ Then write your report to the file your brief names,
332
332
  number on its `deferred` entry, the deferral issues as `{issue, subject}`
333
333
  entries — `filed` for every issue step 5 created, and `unrecorded` for one
334
334
  whose `ledger.mjs filed` failed twice, which the controller's settle records —
335
- `dispositionsRecord` your record's path and `testRun` your test result. Then
336
- send the one line your brief gives, `<your name>: <outcome> <path>`, and exit.
335
+ `dispositionsRecord` your record's path and `testRun` your test result, `null`
336
+ when no test run exists — a `no-op`, or a `failed` that never ran tests; a
337
+ non-null `testRun` always reflects a real run. Then send the one line your
338
+ brief gives, `<your name>: <outcome> <path>`, and exit.
337
339
  **Deferring everything is a normal
338
340
  outcome, not a stall:** nothing is then staged, `git commit` refuses an empty
339
341
  index, `~/.fleet/bin/fleet-run push-sync.mjs --branch <branch>` prints git's
package/package.json CHANGED
@@ -1,6 +1,6 @@
1
1
  {
2
2
  "name": "@feigi/fleet-ctl",
3
- "version": "3.110.2",
3
+ "version": "3.110.4",
4
4
  "description": "Agent fleet: run-team controller, merge bot, PR reviewer, and the ticket pipeline they share",
5
5
  "license": "Apache-2.0",
6
6
  "repository": {
@@ -1,7 +1,7 @@
1
1
  #!/usr/bin/env node
2
- // The run's two metrics duties that keep the router's inputs current: move the
3
- // pending ticket-features rows into the tracked corpus, then refresh the cost
4
- // guard over it.
2
+ // The run's three metrics duties that keep the router's inputs current: move the
3
+ // pending ticket-features rows into the tracked corpus, widen a member-outcomes
4
+ // corpus written before its latest columns, then refresh the cost guard over it.
5
5
  //
6
6
  // metrics-refresh.mjs [--repo <dir>]
7
7
  //
@@ -24,6 +24,20 @@
24
24
  // rule, but phase 0 refuses a second one at its ledger rotation, which runs
25
25
  // after its drain step, so a second controller's drain is not guarded.
26
26
  //
27
+ // MEMBER-OUTCOMES WIDEN. `member-outcomes.mjs` refuses a row that is not exactly
28
+ // as wide as its current columns, so a consumer's `docs/metrics/member-outcomes.tsv`
29
+ // written before `ending`, `dispatch_model` or `work_s` shipped (20, 21 or 22
30
+ // columns; the file has no header row, so width is all that tells) could not be
31
+ // read by phase 0, the close-out scrape or `tier-outcomes.mjs append`. This
32
+ // appends the missing columns to every row, so every consumer widens at its next
33
+ // run start. Cells already there and the leading comments stay as written; an
34
+ // added cell is blank (unknown), except `ending` on an implementer's row, which
35
+ // is `unknown`, because blank there means "not an implementer". A file at the
36
+ // current width is not touched, so a second run changes nothing. Rows of any
37
+ // other width, rows of mixed widths and git conflict markers fail the duty,
38
+ // naming the file, and the file is left as it was; the cost guard after it does
39
+ // not run.
40
+ //
27
41
  // COST GUARD. `pr-cost.mjs --guard` rewrites `.fleet/cost-guard.json`, which
28
42
  // the router reads and fails closed without. It needs all three metrics TSVs;
29
43
  // a workspace that lacks one has no corpus to guard yet (its first run), so the
@@ -32,8 +46,8 @@
32
46
  // (no verdict yet — the baseline cell has too few merged PRs) are verdicts the
33
47
  // router and `fleet-tick.mjs` consume, not failures; any other non-zero exit is.
34
48
  //
35
- // One `metrics-refresh: <duty>: …` line per duty on stdout. Exit 0 when
36
- // neither duty failed; 1 when one failed — named on stderr, and the duty after
49
+ // One `metrics-refresh: <duty>: …` line per duty on stdout. Exit 0 when no
50
+ // duty failed; 1 when one failed — named on stderr, and the duties after
37
51
  // it did not run (each is safe to run again); 2 on a usage error.
