@feigi/fleet-ctl 0.0.0
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/.claude-plugin/plugin.json +9 -0
- package/LICENSE +202 -0
- package/NOTICE +18 -0
- package/agents/fleet-finisher.agent.md +7 -0
- package/agents/fleet-implementer-alt.agent.md +210 -0
- package/agents/fleet-implementer.agent.md +210 -0
- package/agents/fleet-merge-bot.agent.md +7 -0
- package/agents/fleet-review-comments.agent.md +40 -0
- package/agents/fleet-review-correctness.agent.md +43 -0
- package/agents/fleet-review-runner.agent.md +73 -0
- package/agents/fleet-review-silent-failure.agent.md +49 -0
- package/agents/fleet-review-simplify.agent.md +46 -0
- package/agents/fleet-review-snapshot.agent.md +24 -0
- package/agents/fleet-review-tests.agent.md +46 -0
- package/agents/fleet-review-types.agent.md +41 -0
- package/agents/fleet-review-verifier.agent.md +29 -0
- package/commands/review-and-fix.md +98 -0
- package/commands/run-merge-bot.md +357 -0
- package/commands/run-team-help.md +17 -0
- package/package.json +23 -0
- package/scripts/agent-test-dir-prose.test.mjs +59 -0
- package/scripts/agent-test.test.mjs +1636 -0
- package/scripts/ambient-git-vars-mjs-prose.test.mjs +354 -0
- package/scripts/ambient-git-vars-prose.test.mjs +192 -0
- package/scripts/ancestry-check-position-prose.test.mjs +175 -0
- package/scripts/apply-ruleset.test.mjs +420 -0
- package/scripts/arg-header-probes-prose.test.mjs +393 -0
- package/scripts/arg.mjs +600 -0
- package/scripts/arg.test.mjs +1354 -0
- package/scripts/board-cli.test.mjs +694 -0
- package/scripts/board-identity.test.mjs +429 -0
- package/scripts/board-liveness-banner.test.mjs +119 -0
- package/scripts/board-prev-shape.test.mjs +229 -0
- package/scripts/board-tryparse-null.test.mjs +119 -0
- package/scripts/board.html +471 -0
- package/scripts/board.mjs +2037 -0
- package/scripts/board.test.mjs +3086 -0
- package/scripts/brief-after-cut-prose.test.mjs +53 -0
- package/scripts/candidates-exit3-prose.test.mjs +302 -0
- package/scripts/candidates.mjs +674 -0
- package/scripts/candidates.test.mjs +2101 -0
- package/scripts/caveat-unification-prose.test.mjs +98 -0
- package/scripts/ci-comment-rot-prose.test.mjs +2025 -0
- package/scripts/ci-completes-premise-prose.test.mjs +257 -0
- package/scripts/ci-state-prose.test.mjs +113 -0
- package/scripts/ci-state.mjs +761 -0
- package/scripts/ci-state.test.mjs +1640 -0
- package/scripts/ci-vacuous-green.test.mjs +368 -0
- package/scripts/citation-sweep-prose.test.mjs +1572 -0
- package/scripts/claim-lifecycle.test.mjs +185 -0
- package/scripts/claim-ticket.sh +417 -0
- package/scripts/claim-ticket.test.mjs +1121 -0
- package/scripts/closing-keyword-prose.test.mjs +193 -0
- package/scripts/cockpit-launch-idempotent-prose.test.mjs +165 -0
- package/scripts/cockpit-tick-nonblocking.test.mjs +210 -0
- package/scripts/compute-board.mjs +579 -0
- package/scripts/compute-board.test.mjs +1060 -0
- package/scripts/compute-spend.mjs +329 -0
- package/scripts/compute-spend.test.mjs +451 -0
- package/scripts/cross-repo-citation-prose.test.mjs +156 -0
- package/scripts/cwd-isolation-pins.mjs +83 -0
- package/scripts/dedupe-guard-prose.test.mjs +243 -0
- package/scripts/derive-testcmd.sh +239 -0
- package/scripts/derive-testcmd.test.mjs +380 -0
- package/scripts/diff-stats.mjs +231 -0
- package/scripts/diff-stats.test.mjs +408 -0
- package/scripts/dispatch-block-golden-prose.test.mjs +1024 -0
- package/scripts/dispatch-block-pins-prose.test.mjs +615 -0
- package/scripts/drop-merged-label.sh +173 -0
- package/scripts/drop-merged-label.test.mjs +480 -0
- package/scripts/expected-jobs-prose.test.mjs +80 -0
- package/scripts/filing-bar-seat-prose.test.mjs +242 -0
- package/scripts/filing-claim-bar-prose.test.mjs +166 -0
- package/scripts/finisher-dispatch-premise-prose.test.mjs +312 -0
- package/scripts/finisher-halt-prose.test.mjs +106 -0
- package/scripts/finisher-name-prose.test.mjs +134 -0
- package/scripts/finisher-own-tree-prose.test.mjs +285 -0
- package/scripts/finisher-pin-race-prose.test.mjs +339 -0
- package/scripts/fix-applier-correction-rules-prose.test.mjs +201 -0
- package/scripts/fix-applier-result-file-prose.test.mjs +100 -0
- package/scripts/fleet-bootstrap +155 -0
- package/scripts/fleet-heartbeat.mjs +272 -0
- package/scripts/fleet-heartbeat.test.mjs +450 -0
- package/scripts/fleet-provenance +207 -0
- package/scripts/fleet-provenance.test.mjs +64 -0
- package/scripts/fleet-run +148 -0
- package/scripts/fleet-run.test.mjs +90 -0
- package/scripts/fleet-state.mjs +400 -0
- package/scripts/fleet-state.test.mjs +373 -0
- package/scripts/fleet-tick-prose.test.mjs +299 -0
- package/scripts/fleet-tick.mjs +832 -0
- package/scripts/fleet-tick.test.mjs +1332 -0
- package/scripts/force-with-lease-reason-prose.test.mjs +184 -0
- package/scripts/frontmatter-allowlist.json +64 -0
- package/scripts/frontmatter-check.mjs +256 -0
- package/scripts/frontmatter-check.test.mjs +436 -0
- package/scripts/git-env.mjs +124 -0
- package/scripts/git-env.test.mjs +149 -0
- package/scripts/git-status-untracked-mode-sweep.test.mjs +216 -0
- package/scripts/gone-walk.test.mjs +213 -0
- package/scripts/gp-sep-invariant-prose.test.mjs +188 -0
- package/scripts/head-moved-after-label-prose.test.mjs +233 -0
- package/scripts/immutable-body-claim-prose.test.mjs +413 -0
- package/scripts/implementer-model-tier.test.mjs +643 -0
- package/scripts/inflight-citation-prose.test.mjs +144 -0
- package/scripts/inflight-exit2-prose.test.mjs +81 -0
- package/scripts/inflight-probe-guard-prose.test.mjs +109 -0
- package/scripts/inflight-scan-scope-prose.test.mjs +57 -0
- package/scripts/inflight.sh +1106 -0
- package/scripts/inflight.test.mjs +2984 -0
- package/scripts/injection-control-prose.test.mjs +404 -0
- package/scripts/install-root-audit.test.mjs +464 -0
- package/scripts/instrument-check-prose.test.mjs +235 -0
- package/scripts/instruments.sh +394 -0
- package/scripts/instruments.test.mjs +787 -0
- package/scripts/issue-tracker-prose.test.mjs +71 -0
- package/scripts/json.sh +248 -0
- package/scripts/json.test.mjs +364 -0
- package/scripts/ledger-dispatch.test.mjs +516 -0
- package/scripts/ledger-grammar.mjs +85 -0
- package/scripts/ledger-grammar.test.mjs +142 -0
- package/scripts/ledger-read-require-file.test.mjs +197 -0
- package/scripts/ledger.mjs +1753 -0
- package/scripts/ledger.test.mjs +3554 -0
- package/scripts/line-distance-prose.test.mjs +245 -0
- package/scripts/liveness-rationale-prose.test.mjs +83 -0
- package/scripts/locale-pin-prose.test.mjs +189 -0
- package/scripts/member-outcome-prose.test.mjs +167 -0
