@feigi/fleet-ctl 3.23.36 → 3.23.38
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/package.json +1 -1
- package/scripts/cell-readout.mjs +38 -11
- package/scripts/main-checkout.mjs +1 -1
- package/scripts/member-write-guard.mjs +1 -1
- package/scripts/repo-root.mjs +1 -1
- package/skills/run-team/references/reaping.md +1 -1
- package/scripts/agent-test-dir-prose.test.mjs +0 -59
- package/scripts/agent-test.test.mjs +0 -1640
- package/scripts/ambient-git-vars-mjs-prose.test.mjs +0 -397
- package/scripts/ambient-git-vars-prose.test.mjs +0 -192
- package/scripts/ancestry-check-position-prose.test.mjs +0 -175
- package/scripts/apply-ruleset.test.mjs +0 -421
- package/scripts/arg-header-probes-prose.test.mjs +0 -394
- package/scripts/arg.test.mjs +0 -1354
- package/scripts/board-cli.test.mjs +0 -710
- package/scripts/board-identity.test.mjs +0 -429
- package/scripts/board-liveness-banner.test.mjs +0 -119
- package/scripts/board-prev-shape.test.mjs +0 -228
- package/scripts/board-tryparse-null.test.mjs +0 -119
- package/scripts/board.test.mjs +0 -3121
- package/scripts/brief-after-cut-prose.test.mjs +0 -53
- package/scripts/candidates-exit3-prose.test.mjs +0 -302
- package/scripts/candidates.test.mjs +0 -2312
- package/scripts/caveat-unification-prose.test.mjs +0 -98
- package/scripts/cell-readout.test.mjs +0 -615
- package/scripts/ci-comment-rot-prose.test.mjs +0 -2025
- package/scripts/ci-completes-premise-prose.test.mjs +0 -257
- package/scripts/ci-state-prose.test.mjs +0 -113
- package/scripts/ci-state.test.mjs +0 -1933
- package/scripts/ci-vacuous-green.test.mjs +0 -368
- package/scripts/citation-sweep-prose.test.mjs +0 -1712
- package/scripts/claim-lifecycle.test.mjs +0 -187
- package/scripts/claim-ticket.test.mjs +0 -1122
- package/scripts/closing-keyword-prose.test.mjs +0 -193
- package/scripts/cockpit-launch-idempotent-prose.test.mjs +0 -165
- package/scripts/cockpit-tick-nonblocking.test.mjs +0 -210
- package/scripts/compute-board.test.mjs +0 -1214
- package/scripts/compute-spend.test.mjs +0 -517
- package/scripts/consumer-technology-prose.test.mjs +0 -126
- package/scripts/cross-repo-citation-prose.test.mjs +0 -155
- package/scripts/cwd-isolation-pins.mjs +0 -83
- package/scripts/dedupe-guard-prose.test.mjs +0 -243
- package/scripts/delete-merged-branch.test.mjs +0 -576
- package/scripts/derive-testcmd.test.mjs +0 -595
- package/scripts/diff-stats.test.mjs +0 -404
- package/scripts/dispatch-block-golden-prose.test.mjs +0 -1130
- package/scripts/dispatch-block-pins-prose.test.mjs +0 -615
- package/scripts/dispositions-check.test.mjs +0 -936
- package/scripts/dispositions-gate-prose.test.mjs +0 -129
- package/scripts/drop-merged-label.test.mjs +0 -479
- package/scripts/exec-stub.mjs +0 -93
- package/scripts/exec-stub.test.mjs +0 -69
- package/scripts/expected-jobs-prose.test.mjs +0 -80
- package/scripts/fan-out-authorization-prose.test.mjs +0 -80
- package/scripts/filing-bar-seat-prose.test.mjs +0 -242
- package/scripts/filing-claim-bar-prose.test.mjs +0 -166
- package/scripts/finisher-died-prose.test.mjs +0 -31
- package/scripts/finisher-dispatch-premise-prose.test.mjs +0 -312
- package/scripts/finisher-halt-prose.test.mjs +0 -106
- package/scripts/finisher-name-prose.test.mjs +0 -134
- package/scripts/finisher-own-tree-prose.test.mjs +0 -299
- package/scripts/finisher-pin-race-prose.test.mjs +0 -339
- package/scripts/finisher-test-verdict-prose.test.mjs +0 -102
- package/scripts/fix-applier-correction-rules-prose.test.mjs +0 -202
- package/scripts/fix-applier-result-file-prose.test.mjs +0 -149
- package/scripts/fleet-bootstrap.test.mjs +0 -75
- package/scripts/fleet-heartbeat.test.mjs +0 -450
- package/scripts/fleet-provenance.test.mjs +0 -138
- package/scripts/fleet-run.test.mjs +0 -90
- package/scripts/fleet-state.test.mjs +0 -460
- package/scripts/fleet-tick-prose.test.mjs +0 -300
- package/scripts/fleet-tick.test.mjs +0 -2444
- package/scripts/force-with-lease-reason-prose.test.mjs +0 -184
- package/scripts/frontmatter-check.test.mjs +0 -436
- package/scripts/git-env.test.mjs +0 -149
- package/scripts/git-status-untracked-mode-sweep.test.mjs +0 -216
- package/scripts/gone-walk.test.mjs +0 -213
- package/scripts/gp-sep-invariant-prose.test.mjs +0 -188
- package/scripts/head-moved-after-label-prose.test.mjs +0 -233
- package/scripts/immutable-body-claim-prose.test.mjs +0 -204
- package/scripts/implementer-model-tier.test.mjs +0 -235
- package/scripts/inflight-citation-prose.test.mjs +0 -144
- package/scripts/inflight-exit2-prose.test.mjs +0 -81
- package/scripts/inflight-probe-guard-prose.test.mjs +0 -109
- package/scripts/inflight-scan-scope-prose.test.mjs +0 -57
- package/scripts/inflight.test.mjs +0 -2882
- package/scripts/injection-control-prose.test.mjs +0 -404
- package/scripts/install-root-audit.test.mjs +0 -464
- package/scripts/instrument-check-prose.test.mjs +0 -235
- package/scripts/instruments.test.mjs +0 -787
- package/scripts/is-cli.test.mjs +0 -99
- package/scripts/issue-tracker-prose.test.mjs +0 -71
- package/scripts/json.test.mjs +0 -364
- package/scripts/ledger-dispatch.test.mjs +0 -927
- package/scripts/ledger-grammar.test.mjs +0 -227
- package/scripts/ledger-read-require-file.test.mjs +0 -197
- package/scripts/ledger-rotate.test.mjs +0 -142
- package/scripts/ledger.test.mjs +0 -3436
- package/scripts/line-distance-prose.test.mjs +0 -245
- package/scripts/liveness-rationale-prose.test.mjs +0 -83
- package/scripts/locale-pin-prose.test.mjs +0 -189
- package/scripts/main-checkout.test.mjs +0 -332
- package/scripts/member-outcome-prose.test.mjs +0 -167
- package/scripts/member-outcomes-header.test.mjs +0 -129
- package/scripts/member-outcomes-prose.test.mjs +0 -63
- package/scripts/member-outcomes.test.mjs +0 -638
- package/scripts/member-prompt-prose.test.mjs +0 -165
- package/scripts/member-record.test.mjs +0 -541
- package/scripts/member-write-guard.test.mjs +0 -214