38
52
 
39
53
  import { spawnSync } from "node:child_process";
@@ -44,13 +58,15 @@ import { makeDie, defineFlags } from "./arg.mjs";
44
58
  import { gitEnv } from "./git-env.mjs";
45
59
  import { FleetDirUnresolvable, fleetFile } from "./fleet-dir.mjs";
46
60
  import { isCLI } from "./is-cli.mjs";
61
+ import { COLUMNS as MEMBER_COLUMNS, isImplementer } from "./member-outcomes.mjs";
47
62
  import { FEATURE_COLUMNS, parseFeatures } from "./pull-attribution.mjs";
48
63
 
49
64
  const NAME = "metrics-refresh";
50
65
  const PR_COST = fileURLToPath(new URL("./pr-cost.mjs", import.meta.url));
51
66
  const PENDING = join(".fleet", "ticket-features.pending.tsv");
52
67
  const FEATURES = join("docs", "metrics", "ticket-features.tsv");
53
- const CORPUS = [join("docs", "metrics", "member-outcomes.tsv"), join("docs", "metrics", "tier-outcomes.tsv"), FEATURES];
68
+ const MEMBERS = join("docs", "metrics", "member-outcomes.tsv");
69
+ const CORPUS = [MEMBERS, join("docs", "metrics", "tier-outcomes.tsv"), FEATURES];
54
70
  const GUARD = join(".fleet", "cost-guard.json");
55
71
  // pr-cost.mjs --guard: 0 no cell tripped, 3 a cell tripped, 4 no verdict yet —
56
72
  // all three are verdicts its readers consume; anything else is a failure.
@@ -72,6 +88,62 @@ function widen(text, columns) {
72
88
  }).join("\n");
73
89
  }
74
90
 
91
+ // The first width a member-outcomes.tsv was ever written at that a later column widened: the
92
+ // width before `ending` shipped (merged 2026-10-22). Every width from it up to the current
93
+ // one is a file of this corpus written before the columns past that width existed.
94
+ const MEMBER_FIRST_WIDTH = MEMBER_COLUMNS.indexOf("ending");
95
+ const MEMBER_NAME_AT = MEMBER_COLUMNS.indexOf("member");
96
+
97
+ // `text`, a member-outcomes TSV at an earlier width, at the current one, with the columns it
98
+ // lacks appended to every row. The file has no header row, so a row's width is the only thing
99
+ // that says which columns it holds; leading comments and every cell already there are kept.
100
+ // An appended cell is blank, which that file's own rule reads as unknown — except `ending` on
101
+ // an implementer's row, where blank means "not an implementer": that reads `unknown`. Rows of
102
+ // one width other than the current or an earlier one, and a file whose rows are not all of
103
+ // one width, throw: parseTsv's refusal to guess a row's columns stands, and this does not
104
+ // widen what it would not read. A row ending in a CR (a CRLF file) gains its cells before that
105
+ // CR, which stays the row's last byte. A file with no rows comes back as `{ rows: 0 }`, and one
106
+ // already at the current width as null: neither has anything to write.
107
+ function widenMembers(text) {
108
+ const lines = text.split("\n");
109
+ const rows = [];
110
+ lines.forEach((l, i) => { if (l.trim() && !l.startsWith("#")) rows.push(i); });
111
+ if (!rows.length) return { rows: 0 };
112
+ const markerAt = rows.find((i) => /^(<{7}|={7}|>{7})( |\r?$)/.test(lines[i]));
113
+ if (markerAt !== undefined) throw new Error(`git conflict marker — ${lines[markerAt].slice(0, 60)}`);
114
+ const widths = [...new Set(rows.map((i) => lines[i].split("\t").length))].sort((a, b) => a - b);
115
+ if (widths.length > 1) throw new Error(`rows of different widths (${widths.join(", ")}) — a file is written at one width`);
116
+ const [width] = widths;
117
+ if (width === MEMBER_COLUMNS.length) return null;
118
+ if (width < MEMBER_FIRST_WIDTH || width > MEMBER_COLUMNS.length) {
119
+ const row = lines[rows[0]];
120
+ throw new Error(`malformed row: ${width} fields, expected ${MEMBER_COLUMNS.length} (or an earlier ${MEMBER_FIRST_WIDTH} to ${MEMBER_COLUMNS.length - 1}) — ${row.slice(0, 60)}`);
121
+ }
122
+ const added = MEMBER_COLUMNS.slice(width);
123
+ for (const i of rows) {
124
+ const eol = lines[i].endsWith("\r") ? "\r" : "";
125
+ const cells = lines[i].slice(0, lines[i].length - eol.length).split("\t");
126
+ lines[i] = [...cells, ...added.map((c) => (c === "ending" && isImplementer(cells[MEMBER_NAME_AT]) ? "unknown" : ""))].join("\t") + eol;
127
+ }
128