- package/scripts/member-outcomes-header.test.mjs +148 -0
- package/scripts/member-outcomes-prose.test.mjs +63 -0
- package/scripts/member-outcomes.mjs +231 -0
- package/scripts/member-outcomes.test.mjs +507 -0
- package/scripts/member-prompt-prose.test.mjs +161 -0
- package/scripts/member-record.mjs +605 -0
- package/scripts/member-record.test.mjs +516 -0
- package/scripts/member-write-guard.mjs +279 -0
- package/scripts/member-write-guard.test.mjs +211 -0
- package/scripts/merge-commit-probe-prose.test.mjs +122 -0
- package/scripts/merge-gate.mjs +262 -0
- package/scripts/merge-gate.test.mjs +613 -0
- package/scripts/muted-git-guard-sweep.test.mjs +493 -0
- package/scripts/net-ssh-precedence-prose.test.mjs +91 -0
- package/scripts/net-watchdog-prose.test.mjs +79 -0
- package/scripts/net.sh +409 -0
- package/scripts/net.test.mjs +211 -0
- package/scripts/no-ci-gate-prose.test.mjs +108 -0
- package/scripts/no-undo-audit.sh +1134 -0
- package/scripts/no-undo-audit.test.mjs +3965 -0
- package/scripts/no-wave.test.mjs +138 -0
- package/scripts/node-floor-sweep.test.mjs +611 -0
- package/scripts/pointer-target-prose.test.mjs +186 -0
- package/scripts/pr-overlap.mjs +417 -0
- package/scripts/pr-overlap.test.mjs +695 -0
- package/scripts/printf-die-sweep.test.mjs +384 -0
- package/scripts/prompt-renderer.mjs +39 -0
- package/scripts/prompt-renderer.test.mjs +68 -0
- package/scripts/prose-pin.mjs +808 -0
- package/scripts/prose-pin.test.mjs +1034 -0
- package/scripts/prove-merge.sh +228 -0
- package/scripts/prove-merge.test.mjs +995 -0
- package/scripts/pull-table-relabel-prose.test.mjs +60 -0
- package/scripts/quiet-payload-prose.test.mjs +222 -0
- package/scripts/reap.sh +1768 -0
- package/scripts/reap.test.mjs +4395 -0
- package/scripts/reaping-prose.test.mjs +159 -0
- package/scripts/refuter-scratch-prose.test.mjs +135 -0
- package/scripts/release-label-required-prose.test.mjs +71 -0
- package/scripts/release-label-split-prose.test.mjs +169 -0
- package/scripts/release-ticket.sh +1716 -0
- package/scripts/release-ticket.test.mjs +5106 -0
- package/scripts/renovate-release-contract.test.mjs +379 -0
- package/scripts/repo-root.mjs +544 -0
- package/scripts/repo-root.test.mjs +873 -0
- package/scripts/reporter-output-rule-prose.test.mjs +208 -0
- package/scripts/review-agent-report-only.test.mjs +49 -0
- package/scripts/review-core-cwd-isolation.test.mjs +374 -0
- package/scripts/review-core-reads.test.mjs +478 -0
- package/scripts/review-core-refuter-scratch.test.mjs +148 -0
- package/scripts/review-core-snapshot-path.test.mjs +777 -0
- package/scripts/review-core-specialist-scratch.test.mjs +189 -0
- package/scripts/review-core-testcmd.test.mjs +288 -0
- package/scripts/review-core-unrun.test.mjs +530 -0
- package/scripts/review-core-unverified-discriminant.test.mjs +208 -0
- package/scripts/review-core.mjs +1028 -0
- package/scripts/review-eval-resolver-path.test.mjs +55 -0
- package/scripts/review-eval.mjs +197 -0
- package/scripts/review-host-fixture.mjs +73 -0
- package/scripts/review-in-run-retry.test.mjs +187 -0
- package/scripts/review-no-vendor-dispatch.test.mjs +45 -0
- package/scripts/review-path-default.test.mjs +796 -0
- package/scripts/review-runner-spawns.test.mjs +93 -0
- package/scripts/review-runner.test.mjs +155 -0
- package/scripts/review-tier-audit.test.mjs +140 -0
- package/scripts/run-artifact-pr-prose.test.mjs +341 -0
- package/scripts/run-merge-bot-prose.test.mjs +1391 -0
- package/scripts/scratch-partition-invariant-prose.test.mjs +74 -0
- package/scripts/scripts-path-citation-sweep.test.mjs +73 -0
- package/scripts/section-between-guard-prose.test.mjs +378 -0
- package/scripts/select-dimensions.test.mjs +674 -0
- package/scripts/shared-refusal.test.mjs +388 -0
- package/scripts/shell-traps-prose.test.mjs +507 -0
- package/scripts/shortlist.mjs +355 -0
- package/scripts/shortlist.test.mjs +366 -0
- package/scripts/slow-transport.mjs +72 -0
- package/scripts/snapshot-repo.test.mjs +772 -0
- package/scripts/snapshot-runner-audience-prose.test.mjs +129 -0
- package/scripts/spend-view.test.mjs +312 -0
- package/scripts/staleness-qualifier-prose.test.mjs +251 -0
- package/scripts/staleness-verdict-prose.test.mjs +142 -0
- package/scripts/staleness.mjs +414 -0
- package/scripts/staleness.test.mjs +893 -0
- package/scripts/stalled-member-liveness-prose.test.mjs +219 -0
- package/scripts/stash-prohibition-prose.test.mjs +313 -0
- package/scripts/strip-comments.mjs +44 -0
- package/scripts/tier-check.mjs +468 -0
- package/scripts/tier-check.test.mjs +726 -0
- package/scripts/tier-guard-gate-prose.test.mjs +216 -0
- package/scripts/tier-outcomes-header.test.mjs +586 -0
- package/scripts/tier-outcomes.mjs +441 -0
- package/scripts/tier-outcomes.test.mjs +342 -0
- package/scripts/tier-roles.mjs +250 -0
- package/scripts/tier-roles.test.mjs +316 -0
- package/scripts/tracker-block-copy-prose.test.mjs +288 -0
- package/scripts/triage-labels-gloss-prose.test.mjs +97 -0
- package/scripts/unattended-git-sweep.test.mjs +142 -0
- package/scripts/verify-sha-exit2-prose.test.mjs +269 -0
- package/scripts/verify-sha.sh +117 -0
- package/scripts/verify-sha.test.mjs +693 -0
- package/scripts/watcher-degraded-block.test.mjs +372 -0
- package/scripts/watcher-key-prose.test.mjs +106 -0
- package/scripts/within-run-pair-prose.test.mjs +208 -0
- package/scripts/worktree-audit.sh +482 -0
- package/scripts/worktree-audit.test.mjs +1407 -0
- package/scripts/worktree-head-prose.test.mjs +276 -0
- package/scripts/worktree-listing-sweep.test.mjs +217 -0
- package/scripts/worktree-model-prose.test.mjs +216 -0
- package/scripts/worktree.sh +470 -0
- package/skills/next-ticket/SKILL.md +138 -0
- package/skills/run-team/SKILL.md +3603 -0
- package/skills/run-team/references/ci-and-staleness.md +75 -0
- package/skills/run-team/references/correction-tickets.md +37 -0
- package/skills/run-team/references/isolation.md +23 -0
- package/skills/run-team/references/member-lifecycle.md +73 -0
- package/skills/run-team/references/reaping.md +73 -0
- package/skills/sizing-a-ticket/SKILL.md +39 -0
|
@@ -0,0 +1,605 @@
|
|
|
1
|
+
// The member-telemetry adapter (#1342, ruled on #1302). ONE per-member
|
|
2
|
+
// record shape, TWO readers a tree walk chooses by content, never by
|
|
3
|
+
// caller-supplied config: board.mjs (the live spend panel) and
|
|
4
|
+
// member-outcomes.mjs (the scraper) build on the primitives here rather than
|
|
5
|
+
// each inlining a transcript layout, so the fold-back arithmetic, the cwd
|
|
6
|
+
// encoder and the ticket/PR extraction each live in exactly one place.