- package/scripts/merge-commit-probe-prose.test.mjs +0 -122
- package/scripts/merge-gate.test.mjs +0 -613
- package/scripts/muted-git-guard-sweep.test.mjs +0 -493
- package/scripts/net-ssh-precedence-prose.test.mjs +0 -91
- package/scripts/net-watchdog-prose.test.mjs +0 -79
- package/scripts/net.test.mjs +0 -211
- package/scripts/no-ci-gate-prose.test.mjs +0 -108
- package/scripts/no-undo-audit.test.mjs +0 -3961
- package/scripts/no-wave.test.mjs +0 -138
- package/scripts/node-floor-sweep.test.mjs +0 -611
- package/scripts/pointer-target-prose.test.mjs +0 -186
- package/scripts/pr-cost.test.mjs +0 -521
- package/scripts/pr-overlap.test.mjs +0 -692
- package/scripts/printf-die-sweep.test.mjs +0 -384
- package/scripts/prompt-renderer.mjs +0 -40
- package/scripts/prompt-renderer.test.mjs +0 -68
- package/scripts/prose-pin.mjs +0 -807
- package/scripts/prose-pin.test.mjs +0 -1034
- package/scripts/prove-merge.test.mjs +0 -995
- package/scripts/pull-table-relabel-prose.test.mjs +0 -60
- package/scripts/quiet-payload-prose.test.mjs +0 -222
- package/scripts/reap-cherry-check-prose.test.mjs +0 -80
- package/scripts/reap.test.mjs +0 -4432
- package/scripts/reaping-prose.test.mjs +0 -239
- package/scripts/reaping-reap-authorizers-prose.test.mjs +0 -112
- package/scripts/reaping-release-authorizers-prose.test.mjs +0 -85
- package/scripts/recipe-derivation-prose.test.mjs +0 -112
- package/scripts/recipe-prove.test.mjs +0 -1042
- package/scripts/refuter-scratch-prose.test.mjs +0 -135
- package/scripts/release-label-required-prose.test.mjs +0 -71
- package/scripts/release-label-split-prose.test.mjs +0 -184
- package/scripts/release-ticket-reap-pairing-prose.test.mjs +0 -63
- package/scripts/release-ticket.test.mjs +0 -5099
- package/scripts/renovate-release-contract.test.mjs +0 -379
- package/scripts/repo-root.test.mjs +0 -873
- package/scripts/reporter-output-rule-prose.test.mjs +0 -208
- package/scripts/review-agent-report-only.test.mjs +0 -49
- package/scripts/review-core-booking.test.mjs +0 -35
- package/scripts/review-core-cwd-isolation.test.mjs +0 -374
- package/scripts/review-core-reads.test.mjs +0 -478
- package/scripts/review-core-refuter-scratch.test.mjs +0 -148
- package/scripts/review-core-scratch-refusal.test.mjs +0 -64
- package/scripts/review-core-shared-test-run.test.mjs +0 -275
- package/scripts/review-core-snapshot-path.test.mjs +0 -830
- package/scripts/review-core-specialist-scratch.test.mjs +0 -191
- package/scripts/review-core-testcmd.test.mjs +0 -335
- package/scripts/review-core-unrun.test.mjs +0 -566
- package/scripts/review-core-unverified-discriminant.test.mjs +0 -208
- package/scripts/review-eval-resolver-path.test.mjs +0 -55
- package/scripts/review-host-fixture.mjs +0 -89
- package/scripts/review-in-run-retry.test.mjs +0 -211
- package/scripts/review-no-vendor-dispatch.test.mjs +0 -45
- package/scripts/review-path-default.test.mjs +0 -796
- package/scripts/review-runner-spawns.test.mjs +0 -95
- package/scripts/review-runner.test.mjs +0 -166
- package/scripts/review-tier-audit.test.mjs +0 -140
- package/scripts/run-artifact-pr-prose.test.mjs +0 -341
- package/scripts/run-merge-bot-prose.test.mjs +0 -1413
- package/scripts/run-team-local-hook.test.mjs +0 -442
- package/scripts/scratch-partition-invariant-prose.test.mjs +0 -74
- package/scripts/scripts-path-citation-sweep.test.mjs +0 -73
- package/scripts/section-between-guard-prose.test.mjs +0 -379
- package/scripts/select-dimensions.test.mjs +0 -674
- package/scripts/shared-refusal.test.mjs +0 -388
- package/scripts/shell-traps-prose.test.mjs +0 -504
- package/scripts/shortlist.test.mjs +0 -380
- package/scripts/slow-transport.mjs +0 -72
- package/scripts/snapshot-repo.test.mjs +0 -1006
- package/scripts/snapshot-runner-audience-prose.test.mjs +0 -130
- package/scripts/spend-view.test.mjs +0 -312
- package/scripts/staleness-qualifier-prose.test.mjs +0 -251
- package/scripts/staleness-verdict-prose.test.mjs +0 -142
- package/scripts/staleness.test.mjs +0 -893
- package/scripts/stalled-member-liveness-prose.test.mjs +0 -219
- package/scripts/stash-prohibition-prose.test.mjs +0 -313
- package/scripts/strip-comments.mjs +0 -44
- package/scripts/temp-dir.mjs +0 -47
- package/scripts/ticket-router.test.mjs +0 -981
- package/scripts/tier-check.test.mjs +0 -828
- package/scripts/tier-outcomes-header.test.mjs +0 -518
- package/scripts/tier-outcomes.test.mjs +0 -572
- package/scripts/tier-roles.test.mjs +0 -535
- package/scripts/tracker-block-copy-prose.test.mjs +0 -288
- package/scripts/triage-labels-gloss-prose.test.mjs +0 -97
- package/scripts/unattended-git-sweep.test.mjs +0 -142
- package/scripts/verify-sha-exit2-prose.test.mjs +0 -269
- package/scripts/verify-sha.test.mjs +0 -692
- package/scripts/watcher-degraded-block.test.mjs +0 -369
- package/scripts/watcher-key-prose.test.mjs +0 -106
- package/scripts/within-run-pair-prose.test.mjs +0 -212
- package/scripts/worktree-audit.test.mjs +0 -1408
- package/scripts/worktree-head-prose.test.mjs +0 -276
- package/scripts/worktree-listing-sweep.test.mjs +0 -217
- package/scripts/worktree-model-prose.test.mjs +0 -216
- package/scripts/worktree-porcelain.mjs +0 -35
- package/scripts/worktree-porcelain.test.mjs +0 -102
|
@@ -1,191 +0,0 @@
|
|
|
1
|
-
// #1550. PR #1084 (implemented as #1530) added two directory-safety rules to
|
|
2
|
-
// the (since-retired) `review-pr` WORKFLOW's specialist prompt — chain the
|
|
3
|
-
// directory change into the command, and bracket a fixture's own git with
|
|
4
|
-
// `git rev-parse --show-toplevel` — mirroring what PR #1082 had already
|
|
5
|
-
// added to that same file's REFUTER prompt. `scripts/review-core.mjs`
|
|
6
|
-
// carried the same specialist dispatch prompt and, at the time, neither
|
|
7
|
-
// rule, while ordering the same kind of work: "Run any mutation or probe
|
|
8
|
-
// work inside your own copy of the snapshot." `review-core.mjs` is now the
|
|
9
|
-
// only surviving copy of this prompt.