+ return { text: lines.join("\n"), width, rows: rows.length };
129
+ }
130
+
131
+ function widenMemberOutcomes(repo) {
132
+ const file = join(repo, MEMBERS);
133
+ if (!existsSync(file)) return say(`member-outcomes: skipped, ${MEMBERS} not found — no corpus to widen yet`);
134
+ let widened;
135
+ try {
136
+ widened = widenMembers(readFileSync(file, "utf8"));
137
+ } catch (e) {
138
+ throw new Error(`${MEMBERS}: ${e.message}`);
139
+ }
140
+ if (widened === null) return say(`member-outcomes: already ${MEMBER_COLUMNS.length} columns`);
141
+ if (widened.rows === 0) return say("member-outcomes: no rows to widen");
142
+ writeFileSync(`${file}.tmp`, widened.text);
143
+ renameSync(`${file}.tmp`, file);
144
+ say(`member-outcomes: widened ${widened.rows} row${widened.rows === 1 ? "" : "s"} from ${widened.width} to ${MEMBER_COLUMNS.length} columns`);
145
+ }
146
+
75
147
  function drain(repo) {
76
148
  const pending = join(repo, PENDING);
77
149
  if (!existsSync(pending)) return say(`drain: nothing pending in ${PENDING}`);
@@ -140,7 +212,7 @@ function main() {
140
212
  die(`${e.message} — pass --repo <dir>`);
141
213
  }
142
214
  }
143
- for (const [name, duty] of [["drain", drain], ["cost guard", costGuard]]) {
215
+ for (const [name, duty] of [["drain", drain], ["member-outcomes", widenMemberOutcomes], ["cost guard", costGuard]]) {
144
216
  try {
145
217
  duty(repo);
146
218
  } catch (e) {
@@ -26,7 +26,12 @@
26
26
  // is no bail`, gives no reason in `bailDetail`, or names a `pr` or a
27
27
  // `head`, as `$.pr: <pr> is not null, with outcome "blocked" — a block
28
28
  // reports no PR`, or whose `bailed` outcome names a `pr` or a `head`, as
29
- // `$.pr: <pr> is not null, with outcome "bailed" — a bail reports no PR`
29
+ // `$.pr: <pr> is not null, with outcome "bailed" — a bail reports no PR`,
30
+ // and a fix-pr report whose `applied` outcome names no `testRun`, as
31
+ // `$.testRun: null, with outcome "applied" — an applied fix reports the
32
+ // test run behind it`, or that defers a `refuted` finding with no
33
+ // `dispositionsRecord`, as `$.dispositionsRecord: null, with a refuted
34
+ // deferral at $.deferred[<i>] — settle reads its reason from that record`
30
35
  // 2 usage: no or unknown subcommand, a wrong argument count, an unknown
31
36
  // family, any flag; or <file> could not be read; or print could not
32
37
  // write the schema to stdout. Refused on stderr, nothing on stdout
@@ -141,8 +146,10 @@ export const SCHEMAS = {
141
146
  deferred: listOf({ bucket: str, index: int, issue: id }),
142
147
  filed: issues,
143
148
  unrecorded: issues,
144
- dispositionsRecord: { type: "string", pattern: "^/" },
145
- testRun: obj({ cmd: str, exit: int, tests: int, cancelled: int }),
149
+ // `null` where nothing exists: a conflict hold or a red-check fix writes no
150
+ // dispositions record, and a no-op or a failed fix may have run no tests.
151
+ dispositionsRecord: { type: ["string", "null"], pattern: "^/" },
152
+ testRun: obj({ cmd: str, exit: int, tests: int, cancelled: int }, ["object", "null"]),
146
153
  confirmation: strings,
147
154
  }),
148
155
  finisher: report({
@@ -188,9 +195,12 @@ export function mergeBotVerdicts(report) {
188
195
  // text that is not only whitespace, and names no `pr` or `head`, since a block
189
196
  // reports no PR; a `bailed` outcome names no `pr` or `head` either, since a
190
197
  // bail is the pre-code outcome — a PR the member opened is outcome `pr`, which
191
- // settle checks. For a merge-bot report: each PR gets one verdict — merged,
192
- // held, conflict-held or refused — the last it reached in the pass, so a PR
193
- // named twice, in one list or in two, is a violation at its second entry.
198
+ // settle checks. For a fix-pr report: an `applied` outcome names its
199
+ // `testRun`, the run behind the commit, and a `refuted` deferral, under any
200
+ // outcome, names the `dispositionsRecord` settle reads its reason from. For a
201
+ // merge-bot report: each PR gets one verdict — merged, held, conflict-held or
202
+ // refused — the last it reached in the pass, so a PR named twice, in one list
203
+ // or in two, is a violation at its second entry.