|
|
7
|
+
//
|
|
8
|
+
// The record: harness, session, role, agent, model, thinking,
|
|
9
|
+
// subagent_type, tokens_in, tokens_cache_create, tokens_cache_read,
|
|
10
|
+
// tokens_out, cost, wall_s, turns, ticket, pr. `harness` is set by whichever
|
|
11
|
+
// reader produced the row — structural, decided by which root the transcript
|
|
12
|
+
// lives under, before any byte is parsed. `cost` is real (`usage.cost.total`
|
|
13
|
+
// — no pricing table exists in this repo, and inventing one is not this
|
|
14
|
+
// module's business.
|
|
15
|
+
//
|
|
16
|
+
// `thinking` is the harness-written level, blank when a hole is visible
|
|
17
|
+
// (#1302's ruling on "auto"). `subagent_type` (#1066) is what the member was
|
|
18
|
+
// DISPATCHED AS — the agent DEFINITION the dispatch named, never the
|
|
19
|
+
// member's own name: it is the only field that separates a deliberate
|
|
20
|
+
// alternate-tier pair from two members whose models already happened to
|
|
21
|
+
// differ, and both readers keep it because it is derived rather than
|
|
22
|
+
// authored and survives a regeneration. `""` is NOT the hole use; it means
|
|
23
|
+
// the dispatch named no agent definition, the ordinary shape of an untyped
|
|
24
|
+
// `task` call.
|
|
25
|
+
import { readdirSync, readFileSync, realpathSync } from "node:fs";
|
|
26
|
+
import { join, basename, relative, isAbsolute, sep } from "node:path";
|
|
27
|
+
|
|
28
|
+
import { classifyRole, CANONICAL_MEMBER_NAME_PREFIXES } from "./compute-spend.mjs";
|
|
29
|
+
|
|
30
|
+
// ---------------------------------------------------------------------------
|
|
31
|
+
// cwd encoder
|
|
32
|
+
// ---------------------------------------------------------------------------
|
|
33
|
+
|
|
34
|
+
// `realpath` is an injectable seam, not a design nicety: the production
|
|
35
|
+
// default resolves the LIVE filesystem, but a directory a transcript reads
|
|
36
|
+
// may no longer exist by the time anything reads it — a scratch dir cleaned
|
|
37
|
+
// up weeks later, say — and `fs.realpathSync` throws ENOENT on a path it
|
|
38
|
+
// cannot see. Verifying the encoder against a historical transcript, for one
|
|
39
|
+
// callers who need to encode a cwd that is not guaranteed to exist — every
|
|
40
|
+
// caller but board.mjs's own live panel — pass their own resolver.
|
|
41
|
+
//
|
|
42
|
+
// Encodes the way the fleet's own tooling reads it back: every
|
|
43
|
+
// non-alphanumeric character with `-`, so `-Users-x-claude` decodes to
|
|
44
|
+
// `/Users/x/.claude`, not `-Users-x.claude`. Home-relative paths become
|
|
45
|
+
// `~/...`-style (`-` under $HOME) joined by `-` with DOTS PRESERVED
|
|
46
|
+
// (`.claude` -> `-.claude`); non-home paths are realpath-resolved (so `/tmp/x`,
|
|
47
|
+
// a symlink to `/private/tmp/x` on macOS, encodes under the resolved name)
|
|
48
|
+
// and double-dash-wrapped. Both forms are measured against real
|
|
49
|
+
// `~/.omp/agent/sessions/*` directory names in member-record.test.mjs - a
|
|
50
|
+
// `~/dev/fleet-plugin`, `--private-tmp-fx685-scratch--` all exist on disk
|
|
51
|
+
// today.
|
|
52
|
+
export function encodeProjectDir(cwd, { home = process.env.HOME, realpath = realpathSync } = {}) {
|
|
53
|
+
const rel = relative(home, cwd);
|
|
54
|
+
const isHome = !rel.startsWith("..") && !isAbsolute(rel);
|
|
55
|
+
if (isHome) return "-" + rel.split(sep).filter(Boolean).join("-");
|
|
56
|
+
const resolved = realpath(cwd);
|
|
57
|
+
return "--" + resolved.split(sep).filter(Boolean).join("-") + "--";
|
|
58
|
+
}
|
|
59
|
+
|
|
60
|
+
// ---------------------------------------------------------------------------
|
|
61
|
+
// Ticket/PR extraction and model normalisation
|
|
62
|
+
// ---------------------------------------------------------------------------
|
|
63
|
+
// Moved here from member-outcomes.mjs (which re-exports both for the callers
|
|
64
|
+
// and tests that already import them from there) so member-record.mjs, the
|
|
65
|
+
// lower-level module, does not depend upward on either script it feeds.
|
|
66
|
+
// ---------------------------------------------------------------------------
|
|
67
|
+
|
|
68
|
+
// `<synthetic>` is not a model — it is the harness labelling a turn it
|
|
69
|
+
// generated itself, and mapping it to anything would invent a data point.
|
|
70
|
+
//
|
|
71
|
+
// The `[1m]` strip is DEFENSIVE, not load-bearing: it guards a
|
|
72
|
+
// context-window variant spelling (`claude-opus-5[1m]`) this scraper has
|
|
73
|
+
// never actually read from a transcript's `message.model`, where measurement
|
|
74
|
+
// found ZERO bracketed spellings across every file. It fires only if the
|
|
75
|
+
// model spelling ever changes to carry one.
|
|
76
|
+
export function normalizeModel(raw) {
|
|
77
|
+
const s = String(raw ?? "").trim();
|
|
78
|
+
if (!s || s === "<synthetic>") return null;
|
|
79
|
+
return s.replace(/\[[^\]]*\]$/, "");
|
|
80
|
+
}
|
|
81
|
+
|
|
82
|
+
// A member's name is the only place its unit of work is recorded — there is
|
|
83
|
+
// no dispatch sidecar at all (see ompMemberRecord). The one exception is a
|
|
84
|
+
// PR the member OPENED (#2209): an `impl-<ticket>` name can never carry it,
|
|
85
|
+
// so ompMemberRecord reads it off the member's own `gh pr create` result
|
|
86
|
+
// (createdPrNumbers) — this function still answers `pr: ""` for that name.
|
|
87
|
+
//
|
|
88
|
+
// FOUR finisher spellings are live on disk, measured 2026-08-27 across every
|
|
89
|
+
// transcript: finisher-pr-<n>, finish-pr-<n>, finisher-<n>, finish-<n>. All
|
|
90
|
+
// four book a PR, and matching only the first cost 120 of 283 finisher
|
|
91
|
+
// members their join key to tier-outcomes.tsv. The fix-pr-<n> and
|
|
92
|
+
// review-pr-<n> families share the first pattern only because the infix is
|
|
93
|
+
// the same — they are NOT finisher spellings. `finisher-pr-<n>` is the
|
|
94
|
+
// canonical name run-team now fixes (#326); the other three stay matched
|
|
95
|
+
// because the runs that used them are already in the record.