|
|
10
|
-
//
|
|
11
|
-
// Naming the write area is not the same rule and does not close either failure
|
|
12
|
-
// mode. The specialist prompt already said scratch files go in its own
|
|
13
|
-
// directory "and nowhere else", and both measured failures happen with that
|
|
14
|
-
// sentence obeyed as written: a `cd` that silently fails leaves the following
|
|
15
|
-
// `;`-separated `git` running in whatever directory the agent was already in —
|
|
16
|
-
// the incident review-core-refuter-scratch.test.mjs records as commit
|
|
17
|
-
// 020d6ea reaching the checkout during the PR #488 fix-applier run — and an
|
|
18
|
-
// agent that BELIEVES it is already in its scratch copy and is wrong runs
|
|
19
|
-
// `git init` / `git commit` against the repository. A rule about where
|
|
20
|
-
// writes SHOULD go does not detect either.
|
|
21
|
-
//
|
|
22
|
-
// INLINED, NOT SHARED: #496's brief rules the shared-source route out — "the
|
|
23
|
-
// drift hazard is real and is its own ticket, not this one." That drift
|
|
24
|
-
// hazard is exactly what this pin exists to convert into a red: it failed to
|
|
25
|
-
// fire once already, which is why this ticket exists at all.
|
|
26
|
-
//
|
|
27
|
-
// WHY THESE PINS RENDER RATHER THAN GREP. Two reasons, and the second is the
|
|
28
|
-
// one a source-text pin cannot reach. The rules live inside a template
|
|
29
|
-
// literal, so their backticks are written escaped (`` \` ``); a pin written
|
|
30
|
-
// against the source has to tolerate that backslash, and then it passes just
|
|
31
|
-
// as happily on a prompt that shows the specialist a literal `\` instead of a
|
|
32
|
-
// code span. Rendering settles what the agent actually reads. And a bounded
|
|
33
|
-
// slice of the SOURCE would still be satisfied by dead text — the vacuity
|
|
34
|
-
// class strip-comments.mjs exists for — whereas a commented-out prompt does
|
|
35
|
-
// not render at all.
|
|
36
|
-
//
|
|
37
|
-
// THE SLICE IS THE PIN. Each rule appears more than once in this file (the
|
|
38
|
-
// verify/refuter prompt below carries its own, older copy), so every regex
|
|
39
|
-
// here would be satisfied by the refuter's copy alone if it were matched
|
|
40
|
-
// against the whole file — which is exactly the defect this ticket reports,
|
|
41
|
-
// so a pin that could pass on the refuter's copy would have passed before the
|
|
42
|
-
// fix. The extraction is bounded at both ends by the specialist `agent()`
|
|
43
|
-
// call's own template and its `review:` label, so the refuter prompt is
|
|
44
|
-
// outside it by construction. Both directions were measured: reverting the two
|
|
45
|
-
// rules out of the specialist prompt reds all three tests here with the
|
|
46
|
-
// refuter's copy untouched.
|
|
47
|
-
//
|
|
48
|
-
// THE CEILING, same as review-core-refuter-scratch.test.mjs's: nothing here
|
|
49
|
-
// reaches the agent's own obedience. These pins settle what a specialist is
|
|
50
|
-
// TOLD, never where it actually writes.
|
|
51
|
-
import { test } from "node:test";
|
|
52
|
-
import assert from "node:assert/strict";
|
|
53
|
-
import { readFileSync } from "node:fs";
|
|
54
|
-
import { join } from "node:path";
|
|
55
|
-
import { stripComments } from "./strip-comments.mjs";
|
|
56
|
-
import { between } from "./prose-pin.mjs";
|
|
57
|
-
|
|
58
|
-
const REPO = join(import.meta.dirname, "..");
|
|
59
|
-
const SOURCE = readFileSync(join(REPO, "scripts", "review-core.mjs"), "utf8");
|
|
60
|
-
const CODE = stripComments(SOURCE);
|
|
61
|
-
|
|
62
|
-
const TEMPLATE_START = "`Review PR #${pr} (branch ${branch}) for: ";
|
|
63
|
-
const TEMPLATE_END = "label: `review:";
|
|
64
|
-
|
|
65
|
-
// The template's free names, in the order `render` binds them. `sharedRunNote`,
|
|
66
|
-
// `readRules`, `usableDiff` and `environmentNote` are stubs: what they return
|
|
67
|
-
// is not under test here, and binding them keeps a render that stops passing
|
|
68
|
-
// one from dropping a paragraph silently — it throws a ReferenceError instead.
|
|
69
|
-
const SCOPE = ["pr", "branch", "d", "snap", "worktree", "stats", "sharedRunNote", "sharedRun", "failureOwner", "readRules", "usableDiff", "environmentNote"];
|
|
70
|
-
|
|
71
|
-
// `between` owns the bounded-slice extraction (and both failure messages);
|
|
72
|
-
// only the backtick trim below is specific to a template literal and stays
|
|
73
|
-
// here.
|
|
74
|
-
const SLICE = between(
|
|
75
|
-
CODE,
|
|
76
|
-
TEMPLATE_START,
|
|
77
|
-
TEMPLATE_END,
|
|
78
|
-
"review-core.mjs's specialist prompt (opening `Review PR #${pr} (branch ${branch}) for: `, labelled `review:`)",
|
|
79
|
-
);
|
|
80
|
-
|
|
81
|
-
const RENDER = new Function(...SCOPE, "return `" + SLICE.slice(1, SLICE.lastIndexOf("`")) + "`");
|
|
82
|
-
|
|
83
|
-
// One specialist's prompt. Every argument is fixed: these pins are about what
|
|
84
|
-
// the prompt SAYS, not about how it varies, so nothing here needs to.
|
|
85
|
-
const render = () =>
|
|
86
|
-
RENDER(
|
|
87
|
-
7,
|
|
88
|
-
"feature/x",
|
|
89
|
-
{ key: "correctness", prompt: "does it do what it says" },
|
|
90
|
-
{ path: "/scr/run-1/snapshot-abc1234", head: "abc1234", runRoot: "/scr/run-1" },
|
|
91
|
-
"/repo/.worktrees/7-x",
|
|
92
|
-
null,
|
|
93
|
-
() => "SHARED RUN NOTE",
|
|
94
|
-
{},
|
|
95
|
-
"correctness",
|
|
96
|
-
() => "READ RULES",
|
|
97
|
-
() => null,
|
|
98
|
-
() => "TEST ENVIRONMENT",
|
|
99
|
-
);
|
|
100
|
-
|
|
101
|
-
// The first of the two rules, in the prompt that was missing it. ONE
|
|
102
|
-
// contiguous regex rather than two assertions, for the reason
|
|
103
|
-
// refuter-scratch-prose.test.mjs states about the same wording: an unbounded
|
|
104
|
-
// gap lets a spliced sentence carve an exception INTO the rule and stay green.