194
204
  export function reportViolation(family, value) {
195
205
  const found = validate(SCHEMAS[family], value);
196
206
  if (found !== null) return found;
@@ -206,6 +216,15 @@ export function reportViolation(family, value) {
206
216
  if (value[field] !== null) return `$.${field}: ${show(value[field])} is not null, with outcome "bailed" — a bail reports no PR`;
207
217
  }
208
218
  }
219
+ if (family === "fix-pr") {
220
+ if (value.outcome === "applied" && value.testRun === null) {
221
+ return '$.testRun: null, with outcome "applied" — an applied fix reports the test run behind it';
222
+ }
223
+ const refuted = value.deferred.findIndex((d) => d.bucket === "refuted");
224
+ if (refuted !== -1 && value.dispositionsRecord === null) {
225
+ return `$.dispositionsRecord: null, with a refuted deferral at $.deferred[${refuted}] — settle reads its reason from that record`;
226
+ }
227
+ }
209
228
  if (family !== "merge-bot") return null;
210
229
  const first = new Map();
211
230
  for (const { pr, path } of mergeBotVerdicts(value)) {
@@ -136,8 +136,12 @@ recorded nothing — so run it again with `timeout: 0` on the call.
136
136
 
137
137
  Your report is a file, `<scratch>/reports/<your name>.json`, `<your name>` the
138
138
  `fix-pr-` name you were dispatched as, in the shape
139
- `~/.fleet/bin/fleet-run report-schema.mjs print fix-pr` prints. Write it
140
- before your final message, and send nothing until
139
+ `~/.fleet/bin/fleet-run report-schema.mjs print fix-pr` prints. A field with
140
+ nothing real to point at is `null`: `dispositionsRecord` for a conflict hold or
141
+ a red-check fix, which write no dispositions record, and `testRun` when no test
142
+ run exists — a `no-op`, or a `failed` that never ran tests. A non-null value
143
+ always reflects a real record or a real run. Write the report before your
144
+ final message, and send nothing until
141
145
  `~/.fleet/bin/fleet-run report-schema.mjs validate fix-pr <scratch>/reports/<your name>.json`
142
146
  answers `valid`. Your final message is then one line,
143
147
  `<your name>: <outcome> <scratch>/reports/<your name>.json`, `<outcome>` the
@@ -106,11 +106,13 @@ moves `.fleet/ticket-features.pending.tsv` — the row the router script
106
106
  appended for each Pull at dispatch — into `docs/metrics/ticket-features.tsv`,
107
107
  deduped on `session`+`agent`, so this run's own rows are in the file that gets
108
108
  committed; that file is a run artifact too and is committed with the other
109
- two. It then runs `pr-cost.mjs --guard`, which rewrites
110
- `.fleet/cost-guard.json` from all three files. Exit 0: both ran. The guard's
109
+ two. It then widens a `docs/metrics/member-outcomes.tsv` written before the
110
+ newest columns, as at phase 0 — a file already at the current width is left
111
+ as it is — and then runs `pr-cost.mjs --guard`, which rewrites
112
+ `.fleet/cost-guard.json` from all three files. Exit 0: every duty ran. The guard's
111
113
  own exit 3 (a cell tripped) and 4 (no verdict yet, baseline too thin) are
112
114
  verdicts the router and `fleet-tick.mjs` read, never failures; the script
113
- absorbs them. Exit 1 names the duty that failed on stderr and the duty after it
115
+ absorbs them. Exit 1 names the duty that failed on stderr and the duties after it
114
116
  did not run; exit 2 is a usage error. At close-out a failure fails the
115
117
  `metrics` step of `closeout.mjs finish`, so `commit-and-pr` stays pending: fix
116
118
  the cause and run `finish` again — the pending file stays in `.fleet/` and the
@@ -162,7 +162,11 @@ script absorbs them. A workspace that lacks one of the three metrics TSVs
162
162
  has no corpus to guard yet — its first run: the script prints a line naming
163
163
  the missing file, skips the guard and exits 0. Exit 1 or 2 stops the run
164
164
  before its first Pull: report the line the script printed on stderr, fix
165
- the cause and run it again — each duty is safe to repeat.
165
+ the cause and run it again — each duty is safe to repeat. Between the drain
166
+ and the guard the script also widens a `docs/metrics/member-outcomes.tsv`
167
+ written before the newest columns: each row gains them, `ending` reads
168
+ `unknown` on an implementer's row, and nothing else changes. Without that,
169
+ phase 0, the close-out scrape and `tier-outcomes.mjs append` refuse the file.
166
170
 
167
171
  ## Launch the cockpit
168
172