|
|
96
|
+
//
|
|
97
|
+
// merge-bot-<n> is deliberately excluded: its number is a per-run dispatch
|
|
98
|
+
// counter, never a PR, and booking it as a pr would join the row to an
|
|
99
|
+
// unrelated PR's verdict. A single trailing lowercase letter is a retry
|
|
100
|
+
// suffix (-b, -c and -d all observed) and is stripped first, because a
|
|
101
|
+
// re-dispatched member works the same unit. A trailing `-v<n>` (`-v2`,
|
|
102
|
+
// `-v10`, ...) is a DIFFERENT spelling of the same re-dispatch, used when a
|
|
103
|
+
// controller re-dispatches a finisher/reviewer against a PR whose head moved
|
|
104
|
+
// after label (#1482, measured: ~67% of one finisher's tokens fell through
|
|
105
|
+
// to a blank pr column under the old letter-only regex); it is stripped for
|
|
106
|
+
// the same reason, not because it looks like a second-ticket suffix — no
|
|
107
|
+
// naming convention in this repo otherwise uses a literal `-v` + digits
|
|
108
|
+
// tail.
|
|
109
|
+
//
|
|
110
|
+
// The NUMERIC suffix on a TICKET-shaped name (`impl-137-2`) looks like the
|
|
111
|
+
// same retry spelling and is deliberately NOT stripped there. The one real
|
|
112
|
+
// instance on disk describes itself as "Implement 137+138+139 set" — a
|
|
113
|
+
// multi-ticket batch that no single `ticket` value represents. Blank is the
|
|
114
|
+
// honest answer; booking it to 137 would join the row to two tickets it did
|
|
115
|
+
// not do.
|
|
116
|
+
//
|
|
117
|
+
// A PR-shaped name's numeric suffix is a DIFFERENT case: `fix-pr-<n>`,
|
|
118
|
+
// `review-pr-<n>`, `finisher-pr-<n>` and `resolve-pr-<n>` carry exactly ONE
|
|
119
|
+
// number, the PR itself, so a second trailing `-\d+` cannot be a second
|
|
120
|
+
// ticket the way `impl-<ticket>-<n>`'s can — there is only ever one PR per
|
|
121
|
+
// such name. A trailing `-\d+` there (`finisher-pr-1440-2`, `fix-pr-1281-2`)
|
|
122
|
+
// is the same re-dispatch retry the letter and `-v<n>` suffixes above
|
|
123
|
+
// already cover, and is now stripped for PR-shaped names only (#1482,
|
|
124
|
+
// measured: 476,202 cache-create tokens across 7 real rows fell through to a
|
|
125
|
+
// blank pr column this way — more than the 37,580 the `-v<n>` fix above
|
|
126
|
+
// addressed). The ticket-shaped `impl-<ticket>-<n>` family above is
|
|
127
|
+
// untouched: its second-ticket ambiguity is real, and a PR-shaped name's is
|
|
128
|
+
// not.
|
|
129
|
+
//
|
|
130
|
+
// `resolve` joins the `-pr-` alternation for #1250: `resolve-pr-<n>` is a
|
|
131
|
+
// controller-dispatched conflict/rebase resolver against an already-open PR
|
|
132
|
+
// (measured session descriptions: "Resolve conflict on PR 1232", "Rebase
|
|
133
|
+
// and resolve conflicts for PR #1310") — the same shape as `fix-pr-<n>`'s
|
|
134
|
+
// applier and `review-pr-<n>`'s reviewer, just not a name run-team's own
|
|
135
|
+
// naming convention fixes, so it stays out of the canonical list in
|
|
136
|
+
// SKILL.md/member-lifecycle.md the same way `finish-<n>`/`finisher-<n>` do.
|
|
137
|
+
// Unrecognised, a `resolve-pr-<n>` member fell through to `{ticket:"",
|
|
138
|
+
// pr:""}`, losing its join key into tier-outcomes.tsv exactly the way an
|
|
139
|
+
// unmatched finisher spelling once did (#1072).
|
|
140
|
+
export function parseMemberName(name) {
|
|
141
|
+
const s = String(name ?? "").trim().replace(/-(?:[a-z]|v\d+)$/, "");
|
|
142
|
+
let m = /^(?:fix|review|finish(?:er)?|resolve)-pr-(\d+)(?:-\d+)?$/.exec(s);
|
|
143
|
+
if (m) return { ticket: "", pr: m[1] };
|
|
144
|
+
m = /^finish(?:er)?-(\d+)$/.exec(s);
|
|
145
|
+
if (m) return { ticket: "", pr: m[1] };
|
|
146
|
+
m = /^impl-(\d+)$/.exec(s);
|
|
147
|
+
if (m) return { ticket: m[1], pr: "" };
|
|
148
|
+
return { ticket: "", pr: "" };
|
|
149
|
+
}
|
|
150
|
+
|
|
151
|
+
// ---------------------------------------------------------------------------
|
|
152
|
+
// transcript reader
|
|
153
|
+
// ---------------------------------------------------------------------------
|
|
154
|
+
|
|
155
|
+
// An omp transcript line carries no `sessionId`/`parentUuid` — session id
|
|
156
|
+
// lives in the DIRECTORY name, never per line; omp's envelope is
|
|
157
|
+
// `{type,id,parentId,timestamp,message}`. That is the structural signature
|
|
158
|
+
// this reader refuses on if it is ever absent — a validation CHECK inside
|
|
159
|
+
// this reader, per #1302's ruling: every root this module is ever handed is
|
|
160
|
+
// an omp `~/.omp/agent/sessions/**` tree, so a line failing this shape
|
|
161
|
+
// signals a corrupted or foreign file, not a harness to dispatch to.
|
|
162
|
+
//
|
|
163
|
+
// Both halves of that signature are checked, not just the blocklist half: a
|
|
164
|
+
// line carrying neither Claude's keys NOR omp's own `type` field (e.g. a
|
|
165
|
+
// foreign/corrupted `{"foo":"bar"}`) used to sail past the blocklist-only
|
|
166
|
+
// check below and fold into a fabricated all-null/zero member record instead
|
|
167
|
+
// of the refusal this comment already promised. `type` is the one envelope
|
|
168
|
+
// field foldOmpTranscript's own loop dispatches every branch on below, so
|
|
169
|
+
// requiring it costs nothing a real omp line does not already carry.
|
|
170
|
+
function assertOmpShaped(d, filePath) {
|
|
171
|
+
const shaped =
|
|
172
|
+
d !== null &&
|
|
173
|
+
typeof d === "object" &&
|
|
174
|
+
typeof d.type === "string" &&
|
|
175
|
+
d.type !== "" &&
|
|
176
|
+
!Object.prototype.hasOwnProperty.call(d, "sessionId") &&
|
|
177
|
+
!Object.prototype.hasOwnProperty.call(d, "parentUuid");
|
|
178
|
+
if (!shaped) {
|
|
179
|
+
throw new Error(`member-record: not an omp-shaped transcript line, refusing to parse it: ${filePath}`);
|
|
180
|
+
}
|
|
181
|
+
}
|
|
182
|
+
|
|
183
|
+
// `openedPrs` (#2209) is every PR number the member's OWN `gh pr create`
|
|
184
|
+
// printed — the source of `pr` for a member whose name carries none, which is
|
|
185
|
+
// every implementer (`impl-<ticket>` names a ticket, never a PR). Chosen over
|
|
186
|
+
// the ledger's `impl-<N>=PR#M` settle token because that token is written by a
|
|
187
|
+
// live run only: a regeneration over past sessions has the transcript and
|
|
188
|
+
// nothing else, and this module stays pure over transcripts. Measured
|
|
189
|
+
// 2026-09-29 across every real `~/.omp/agent/sessions/**/*.jsonl`: 443
|
|
190
|
+
// transcripts printed a created PR from a `bash` toolCall whose result held
|
|
191
|
+
// gh's stdout — the URL on a line of its own — then omp's "Wall time"
|
|
192
|
+
// trailer. A create is also sometimes run via `eval` instead of `bash` —
|
|
193
|
+
// measured on impl-1113 (a Python subprocess, `gh pr create` printing PR
|
|
194
|
+
// #1535) — and stays blank there too: `eval` is not read, because a cell
|
|
195
|
+
// that only READS transcripts (or `gh pr view`s) prints the very same URLs.