|
|
105
|
-
// Both gaps are the literal punctuation the rendered prompt carries around
|
|
106
|
-
// "never" ("…`, " before it, a line break plus backtick after) rather than a
|
|
107
|
-
// free-text `.{0,N}` span — mutation-tested: an exception clause spliced into
|
|
108
|
-
// either gap (e.g. "never except during setup steps") reds this assertion.
|
|
109
|
-
//
|
|
110
|
-
// Both spellings are required, and the negative one is what carries the rule —
|
|
111
|
-
// `cd "$D" && git …` alone reads as a suggestion, and the observed failure is
|
|
112
|
-
// an agent writing the other form.
|
|
113
|
-
test("the rendered specialist prompt chains cd into the git command, never semicolon", () => {
|
|
114
|
-
assert.match(
|
|
115
|
-
render(),
|
|
116
|
-
/cd\s+"\$D"\s+&&\s+git\s+…`,\s+never\s+`cd\s+"\$D";\s+git\s+…/s,
|
|
117
|
-
"review-core.mjs's specialist prompt carries no cd-chaining rule — a silently failed `cd` leaves the following `git` " +
|
|
118
|
-
"running in the checkout, which is what produced commit 020d6ea during the PR #488 fix-applier run (#1550)",
|
|
119
|
-
);
|
|
120
|
-
});
|
|
121
|
-
|
|
122
|
-
// The second rule. Both halves, because they have OPPOSITE expected outcomes:
|
|
123
|
-
// a lone "resolves to your scratch path" guard is unsatisfiable before `git
|
|
124
|
-
// init` — a fresh scratch dir has no toplevel and exits 128 — and a guard that
|
|
125
|
-
// cannot pass on the clean path gets ignored. Naming a path alone does not
|
|
126
|
-
// catch the observed failure either, which was an agent BELIEVING it was
|
|
127
|
-
// already in scratch and being wrong. Gaps are the literal prose between the
|
|
128
|
-
// four clauses, typed out rather than left as a free-text `.{0,N}` span —
|
|
129
|
-
// mutation-tested: a short exception clause spliced into any of the first,
|
|
130
|
-
// third, or fourth gaps (e.g. "except in CI runners," right after the
|
|
131
|
-
// leading `:`, or "except during rebase," right after "not a failure;")
|
|
132
|
-
// reds this assertion; the same splices pass a `.{0,N}`-gapped version of
|
|
133
|
-
// this regex undetected.
|
|
134
|
-
//
|
|
135
|
-
// The realpath tail is part of the pin, not decoration on it, and carries the
|
|
136
|
-
// same typed gaps for the same reason. PR #1567 added that remedy to all four
|
|
137
|
-
// copies of this rule but the pin to only the refuter's two, so the clause sat
|
|
138
|
-
// unpinned in both specialist prompts: deleting it from both (measured, #1570)
|
|
139
|
-
// left this file and its review-pr sibling at 6/6 and the four refuter/prose
|
|
140
|
-
// pins at 41/41, all green. It is also the half that makes the rule
|
|
141
|
-
// satisfiable in the OTHER direction — `--show-toplevel` answers
|
|
142
|
-
// `/private/tmp/…` for a `/tmp` scratch dir on macOS, so a raw string compare
|
|
143
|
-
// reports a mismatch on the clean path, and a guard that cannot pass there
|
|
144
|
-
// gets ignored exactly as a lone `git init` half would.
|
|
145
|
-
//
|
|
146
|
-
// Both directions on the tail itself: an exception clause spliced into its
|
|
147
|
-
// leading gap ("…your scratch path, except in CI runners — compare resolved
|
|
148
|
-
// forms…") reds this assertion, where the `.{0,80}`-gapped spelling in
|
|
149
|
-
// `ed9f2967^:plugin/scripts/review-pr-specialist-scratch.test.mjs` let it
|
|
150
|
-
// through; rewrapping the clause to one line and to a narrower column stays
|
|
151
|
-
// green, since every gap here is `\s+` (all measured).
|
|
152
|
-
test("the rendered specialist prompt requires a toplevel assertion around git init/commit, including the realpath remedy for macOS's /private/tmp symlink", () => {
|
|
153
|
-
assert.match(
|
|
154
|
-
render(),
|
|
155
|
-
/`git\s+rev-parse\s+--show-toplevel`:\s+before\s+`git\s+init`\s+it\s+must\s+NOT\s+resolve\s+to\s+the\s+repository,\s+and\s+a\s+fresh\s+scratch\s+dir's\s+`fatal:\s+not\s+a\s+git\s+repository`\s+\(exit\s+128\)\s+is\s+the\s+pass,\s+not\s+a\s+failure;\s+before\s+any\s+`git\s+commit`\s+it\s+must\s+resolve\s+to\s+your\s+scratch\s+path\s+—\s+compare\s+resolved\s+forms\s+\(`realpath`\),\s+since\s+`--show-toplevel`\s+can\s+report\s+`\/private\/tmp\/…`\s+for\s+a\s+`\/tmp`\s+scratch\s+dir\s+on\s+macOS/s,
|
|
156
|
-
"review-core.mjs's specialist prompt carries no toplevel assertion around a fixture's own git init/commit, or " +
|
|
157
|
-
"dropped the realpath remedy for macOS's /private vs /tmp symlink — an agent that wrongly believes it is " +
|
|
158
|
-
"already in its scratch copy runs `git init`/`git commit` against the repository, and a raw compare against " +
|
|
159
|
-
"`--show-toplevel` reads a `/private/tmp` answer for a `/tmp` scratch dir as a mismatch on the clean path (#1550, #1570)",
|
|
160
|
-
);
|
|
161
|
-
});
|
|
162
|
-
|
|
163
|
-
// The other direction, and the reason a prose pin on an ADDED rule needs one: a
|
|
164
|
-
// prompt that gains rules can displace the instructions those rules exist to
|
|
165
|
-
// make safe, and both of the added rules are written in terms of the
|
|
166
|
-
// specialist's own scratch area — "your scratch path" names nothing once the
|
|
167
|
-
// directory sentence goes, and a rule about a fixture's git is inert if the
|
|
168
|
-
// prompt no longer orders mutation work in a copy of its own.
|
|
169
|
-
//
|
|
170
|
-
// The scratch sentence and the first rule are asserted as one bounded span,
|
|
171
|
-
// with the gap between them typed as `\s+` — the rendered prompt carries
|
|
172
|
-
// exactly one space there, never a free-text `.{0,N}` — so a sentence spliced
|
|
173
|
-
// between them (mutation-tested: "Except for git fixtures, which may live
|
|
174
|
-
// anywhere.") reds this assertion instead of leaving both halves present and
|
|
175
|
-
// passing.