|
|
196
|
+
//
|
|
197
|
+
// Three filters keep a URL that is NOT a PR this member opened out:
|
|
198
|
+
// - the CALL must invoke `gh pr create` — first in the command, after a
|
|
199
|
+
// shell separator (`;`, `&`, `|`, `(`, newline) or an `if`/`then`/`do`/`!`
|
|
200
|
+
// keyword, optionally through `rtk`/`env`/`command`/`timeout <n>`, a
|
|
201
|
+
// `NAME=value` prefix, or a path-qualified `.../gh` — a `gh pr view`/
|
|
202
|
+
// `gh pr list` result prints bare PR URLs too, and `gh pr create` quoted
|
|
203
|
+
// as text is no invocation (measured false-miss shapes: #2213 review);
|
|
204
|
+
// - the URL must be a LINE of its own, the shape gh prints, never a URL
|
|
205
|
+
// inside prose;
|
|
206
|
+
// - a URL right after gh's "... already exists:" refusal is skipped: that PR
|
|
207
|
+
// was opened by someone else for the same branch.
|
|
208
|
+
// `isError` is deliberately NOT a filter: a create chained before a failing
|
|
209
|
+
// command (`gh pr create ... && gh pr edit --add-label x`) exits non-zero
|
|
210
|
+
// having still opened the PR it printed.
|
|
211
|
+
//
|
|
212
|
+
// This is a text heuristic, not a shell parser, so it can also count a URL
|
|
213
|
+
// that is NOT the member's own opened PR: it does not know quoting, so a
|
|
214
|
+
// separator character reused inside a quoted string right before the
|
|
215
|
+
// literal text `gh pr create` (e.g. `echo "a; gh pr create"`) reads as an
|
|
216
|
+
// invocation too; a `--dry-run` create still counts a URL it merely
|
|
217
|
+
// previews; a `gh pr create` mentioned earlier in the same multi-line
|
|
218
|
+
// command (a heredoc body, a failed retry) can pair with an unrelated
|
|
219
|
+
// later URL in the result; and a `create || gh pr view` fallback's own
|
|
220
|
+
// `view` copy of an "already exists:" URL is not caught by the one-line-back
|
|
221
|
+
// skip above. Measured 0 real occurrences of any of these across the whole
|
|
222
|
+
// local `~/.omp/agent/sessions/**/*.jsonl` corpus and every filled `pr` in
|
|
223
|
+
// the committed TSV (#2213 review); accepted as known gaps rather than
|
|
224
|
+
// hand-rolling shell-quote/heredoc awareness into a single regex.
|
|
225
|
+
const GH_PR_CREATE_RE =
|
|
226
|
+
/(?:^|[;&|(\n]|\b(?:if|then|do|!)\s)\s*(?:(?:env|command|rtk|timeout\s+\S+)\s+|[A-Za-z_]\w*=\S*\s+)*(?:\S*\/)?gh\s+pr\s+create\b/;
|
|
227
|
+
function createdPrNumbers(text) {
|
|
228
|
+
const out = [];
|
|
229
|
+
const lines = text.split("\n");
|
|
230
|
+
for (let i = 0; i < lines.length; i++) {
|
|
231
|
+
const m = /^https?:\/\/[^/\s]+\/[^/\s]+\/[^/\s]+\/pull\/(\d+)\s*$/.exec(lines[i]);
|
|
232
|
+
if (m && !(i > 0 && /already exists:\s*$/.test(lines[i - 1]))) out.push(m[1]);
|
|
233
|
+
}
|
|
234
|
+
return out;
|
|
235
|
+
}
|
|
236
|
+
|
|
237
|
+
// Folds one omp subagent transcript into per-turn totals. One
|
|
238
|
+
// `message.usage` per assistant turn already (no cross-line fold-back
|
|
239
|
+
// needed), a real per-turn dollar cost at `usage.cost.total`, and the
|
|
240
|
+
// declared thinking level on its own `thinking_level_change` event rather
|
|
241
|
+
// than repeated per line. Measured 2026-09-08/09 against real
|
|
242
|
+
// `~/.omp/agent/sessions/**/*.jsonl` files:
|
|
243
|
+
// {"type":"session",...,"cwd":"/Users/chris/dev/fleet-plugin"}
|
|
244
|
+
// {"type":"thinking_level_change",...,"thinkingLevel":"high","configured":null}
|
|
245
|
+
// {"type":"message","message":{"role":"assistant","model":"claude-sonnet-5",
|
|
246
|
+
// "usage":{"input":2,"output":201,"cacheRead":0,"cacheWrite":31319,
|
|
247
|
+
// "totalTokens":31522,"cost":{"input":4e-6,"output":0.00201,
|
|
248
|
+
// "cacheRead":0,"cacheWrite":0.0783,"total":0.0803}}}}
|
|
249
|
+
//
|
|
250
|
+
// `thinking` reads the harness-WRITTEN level, never the frontmatter — that
|
|
251
|
+
// is what makes #1298's declared-vs-resolved comparison possible (#1302's
|
|
252
|
+
// ruling). It stays `null` here (ompMemberRecord turns that into the
|
|
253
|
+
// record's `-`) when no `thinking_level_change` line exists, rather than
|
|
254
|
+
// guessing the default: a member that is not a fleet definition genuinely
|
|
255
|
+
// has no recorded level, and the hole must stay visible.
|
|
256
|
+
//
|
|
257
|
+
// `session_init.task` is the closest thing to a dispatch sidecar this
|
|
258
|
+
// module has — there is no separate description file at all.
|
|
259
|
+
// ompMemberRecord below uses it, together with the transcript's own nesting
|
|
260
|
+
// depth AND the AgentId itself, as REAL classifyRole() signals. A
|
|
261
|
+
// canonically-stemmed AgentId (`impl-<n>`, `fix-pr-<n>`, `finisher-<n>`,
|
|
262
|
+
// `review-pr-<n>`, `merge-bot-<n>`) IS matched against classifyRole, as
|
|
263
|
+
// `memberName`, per #1506; only a non-canonical AgentId — a generated word
|
|
264
|
+
// pair that names nothing — is still never matched.
|
|
265
|
+
//
|
|
266
|
+
// `resolvedModelIdentity` (#1345) is `session_init`'s OWN field, written at
|
|
267
|
+
// DISPATCH — before the member's first assistant turn exists, which is what
|
|
268
|
+
// makes it different from `model` above: a member still working folds to
|
|
269
|
+
// `model: null` (no assistant turn yet) but already carries
|
|
270
|
+
// `resolvedModelIdentity` (measured `anthropic/claude-opus-5` (61),
|
|
271
|
+
// `anthropic/claude-sonnet-5` (103), `anthropic/claude-haiku-4-5` (68)
|
|
272
|
+
// across real `~/.omp/agent/sessions/**` — always provider-prefixed, never a
|
|
273
|
+
// bare alias). `model` itself is left untouched by this addition: board.mjs
|
|
274
|
+
// and member-outcomes.mjs read `model` for cost/spend attribution, where the
|
|
275
|
+
// per-turn value (which can in principle change mid-run) is the fact they
|
|
276
|
+
// want, not the dispatch-time identity.
|
|
277
|
+
//
|
|
278
|
+
// `agent` (#1066) is `session_init`'s dispatch-time record of WHICH AGENT
|
|
279
|
+
// DEFINITION this member is. Measured 2026-09-12 across real
|
|
280
|
+
// `~/.omp/agent/sessions/**`: present on every one of the 1,191 transcripts
|
|
281
|
+
// carrying a `session_init` line (`fleet-implementer` 51,
|
|
282
|
+
// `fleet-implementer-alt` 17, the default `task` 220, plus the review
|
|
283
|
+
// fan-out's own definitions), absent only where the line itself is.
|
|
284
|
+
//
|
|
285
|
+
// `entries` (#1717) is the per-agent tool stream compute-spend.mjs's
|
|
286
|
+
// attributeTools reads: `assistant`/`result` entries. Measured 2026-09-26
|
|
287
|
+
// across 5,085 real `~/.omp/agent/sessions/**/*.jsonl` files, the call is a
|
|
288
|
+
// block on the assistant message and the result a line of its own:
|
|
289
|
+
// {"type":"message","message":{"role":"assistant","stopReason":"toolUse",
|
|
290
|
+
// "content":[{"type":"toolCall","id":"toolu_01JV…","name":"bash",
|
|
291
|
+
// "arguments":{…},"intent":"…"}],"usage":{…}}}
|
|
292
|
+
// {"type":"message","message":{"role":"toolResult","toolCallId":"toolu_01JV…",
|
|
293
|
+
// "toolName":"bash","content":[{"type":"text","text":"…"}],"details":{…},
|
|
294
|
+
// "isError":false,"timestamp":1790415990525}}
|
|
295
|
+
// All 176,056 `toolCall` blocks carried `id` and `name`; the ~29-call gap
|
|
296
|
+
// against the 176,027 results below is orphan calls whose result never
|
|
297
|
+
// landed (aborted before the tool replied) — attributeTools still counts
|
|
298
|
+
// them under their own name via `calls++`, only `resultChars`/`cacheWrite`
|
|
299
|
+
// stay at 0. Each of the 176,027 results sat on its own line, so parallel
|
|
300
|
+
// calls arrive as consecutive `toolResult` lines, which attributeTools
|
|
301
|
+
// accumulates into one batch (run lengths 2 through 14 matched the
|
|
302
|
+
// calls-per-turn counts to within one).
|
|
303
|
+
//
|
|
304
|
+
// A result's `chars` sums its text blocks' lengths (`content` was an array
|
|
305
|
+
// every time: 179,380 text blocks, 14 image blocks — an unmeasured
|
|
306
|
+
// non-array `content` folds to 0 blocks and `chars: 0`). A non-text block
|
|
307
|
+
// (an image) counts at its JSON length. `prunedAt` (194) holds a
|
|
308
|
+
// placeholder such as "[Uneventful result elided]" instead of the output,
|
|
309
|
+
// so its `chars` is the placeholder's length.
|
|
310
|
+
//
|
|
311
|
+
// The result line names its tool (`toolName`), but the stream stays
|
|
312
|
+
// id-only, so a result whose id no `toolCall` block carries books as
|
|
313
|
+
// `unknown`. Measured, that is a turn aborted mid-stream (2 results): the
|
|
314
|
+
// tool's "not executed" result is written first, then the assistant message
|
|
315
|
+
// persists with empty content, so the call never lands. That aborted
|
|
316
|
+
// message is still an `assistant` entry, as is every line `turns` counts.
|
|
317
|
+
// 142 aborted or errored turns followed a result batch. When one never
|
|
318
|
+
// reached the API its usage is all zero, and attributeTools books the batch
|
|
319
|
+
// before it at 0.
|
|
320
|
+
//
|
|
321
|
+
// `openedPrs` (#2209) folds the `bash` calls that invoke `gh pr create`
|
|
322
|
+
// against their own results, via createdPrNumbers above — distinct PR
|
|
323
|
+
// numbers in first-seen order.
|
|
324
|
+
export function foldOmpTranscript(jsonlText, filePath) {
|
|
325
|
+
let model = null, thinking = null, task = null, resolvedModelIdentity = null, agent = null;
|
|
326
|
+
let firstTs = null, lastTs = null;
|
|
327
|
+
let input = 0, cacheWrite = 0, cacheRead = 0, output = 0, cost = 0, turns = 0;
|
|
328
|
+
let sawCost = false;
|
|
329
|
+
let malformedNonLastLines = 0;
|
|
330
|
+
const entries = [];
|
|
331
|
+
const createCallIds = new Set();
|
|
332
|
+
const openedPrs = new Set();
|
|
333
|
+
const lines = String(jsonlText ?? "").split("\n");
|
|
334
|
+
for (let i = 0; i < lines.length; i++) {
|
|
335
|
+
const raw = lines[i];
|
|
336
|
+
if (!raw.trim()) continue;
|
|
337
|
+
let d;
|
|
338
|
+
// A torn tail (transcript read mid-write) is dropped, and costs at most
|
|
339
|
+
// that one turn either way. A MIDDLE line failing the same parse is not
|
|
340
|
+
// a live write in progress — it is lost data, and (#1717) it can desync
|
|
341
|
+
// a toolCall from its toolResult, not just a turn's totals, so it is
|
|
342
|
+
// counted, never for the last line.
|
|
343
|
+
try { d = JSON.parse(raw); }
|
|
344
|
+
catch { if (i !== lines.length - 1) malformedNonLastLines++; continue; }
|
|
345
|
+
assertOmpShaped(d, filePath);
|
|
346
|
+
if (typeof d.timestamp === "string") { firstTs ??= d.timestamp; lastTs = d.timestamp; }
|
|
347
|
+
if (d.type === "thinking_level_change" && typeof d.thinkingLevel === "string") thinking = d.thinkingLevel;
|
|
348
|
+
if (d.type === "session_init") {
|
|
349
|
+
if (typeof d.task === "string") task = d.task;
|
|
350
|
+
if (typeof d.resolvedModelIdentity === "string") resolvedModelIdentity = d.resolvedModelIdentity;
|
|
351
|
+
if (typeof d.agent === "string") agent = d.agent;
|
|
352
|
+
}
|
|
353
|
+
const m = d.message;
|
|
354
|
+
if (d.type === "message" && m?.role === "assistant") {
|
|
355
|
+
const blocks = Array.isArray(m.content) ? m.content : [];
|
|
356
|
+
for (const c of blocks) {
|
|
357
|
+
if (c?.type === "toolCall" && c.name === "bash" && typeof c.arguments?.command === "string"
|
|
358
|
+
&& GH_PR_CREATE_RE.test(c.arguments.command)) createCallIds.add(c.id);
|
|
359
|
+
}
|
|
360
|
+
if (m.usage) {
|
|
361
|
+
const u = m.usage;
|
|
362
|
+
if (typeof m.model === "string" && m.model) model = m.model;
|
|
363
|
+
const cw = Number(u.cacheWrite ?? 0);
|
|
364
|
+
input += Number(u.input ?? 0);
|
|
365
|
+
cacheWrite += cw;
|
|
366
|
+
cacheRead += Number(u.cacheRead ?? 0);
|
|
367
|
+
output += Number(u.output ?? 0);
|
|
368
|
+
if (u.cost && typeof u.cost.total === "number") { cost += u.cost.total; sawCost = true; }
|
|
369
|
+
turns++;
|
|
370
|
+
entries.push({
|
|
371
|
+
kind: "assistant", cacheWrite: cw,
|
|
372
|
+
tools: blocks.filter((c) => c?.type === "toolCall").map((c) => ({ id: c.id, name: c.name })),
|
|
373
|
+
});
|
|
374
|
+
}
|
|
375
|
+
} else if (d.type === "message" && m?.role === "toolResult") {
|
|
376
|
+
const blocks = Array.isArray(m.content) ? m.content : [];
|
|
377
|
+
const chars = blocks.reduce(
|
|
378
|
+
(n, b) => n + (b?.type === "text" && typeof b.text === "string" ? b.text.length : JSON.stringify(b).length), 0);
|
|
379
|
+
entries.push({ kind: "result", results: [{ id: m.toolCallId, chars }] });
|
|
380
|
+
if (createCallIds.has(m.toolCallId)) {
|
|
381
|
+
const text = blocks.map((b) => (b?.type === "text" && typeof b.text === "string" ? b.text : "")).join("\n");
|
|
382
|
+
for (const n of createdPrNumbers(text)) openedPrs.add(n);
|
|
383
|
+
}
|
|
384
|
+
}
|
|
385
|
+
}
|
|
386
|
+
const span = firstTs && lastTs ? (Date.parse(lastTs) - Date.parse(firstTs)) / 1000 : 0;
|
|
387
|
+
return {
|
|
388
|
+
model, thinking, task, resolvedModelIdentity, agent,
|
|
389
|
+
input, cacheWrite, cacheRead, output,
|
|
390
|
+
cost: sawCost ? cost : null,
|
|
391
|
+
turns,
|
|
392
|
+
wallS: Number.isFinite(span) ? Math.round(span) : 0,
|
|
393
|
+
entries,
|
|
394
|
+
malformedNonLastLines,
|
|
395
|
+
openedPrs: [...openedPrs],
|
|
396
|
+
};
|
|
397
|
+
}
|
|
398
|
+
|
|
399
|
+
// One member record from one omp transcript's fold. `member` is the AgentId
|
|
400
|
+
// (the filename stem, e.g. `InstallVerifySearch`) — there is no separate
|
|
401
|
+
// display name, so `ticket`/`pr` extraction runs against it directly — except
|
|
402
|
+
// a `pr` the name cannot carry, read off the fold's `openedPrs` (#2209).