|
|
176
|
-
test("the added rules still sit on a prompt that names a write area and orders work in its own copy", () => {
|
|
177
|
-
const prompt = render();
|
|
178
|
-
assert.match(
|
|
179
|
-
prompt,
|
|
180
|
-
/Run any mutation or probe work inside your own copy of the snapshot\./,
|
|
181
|
-
"review-core.mjs's specialist is no longer told to do mutation or probe work in a copy of its own — the two " +
|
|
182
|
-
"directory rules now guard work the prompt never scopes to a copy at all",
|
|
183
|
-
);
|
|
184
|
-
assert.match(
|
|
185
|
-
prompt,
|
|
186
|
-
/Scratch\s+files\s+go\s+in\s+\/scr\/run-1\/correctness\/\s+and\s+nowhere\s+else\.\s+Chain\s+the\s+directory\s+change\s+into\s+the\s+command/s,
|
|
187
|
-
"review-core.mjs's specialist prompt no longer names the scratch directory immediately before the cd rule — " +
|
|
188
|
-
"either the write area is gone, leaving 'your scratch path' naming nothing, or a sentence between them carves " +
|
|
189
|
-
"an exception into it",
|
|
190
|
-
);
|
|
191
|
-
});
|
|
@@ -1,335 +0,0 @@
|
|
|
1
|
-
import { test } from "node:test";
|
|
2
|
-
import assert from "node:assert/strict";
|
|
3
|
-
import { readFileSync } from "node:fs";
|
|
4
|
-
import { join } from "node:path";
|
|
5
|
-
import { stripComments } from "./strip-comments.mjs";
|
|
6
|
-
import { between } from "./prose-pin.mjs";
|
|
7
|
-
import { resolveTestCmd, SNAPSHOT_SCHEMA } from "./review-core.mjs";
|
|
8
|
-
|
|
9
|
-
// The review host used to default `testCmd` to a literal string naming THIS
|
|
10
|
-
// repo's own test path — silent green everywhere else, since `worktree` is a
|
|
11
|
-
// caller-supplied argument and a glob matching nothing exits 0 reporting
|
|
12
|
-
// `tests 0` (#142). It is now READ from the repo under review's Recipe cache by
|
|
13
|
-
// the snapshot agent (see the "derives this repository's own test command"
|
|
14
|
-
// paragraph below), through derive-testcmd.sh — see derive-testcmd.test.mjs
|
|
15
|
-
// for that script's own coverage, exercised against fixture repos that are
|
|
16
|
-
// NOT this one, which is the guard #142's acceptance criteria required and an
|
|
17
|
-
// archive of this repo could not have been.
|
|
18
|
-
//
|
|
19
|
-
// This file covers what remains review-core.mjs's own responsibility:
|
|
20
|
-
// resolveTestCmd's resolution order and refusal, the snapshot agent's prompt
|
|
21
|
-
// and schema actually carrying the derivation, and the shared test-run
|
|
22
|
-
// prompt (#2315) handing testCmd over verbatim with the 'tests 0' rule — the
|
|
23
|
-
// one prompt that runs it, since no specialist runs the full suite any more.
|
|
24
|
-
const REPO = join(import.meta.dirname, "..");
|
|
25
|
-
const SOURCE = readFileSync(join(REPO, "scripts", "review-core.mjs"), "utf8");
|
|
26
|
-
|
|
27
|
-
// Every pin below runs against CODE, not SOURCE — the same policy, and now the
|
|
28
|
-
// same stripper, as `review-core-reads.test.mjs`. Written against raw source,
|
|
29
|
-
// this file's schema-declaration pin was VACUOUS: wrapping the real
|
|
30
|
-
// `testCmd: { type: "string" },` in a `/* */` block left the suite 9 pass / 0
|
|
31
|
-
// fail while `additionalProperties: false` silently dropped the field at
|
|
32
|
-
// runtime, breaking the very derivation #142 adds. The `^\s*` anchor those
|
|
33
|
-
// pins use rejects a leading `//` and nothing else. Measured, #142 review.
|
|
34
|
-
const CODE = stripComments(SOURCE);
|
|
35
|
-
|
|
36
|
-
test("an explicit override always wins, whatever the snapshot derived", () => {
|
|
37
|
-
assert.equal(resolveTestCmd("npm test --", { testCmd: "node --test" }), "npm test --");
|
|
38
|
-
assert.equal(resolveTestCmd("npm test --", null), "npm test --");
|
|
39
|
-
assert.equal(resolveTestCmd("npm test --", { testCmdError: "refused" }), "npm test --");
|
|
40
|
-
});
|
|
41
|
-
|
|
42
|
-
test("no override falls back to the snapshot agent's derivation", () => {
|
|
43
|
-
assert.equal(resolveTestCmd(undefined, { testCmd: "node --test" }), "node --test");
|
|
44
|
-
assert.equal(resolveTestCmd(null, { testCmd: "npm test --" }), "npm test --");
|
|
45
|
-
});
|
|
46
|
-
|
|
47
|
-
// The whole point of #142: neither an override nor a derivation must REFUSE
|
|
48
|
-
// the review, never fall back to a guessed default. The reason passed through
|
|
49
|
-
// is derive-testcmd.sh's own, which names the derivation step (ADR 0015).
|
|
50
|
-
test("no override and no derivation refuses, naming the snapshot agent's reason", () => {
|
|
51
|
-
assert.throws(
|
|
52
|
-
() => resolveTestCmd(undefined, { testCmdError: "derive-testcmd: no Recipe cache at /r/.fleet/recipe.json" }),
|
|
53
|
-
/no test command for this repository — derive-testcmd: no Recipe cache at \/r\/\.fleet\/recipe\.json/,
|
|
54
|
-
);
|
|
55
|
-
});
|
|
56
|
-
|
|
57
|
-
// With no reason from the snapshot agent, the refusal still has to tell its
|
|
58
|
-
// reader what to RUN — the derivation step — rather than only that nothing
|
|
59
|
-
// was derived: a controller or reviewer reading it has no other pointer.
|
|
60
|
-
test("no override, no snapshot at all, and no testCmdError still refuses, naming the derivation step", () => {
|
|
61
|
-
for (const snap of [null, {}]) {
|
|
62
|
-
assert.throws(
|
|
63
|
-
() => resolveTestCmd(undefined, snap),
|
|
64
|
-
/no test command for this repository — the snapshot agent read no Recipe cache; run the Recipe derivation step \(run-team phase 0/,
|
|
65
|
-
);
|
|
66
|
-
}
|
|
67
|
-
});
|
|
68
|
-
|
|
69
|
-
// A falsy-but-present override (`""`) is caller error, not "no override" —
|
|
70
|
-
// the `||`/truthiness check above only distinguishes explicit from absent,
|
|
71
|
-
// and this pins that an empty string does NOT silently pass through as a
|
|
72
|
-
// command. It falls to the snapshot the same as `undefined` would.
|
|
73
|
-
test("an empty-string override is not treated as an explicit command", () => {
|
|
74
|
-
assert.equal(resolveTestCmd("", { testCmd: "node --test" }), "node --test");
|
|
75
|
-
});
|
|
76
|
-
|
|
77
|
-
// The function is worthless if nothing calls it — pin the CALL SITE, not
|
|
78
|
-
// just the lifted copy. select-dimensions.test.mjs's #118 regression is this
|
|
79
|
-
// exact defect: replacing the call with a bare default left every test above
|
|
80
|
-
// green while the feature disconnected.
|
|
81
|
-
test("runReview actually calls resolveTestCmd once the snapshot is validated", () => {
|
|
82
|
-
assert.match(
|
|
83
|
-
CODE,
|
|
84
|
-
/^\s*const testCmd = resolveTestCmd\(A\.testCmd, snap\);$/m,
|
|
85
|
-
"the testCmd call site changed — the derivation may be disconnected",
|
|
86
|
-
);
|
|
87
|
-
// Textually after the snapshot's own validity guard, not before — snap must
|
|
88
|
-
// be known good (or the throw above already fired) before this reads it.
|
|
89
|
-
// Anchored on the guard's INVOCATION — not its declaration, and not one of
|
|
90
|
-
// its return messages. The declaration sits above every top-level statement
|
|
91
|
-
// in this file, so ordering against it is near-tautological: measured, it
|
|
92
|
-
// stays green with the resolveTestCmd call moved ahead of the guard, which
|
|
93
|
-
// is the one regression this assertion's message names. A message anchor is
|
|
94
|
-
// reword-fragile instead — #539 split the single "no tree" message into
|
|
95
|
-
// three, and the retired literal leaves guardAt at -1, which reds this
|
|
96
|
-
// assertion loudly rather than silently, but reds it all the same. The call
|
|
97
|
-
// site is the only anchor that is both reword-proof and order-sensitive.
|
|
98
|
-
const guardAt = CODE.indexOf("const missingReason = snapshotMissing(snap, runRootPrefix);");
|
|
99
|
-
const callAt = CODE.indexOf("const testCmd = resolveTestCmd(");
|
|
100
|
-
assert.ok(guardAt !== -1 && callAt !== -1 && callAt > guardAt, "resolveTestCmd is called before snap is validated");
|
|
101
|
-
});
|
|
102
|
-
|
|
103
|
-
// A hardcoded default anywhere in this file is the exact regression #142
|
|
104
|
-
// fixes — `A.testCmd || "<literal>"` is the shape it used to take.
|
|
105
|
-
test("no hardcoded testCmd default remains", () => {
|
|
106
|
-
assert.doesNotMatch(
|
|
107
|
-
CODE,
|
|
108
|
-
/A\.testCmd\s*\|\|\s*"/,
|
|
109
|
-
"a literal testCmd default reappeared — it must come from resolveTestCmd/derive-testcmd.sh instead",
|
|
110
|
-
);
|
|
111
|
-
});
|
|
112
|
-
|
|
113
|
-
// The snapshot agent's return schema gains testCmd/testCmdError the same way
|
|
114
|
-
// diffPath/diffLines/prHead were added: `additionalProperties: false` drops
|
|
115
|
-
// an undeclared field silently, so BOTH the prompt asking for it and the
|
|
116
|
-
// schema declaring it have to be pinned, or the feature disconnects in one
|
|
117
|
-
// token exactly like `review-core-reads.test.mjs`'s "the snapshot agent asks for
|
|
118
|
-
// the diff facts AND declares them in its schema" records happening to the
|
|
119
|
-
// diff facts.
|
|
120
|
-
test("the snapshot agent is told to derive testCmd AND the schema declares it", () => {
|
|
121
|
-
const snapshot = between(CODE, "const snap = await agent(", "if (snap) {", "the snapshot agent dispatch");
|
|
122
|
-
|
|
123
|
-
assert.match(
|
|
124
|
-
snapshot,
|
|
125
|
-
/~\/\.fleet\/bin\/fleet-run derive-testcmd\.sh \$\{worktree\} test/,
|
|
126
|
-
"the snapshot agent no longer reads the Test entrypoint through derive-testcmd.sh via the Resolver",
|
|
127
|
-
);
|
|
128
|
-
// Deriving the command is half the job — it also has to RUN where the
|
|
129
|
-
// specialists are told to run it. `git archive` carries tracked files only,
|
|
130
|
-
// so `npm test --` (the derivation's first branch, for any repo whose
|
|
131
|
-
// scripts.test invokes a node_modules binary) exits 127 in the snapshot
|
|
132
|
-
// while passing in the worktree the derivation was run against. Measured on
|
|
133
|
-
// a synthetic repo during the #142 review: worktree exit 0, archive exit 127.
|
|
134
|
-
assert.match(
|
|
135
|
-
snapshot,
|
|
136
|
-
// `"$SNAP/node_modules"` since #1129: the destination is a per-run shell
|
|
137
|
-
// variable now, and the symlink has to land in the tree this run actually
|
|
138
|
-
// extracted — a link left at the old bare `${scratch}/snapshot` would
|
|
139
|
-
// provision node_modules for a directory no specialist is pointed at.
|
|
140
|
-
// Quote-tolerant on purpose. The needle's job is that the symlink lands in
|
|
141
|
-
// THIS run's destination — `$SNAP`, never the old bare `${scratch}/snapshot`
|
|
142
|
-
// — and `"$SNAP/node_modules"` and `"$SNAP"/node_modules` are the same word
|
|
143
|
-
// to any POSIX shell, `/node_modules` carrying no metacharacters. Pinning
|
|
144
|
-
// one of the two rejected a behaviour-identical rewrite as a #1129
|
|
145
|
-
// regression, which is how a pin teaches its next reader to weaken it.
|
|
146
|
-
//
|
|
147
|
-
// The `[ -n "$SNAP" ]` half is not decoration either: this is the only
|
|
148
|
-
// command in the block whose target no earlier line created, so it is the
|
|
149
|
-
// only one that would still act on an empty `$SNAP` — writing a symlink at
|
|
150
|
-
// `/node_modules`, outside the run root entirely.
|
|
151
|
-
// Order-agnostic between the two `[ ]` tests, for the same reason the
|
|
152
|
-
// quotes are optional: swapping them is behaviour-identical, so pinning
|
|
153
|
-
// one arrangement would red a correct change.
|
|
154
|
-
/if [^\n]*\[ -n "\$SNAP" \][^\n]*; then ln -s \$\{worktree\}\/node_modules "?\$SNAP"?\/node_modules"?/,
|
|
155
|
-
"the snapshot no longer provisions node_modules — a derived `npm test --` cannot run in it",
|
|
156
|
-
);
|
|
157
|
-
// These names live inside a template literal, so each backtick is a
|
|
158
|
-
// BACKSLASH-backtick in the source text — `\\?` matches it either way, the
|
|
159
|
-
// same idiom `review-core-reads.test.mjs` uses for diffPath/prHead/diffLines.