|
|
403
|
+
//
|
|
404
|
+
// `role` is NEVER guessed off the bare AgentId ALONE — a generated CamelCase
|
|
405
|
+
// word pair names nothing classifyRole can read. FOUR real signals exist:
|
|
406
|
+
// `session_init.agent` (the agent DEFINITION the dispatch named),
|
|
407
|
+
// `session_init.task` (the dispatch prompt, when present), `spawnDepth` (the
|
|
408
|
+
// transcript's own nesting depth, supplied by readOmpSession from the walk —
|
|
409
|
+
// a fact about where the file lives, not a guess about what it is), and —
|
|
410
|
+
// per #1506's ruling below — the AgentId itself, handed to classifyRole as
|
|
411
|
+
// `memberName`. Depth matters on its own: classifyRole checks depth BEFORE
|
|
412
|
+
// any text match, specifically so a nested member whose task happens to
|
|
413
|
+
// read like a reviewer's ("Review PR 1353 correctness") still books as the
|
|
414
|
+
// fan-out specialist it structurally is, not a reviewer. None present
|
|
415
|
+
// yields `"-"` — the same visible-hole spelling as `thinking`, never a
|
|
416
|
+
// default like "other", which only makes sense where a real dispatch record
|
|
417
|
+
// backs it.
|
|
418
|
+
//
|
|
419
|
+
// `agent` reaches classifyRole as its `agentDefinition` (#1486, #1505). It
|
|
420
|
+
// is the same value the `subagent_type` column below already records, and
|
|
421
|
+
// withholding it here made classifyRole's FIRST branch — the one whose
|
|
422
|
+
// comment says memory-system work must "never land in review spend" —
|
|
423
|
+
// structurally unreachable, leaving every row's role decided by
|
|
424
|
+
// dispatch-prompt prose alone. Measured 2026-09-16 before the fix: 50
|
|
425
|
+
// omp/memory-proxy rows, not one of them `role=memory`, and a single
|
|
426
|
+
// definition (`fleet-review-verifier`) split across four buckets on nothing
|
|
427
|
+
// but how each prompt happened to read.
|
|
428
|
+
//
|
|
429
|
+
// #1506's gap: `memberName` used to be left unset here on the theory that
|
|
430
|
+
// the AgentId genuinely is not a name — a generated CamelCase word pair
|
|
431
|
+
// (`InstallVerifySearch`) names nothing the classifier can read. True for
|
|
432
|
+
// the ordinary case, but false for members dispatched under run-team's own
|
|
433
|
+
// naming convention: `impl-<n>`, `fix-pr-<n>`, `finisher-<n>`, `finish-<n>`,
|
|
434
|
+
// `review-pr-<n>` and `merge-bot-<n>` ARE the dispatch name — exactly why
|
|
435
|
+
// `parseMemberName` runs against this same stem below. Those members are
|
|
436
|
+
// dispatched under the generic default `task` definition (`folded.agent` is
|
|
437
|
+
// `undefined`), so the definition-based fix above cannot reach them, and
|
|
438
|
+
// they fall through to prose classification of `folded.task` alone.
|
|
439
|
+
// Measured against docs/metrics/member-outcomes.tsv (#1506): 478 omp rows
|
|
440
|
+
// carry such an AgentId, 52 of them booked `other` for want of this signal —
|
|
441
|
+
// 31 `fix-pr-*`, 9 `impl-*` and 12 `merge-bot-*` members whose dispatch
|
|
442
|
+
// prompt never happens to name the role.
|
|
443
|
+
//
|
|
444
|
+
// RULED: pass the AgentId unconditionally — NOT gated on looking
|
|
445
|
+
// name-shaped first. Every name-driven branch in classifyRole is
|
|
446
|
+
// hyphen-anchored (`^impl-`, `review-pr-`, `fix-pr-`, `^finish-`) or a
|
|
447
|
+
// multi-word phrase ("implement ticket", "review pr"), so a stray generated
|
|
448
|
+
// word pair cannot coincidentally satisfy one; the one bare-word pattern,
|
|
449
|
+
// `finisher`, carries that risk today and has not fired on the 4,574
|
|
450
|
+
// sidecars measured for #1505. `OMP_CANONICAL_STEM_RE` below exists only to
|
|
451
|
+
// widen `hasRoleSignal` itself: a canonically-named member whose transcript
|
|
452
|
+
// predates #1343 (no `session_init` line at all, so neither `task` nor
|
|
453
|
+
// `agent` exist) still holds a readable identity and must not fall back to
|
|
454
|
+
// the "-" hole. Measured 2026-09-16: zero such rows on disk today, but the
|
|
455
|
+
// gate exists so the design does not assume that stays true forever. Built
|
|
456
|
+
// from `CANONICAL_MEMBER_NAME_PREFIXES` (compute-spend.mjs) rather than its
|
|
457
|
+
// own copy of the prefix list, so the two cannot drift apart.
|
|
458
|
+
//
|
|
459
|
+
// REJECTED: dispatching these roles under real agent definitions instead
|
|
460
|
+
// (the issue's other named approach) — a controller/dispatch-convention
|
|
461
|
+
// change, out of scope for a classifier fix, and unlike this one it cannot
|
|
462
|
+
// repair the 478 historical rows already on disk.
|
|
463
|
+
const OMP_CANONICAL_STEM_RE = new RegExp(`^(?:${CANONICAL_MEMBER_NAME_PREFIXES})-`);
|
|
464
|
+
export function ompMemberRecord(folded, agentStem, spawnDepth = 0) {
|
|
465
|
+
if (!folded.model) return null; // no assistant turn — not a real member transcript
|
|
466
|
+
const member = agentStem;
|
|
467
|
+
const { ticket, pr: namedPr } = parseMemberName(member);
|
|
468
|
+
// A name-carried PR (`fix-pr-<n>`, `finisher-<n>`, ...) is the member's unit
|
|
469
|
+
// of work by construction and wins. Otherwise (#2209) the one PR its own
|
|
470
|
+
// `gh pr create` printed; none, or more than one (measured: impl-1578 opened
|
|
471
|
+
// its real PR plus a throwaway probe PR), stays blank — never a guess.