|
|
160
|
-
const B = "\\\\?`";
|
|
161
|
-
assert.match(
|
|
162
|
-
snapshot,
|
|
163
|
-
new RegExp(`Report\\s+${B}testCmd${B}\\s*=\\s*its\\s+stdout\\s+ONLY\\s+if\\s+it\\s+exited\\s+0`),
|
|
164
|
-
"testCmd is not bound to the script's stdout and gated on its exit code",
|
|
165
|
-
);
|
|
166
|
-
assert.match(
|
|
167
|
-
snapshot,
|
|
168
|
-
new RegExp(`report\\s+${B}testCmdError${B}\\s*=\\s*its\\s+stderr`),
|
|
169
|
-
"testCmdError is not bound to the script's stderr on refusal",
|
|
170
|
-
);
|
|
171
|
-
|
|
172
|
-
// Scoped to `SNAPSHOT_SCHEMA.properties` directly — a real import now,
|
|
173
|
-
// never a second copy of its JSON re-parsed from source text. A field
|
|
174
|
-
// declared ANYWHERE else is undeclared as far as
|
|
175
|
-
// `additionalProperties: false` is concerned.
|
|
176
|
-
for (const field of ["testCmd", "testCmdError"]) {
|
|
177
|
-
assert.ok(
|
|
178
|
-
Object.prototype.hasOwnProperty.call(SNAPSHOT_SCHEMA.properties, field),
|
|
179
|
-
`${field} is not declared in SNAPSHOT_SCHEMA.properties — additionalProperties:false drops it`,
|
|
180
|
-
);
|
|
181
|
-
}
|
|
182
|
-
// Both stay OUT of `required`: a network or git failure inside
|
|
183
|
-
// derive-testcmd.sh must not abort a snapshot that is otherwise good —
|
|
184
|
-
// resolveTestCmd is what turns a missing derivation into a refusal, not a
|
|
185
|
-
// required-field validation error one layer down. `pathVerified` joins
|
|
186
|
-
// path+head instead (#140): unlike testCmd, its absence must abort. So does
|
|
187
|
-
// `repoVerified` (#1056) — whether the snapshot is a repository holding the
|
|
188
|
-
// reviewed tree decides what a suite run in there is evidence about.
|
|
189
|
-
// `environmentNote`'s strict `=== true` check already guarantees an omitted
|
|
190
|
-
// field reads as UNVERIFIED either way, so the risk `required` guards
|
|
191
|
-
// against is different: omitted from `required`, the snapshot agent could
|
|
192
|
-
// silently drop the field from its structured output — especially under
|
|
193
|
-
// omp's permissive schema-retry-exhaustion mode — and the caller would
|
|
194
|
-
// never find out the measurement was never taken.
|
|
195
|
-
assert.deepEqual(
|
|
196
|
-
SNAPSHOT_SCHEMA.required,
|
|
197
|
-
["runRoot", "path", "head", "pathVerified", "repoVerified"],
|
|
198
|
-
"required must stay path+head+pathVerified+repoVerified only",
|
|
199
|
-
);
|
|
200
|
-
});
|
|
201
|
-
|
|
202
|
-
// derive-testcmd.sh's exit-3 refusals are about the environment and name no
|
|
203
|
-
// derivation step, so the prompt must not tell the snapshot agent that every
|
|
204
|
-
// refusal does.
|
|
205
|
-
test("the snapshot agent is told a cache refusal names the derivation step and an exit-3 refusal names an environment fault", () => {
|
|
206
|
-
const snapshot = between(CODE, "const snap = await agent(", "if (snap) {", "the snapshot agent dispatch");
|
|
207
|
-
assert.match(snapshot, /a\s+refusal\s+about\s+the\s+cache\s+names\s+the\s+step\s+that\s+derives\s+it/);
|
|
208
|
-
assert.match(snapshot, /an\s+exit-3\s+refusal\s+names\s+an\s+environment\s+fault,\s+and\s+the\s+caller\s+fixes\s+that\s+instead\s+of\s+deriving/);
|
|
209
|
-
// The two clauses above are the only places the prompt may speak of a
|
|
210
|
-
// refusal: any other sentence about one, however worded or capitalised, is a
|
|
211
|
-
// fresh chance to say that every refusal names the derivation step.
|
|
212
|
-
const outside = snapshot
|
|
213
|
-
.replace(/a\s+refusal\s+about\s+the\s+cache\s+names\s+the\s+step\s+that\s+derives\s+it/, "")
|
|
214
|
-
.replace(/an\s+exit-3\s+refusal\s+names\s+an\s+environment\s+fault/, "");
|
|
215
|
-
assert.doesNotMatch(outside, /\brefusals?\b/i, "the prompt speaks of a refusal outside the two clauses that say which kind names the derivation step");
|
|
216
|
-
});
|
|
217
|
-
|
|
218
|
-
// #2315. The prompt that RUNS the command is the shared test run's — every
|
|
219
|
-
// specialist used to run it, which made one review up to six full sweeps.
|
|
220
|
-
// Bounded at both ends: an unbounded slice runs to EOF, where the specialist
|
|
221
|
-
// and refuter prompts could satisfy the assertions below instead.
|
|
222
|
-
function testRunPrompt() {
|
|
223
|
-
return between(CODE, "`Run this repository's test command ONCE", "{ label: `test-run${forPr}`", "the shared test-run prompt");
|
|
224
|
-
}
|
|
225
|
-
|
|
226
|
-
// The prompt is what the test-run agent actually obeys, so the reading rule has
|
|
227
|
-
// to be in it. Without this, an agent that guesses a glob still reports
|
|
228
|
-
// `tests 0` as a pass — the exact failure the derivation alone does not cover.
|
|
229
|
-
test("the test-run prompt hands the command over verbatim and rules 'tests 0' a failure", () => {
|
|
230
|
-
const prompt = testRunPrompt();
|
|
231
|
-
// The command block must INTERPOLATE testCmd, not restate it. A hardcoded
|
|
232
|
-
// copy drifts from the resolved value the moment either one changes, and a
|
|
233
|
-
// caller passing args.testCmd (or the derivation) would be handed the wrong
|
|
234
|
-
// command. Run from the snapshot's root, its output captured to the log
|
|
235
|
-
// every specialist is pointed at.
|
|
236
|
-
assert.match(
|
|
237
|
-
prompt,
|
|
238
|
-
/\{ cd "\$\{snap\.path\}" && \$\{testCmd\}; \} > "\$\{logPath\}" 2>&1; echo "TEST_RUN_EXIT=\$\?"/,
|
|
239
|
-
"the prompt no longer runs the interpolated command verbatim from the snapshot's root into the log",
|
|
240
|
-
);
|
|
241
|
-
// Match the ruling itself, not the phrase `tests 0` — that appears in the
|
|
242
|
-
// surrounding explanation too, so a looser assertion passes with the rule
|
|
243
|
-
// deleted (observed: it did).
|
|
244
|
-
assert.match(prompt, /'tests 0' is a FAILED run/, "the prompt no longer rules a zero-test run a failure");
|
|
245
|
-
// The RULING was pinned and the INSTRUCTION was not, so the sentence telling
|
|
246
|
-
// the agent not to swap the command out could be deleted with this suite
|
|
247
|
-
// green (#143). Everything around it explains why; this is the only clause
|
|
248
|
-
// that actually forbids anything.
|
|
249
|
-
assert.match(
|
|
250
|
-
prompt,
|
|
251
|
-
/Do not substitute a command of your own\./,
|
|
252
|
-
"the prompt no longer forbids substituting a command — only explains why one would be wrong",
|
|
253
|
-
);
|
|
254
|
-
// Once is the whole point of #2315: a retry by the agent is a second full run.
|
|
255
|
-
// The clause, not the phrase — "Run it exactly once, then again when it
|
|
256
|
-
// fails" still contains "Run it exactly once".
|
|
257
|
-
assert.match(
|
|
258
|
-
prompt,
|
|
259
|
-
/Run it exactly once: not again when it fails, and not again when it hits the\s+deadline\./,
|
|
260
|
-
"the prompt no longer forbids running the command a second time",
|
|
261
|
-
);
|
|
262
|
-
});
|
|
263
|
-
|
|
264
|
-
// How the one run is held and read. Each clause is a behavior the agent obeys
|
|
265
|
-
// and nothing downstream re-checks: a backgrounded or polled run is the
|
|
266
|
-
// unbounded load #2315 measured, a deadline that is far too short reads every
|
|
267
|
-
// heavy suite as no counts, a `tail` too short misses the summary, and a 0
|
|
268
|
-
// typed in for a count the log never stated turns "no counts" (every dimension
|
|
269
|
-
// unrun) into "tests 0" or a clean-looking pass — the distinction
|
|
270
|
-
// TEST_RUN_SCHEMA's absent-count rule exists for.
|
|
271
|
-
test("the test-run prompt holds the run in the foreground under a deadline, and never invents a count", () => {
|
|
272
|
-
const prompt = testRunPrompt();
|
|
273
|
-
assert.match(prompt, /ONE blocking foreground command — never backgrounded, never polled —/, "the prompt no longer forbids a backgrounded or polled run");
|
|
274
|
-
assert.match(prompt, /with a command deadline of 1800 seconds:/, "the prompt's command deadline changed — size it against a full suite under fleet load");
|
|
275
|
-
assert.match(prompt, /\\`tail -n 40 "\$\{logPath\}"\\`/, "the prompt no longer reads enough of the log's tail to hold a runner's summary");
|
|
276
|
-
assert.match(
|
|
277
|
-
prompt,
|
|
278
|
-
/omit every count and say what happened in \\`error\\`\. Never\s+write 0 for a count the log does not state/,
|
|
279
|
-
"the prompt no longer forbids writing 0 for a count the log does not state",
|
|
280
|
-
);
|
|
281
|
-
});
|
|
282
|
-
|
|
283
|
-
// #143's second correction. The rationale once asserted, present tense and as
|
|
284
|
-
// fact about the run being described, that a bare runner "tears down a shared
|
|
285
|
-
// container mid-run for every sibling". Measured against this repo:
|
|
286
|
-
// this repo's Test entrypoint is a plain `node --test` run and
|
|
287
|
-
// `commands/review-and-fix.md` records that this repo has no
|
|
288
|
-
// compose file, no `globalSetup`, and no vitest — so no teardown can happen,
|
|
289
|
-
// and an agent that checks the reason it was given finds it false.
|
|
290
|
-
//
|
|
291
|
-
// The rule is still worth carrying, because the review host reviews repos that DO
|
|
292
|
-
// have a stack. It has to be stated as a conditional about those repos rather
|
|
293
|
-
// than as a fact about this run.
|
|
294
|
-
test("the anti-substitution rationale is portable, not a present-tense claim about this run", () => {
|
|
295
|
-
const prompt = testRunPrompt();
|
|
296
|
-
assert.doesNotMatch(
|
|
297
|
-
prompt,
|
|
298
|
-
/tears down a shared container mid-run for every sibling/,
|
|
299
|
-
"the prompt again asserts a container teardown as fact about a run that cannot have one (#143)",
|
|
300
|
-
);
|
|
301
|
-
// The conditional that replaced it, and the half that is NOT conditional: a
|
|
302
|
-
// zero-match glob exits 0 in every repo, so hedging that one would weaken a
|
|
303
|
-
// rule this repo has actually measured (#142).
|
|
304
|
-
assert.match(prompt, /In a repo that has a shared test stack/, "the container rationale is no longer scoped to the repos it can happen in");
|
|
305
|
-
assert.match(prompt, /a guessed glob is\s+worse in every repo/, "the zero-match-glob half is no longer stated as holding everywhere");
|
|
306
|
-
});
|
|
307
|
-
|
|
308
|
-
// The rule #143 widened, now in the one prompt that runs the suite. The
|
|
309
|
-
// classifier catches the zero-pass case on its own (`review-core-unrun.test.mjs`),
|
|
310
|
-
// but the partial-tree case it CANNOT: `unrunReason` is pure and never learns
|
|
311
|
-
// how many tests the whole tree has, so the only guard is the run itself
|
|
312
|
-
// starting from the snapshot's root. If this instruction goes, that case has no
|
|
313
|
-
// other guard.
|
|
314
|
-
test("the test-run prompt rules a no-work run too: zero passes, and a count below the whole tree", () => {
|
|
315
|
-
const prompt = testRunPrompt();
|
|
316
|
-
assert.match(prompt, /0 passes with no failures is\s+everything skipped/, "the prompt no longer rules an all-skipped run a no-work run (#143)");
|
|
317
|
-
assert.match(
|
|
318
|
-
prompt,
|
|
319
|
-
/a count well below what the whole tree reports, which means it ran a\s+PARTIAL copy/,
|
|
320
|
-
"the prompt no longer names a partial-tree run — the case nothing downstream can catch (#143)",
|
|
321
|
-
);
|
|
322
|
-
});
|
|
323
|
-
|
|
324
|
-
// #2315's other half: the specialist prompt must hand the shared run over and
|
|
325
|
-
// never the command to run. A specialist prompt that interpolates `testCmd`
|
|
326
|
-
// again is six full sweeps per review again.
|
|
327
|
-
test("the specialist prompt carries the shared run's note, and never the command to run", () => {
|
|
328
|
-
const prompt = between(CODE, "READ ONLY FROM THE SNAPSHOT", "Scratch files go in", "the specialist prompt");
|
|
329
|
-
assert.doesNotMatch(prompt, /\$\{testCmd\}/, "the specialist prompt hands out testCmd to run again — one full sweep per dimension");
|
|
330
|
-
assert.match(
|
|
331
|
-
prompt,
|
|
332
|
-
/\$\{sharedRunNote\(sharedRun, failureOwner, d\.key\)\}/,
|
|
333
|
-
"the specialist prompt no longer carries the shared run's note",
|
|
334
|
-
);
|
|
335
|
-
});
|