|
|
472
|
+
// `folded.openedPrs` is `foldOmpTranscript`'s own unconditional return
|
|
473
|
+
// shape (always an array, never absent), so no `?.` is needed here.
|
|
474
|
+
const pr = namedPr || (folded.openedPrs.length === 1 ? folded.openedPrs[0] : "");
|
|
475
|
+
const hasRoleSignal = spawnDepth >= 1 || typeof folded.task === "string" || typeof folded.agent === "string"
|
|
476
|
+
|| OMP_CANONICAL_STEM_RE.test(member);
|
|
477
|
+
const role = hasRoleSignal
|
|
478
|
+
? classifyRole({ agentDefinition: folded.agent, memberName: member, description: folded.task, spawnDepth })
|
|
479
|
+
: "-";
|
|
480
|
+
return {
|
|
481
|
+
harness: "omp",
|
|
482
|
+
role,
|
|
483
|
+
member,
|
|
484
|
+
model: folded.model,
|
|
485
|
+
// Additive only (#1345) — `model` above stays the per-turn value
|
|
486
|
+
// board.mjs/member-outcomes.mjs already key cost/spend attribution on;
|
|
487
|
+
// this is the dispatch-time identity `session_init` wrote before any
|
|
488
|
+
// turn existed, `null` when the transcript predates #1343 or carries no
|
|
489
|
+
// `session_init` line at all (never guessed).
|
|
490
|
+
resolvedModelIdentity: folded.resolvedModelIdentity ?? null,
|
|
491
|
+
thinking: folded.thinking ?? "-",
|
|
492
|
+
// The dispatch record's own agent definition, `""` when the transcript
|
|
493
|
+
// carries no `session_init` line to read one from.
|
|
494
|
+
subagent_type: folded.agent ?? "",
|
|
495
|
+
tokens_in: folded.input, tokens_cache_create: folded.cacheWrite,
|
|
496
|
+
tokens_cache_read: folded.cacheRead, tokens_out: folded.output,
|
|
497
|
+
cost: folded.cost,
|
|
498
|
+
wall_s: folded.wallS, turns: folded.turns,
|
|
499
|
+
ticket, pr,
|
|
500
|
+
};
|
|
501
|
+
}
|
|
502
|
+
|
|
503
|
+
// ompMemberRecord straight from the transcript text. board.mjs's live spend
|
|
504
|
+
// panel calls the two halves itself instead, because it needs the fold's tool
|
|
505
|
+
// stream (#1717) beside the record, and folding the file twice for it would
|
|
506
|
+
// parse every transcript twice on every tick.
|
|
507
|
+
export function readOmpMember(jsonlText, filePath, agentStem, spawnDepth = 0) {
|
|
508
|
+
return ompMemberRecord(foldOmpTranscript(jsonlText, filePath), agentStem, spawnDepth);
|
|
509
|
+
}
|
|
510
|
+
|
|
511
|
+
// One omp session directory's member transcripts, as the walk both readers of
|
|
512
|
+
// that directory need it: readOmpSession below, and board.mjs's live spend
|
|
513
|
+
// panel (#1716), which folds each file itself so it can keep its own
|
|
514
|
+
// per-transcript skip tally. RECURSIVE: a member can itself dispatch further
|
|
515
|
+
// members (measured on disk — a research session's `Facts1303/` held seven
|
|
516
|
+
// more `.jsonl` files one level down), and `agent` is the path-relative stem
|
|
517
|
+
// so those nest instead of colliding. `spawnDepth` is read straight off
|
|
518
|
+
// that path — one `/` per nesting level — and handed to ompMemberRecord as a
|
|
519
|
+
// real fact about the walk, not a guess about the member.
|
|
520
|
+
//
|
|
521
|
+
// Throws when the directory itself cannot be listed; the caller decides what
|
|
522
|
+
// that means.
|
|
523
|
+
export function ompSessionTranscripts(sessionDir) {
|
|
524
|
+
return readdirSync(sessionDir, { recursive: true })
|
|
525
|
+
.filter((f) => f.endsWith(".jsonl"))
|
|
526
|
+
.map((f) => ({
|
|
527
|
+
file: join(sessionDir, f),
|
|
528
|
+
agent: f.replace(/\.jsonl$/, ""),
|
|
529
|
+
spawnDepth: (f.match(/[\\/]/g) ?? []).length,
|
|
530
|
+
}));
|
|
531
|
+
}
|
|
532
|
+
|
|
533
|
+
// Walks one omp session directory into records, via the walk above.
|
|
534
|
+
//
|
|
535
|
+
// Deliberately NOT wrapped in a blanket try/catch around readOmpMember: the
|
|
536
|
+
// wrong-shape refusal (assertOmpShaped, inside foldOmpTranscript) must
|
|
537
|
+
// propagate all the way out of readMembers, uncaught, per #1342's acceptance
|
|
538
|
+
// criterion. Only the directory listing and the raw file read are given the
|
|
539
|
+
// ordinary per-member tolerance.
|
|
540
|
+
export function readOmpSession(sessionDir) {
|
|
541
|
+
let transcripts;
|
|
542
|
+
try { transcripts = ompSessionTranscripts(sessionDir); }
|
|
543
|
+
catch { return []; }
|
|
544
|
+
const session = basename(sessionDir);
|
|
545
|
+
const rows = [];
|
|
546
|
+
for (const { file, agent, spawnDepth } of transcripts) {
|
|
547
|
+
let text;
|
|
548
|
+
try { text = readFileSync(file, "utf8"); }
|
|
549
|
+
catch { continue; }
|
|
550
|
+
const rec = readOmpMember(text, file, agent, spawnDepth); // may throw — see comment above
|
|
551
|
+
if (!rec) continue;
|
|
552
|
+
rows.push({ ...rec, session, agent });
|
|
553
|
+
}
|
|
554
|
+
return rows;
|
|
555
|
+
}
|
|
556
|
+
|
|
557
|
+
// An omp session directory is identified STRICTLY by name — `<ISO>_<uuid>`,
|
|
558
|
+
// e.g. `2026-09-08T13-13-27-300Z_01a08126-…` — never by holding `.jsonl`
|
|
559
|
+
// files directly. That second test used to be a shortcut, and it fired one
|
|
560
|
+
// level too high on the real tree: `~/.omp/agent/sessions/-dev-fleet-plugin/`
|
|
561
|
+
// holds the project's MAIN-session transcripts as plain files SIBLING to the
|
|
562
|
+
// per-session directories, so `findOmpSessionDirs` recursing from the
|
|
563
|
+
// encoded-cwd dir hit the `.jsonl` test there first, returned the whole
|
|
564
|
+
// project as "one session", stamped every row `session=-dev-fleet-plugin`
|
|
565
|
+
// instead of the `<ISO>_<uuid>` name #1302 rules the row key on, and booked
|
|
566
|
+
// the controller's own top-level transcripts as members. Name-only matching
|
|
567
|
+
// costs nothing a real fixture needs: every fixture in this repo already
|
|
568
|
+
// names its session dir in the real shape.
|
|
569
|
+
const OMP_SESSION_DIR_RE = /^\d{4}-\d{2}-\d{2}T[\d-]+Z_[0-9a-f-]+$/i;
|
|
570
|
+
|
|
571
|
+
// Exposed for callers that already hold one EXPLICIT directory rather than a
|
|
572
|
+
// tree to search — member-outcomes.mjs's CLI, so the pattern is defined once.
|
|
573
|
+
export function isOmpSessionDirName(name) {
|
|
574
|
+
return OMP_SESSION_DIR_RE.test(name);
|
|
575
|
+
}
|
|
576
|
+
|
|
577
|
+
function findOmpSessionDirs(root) {
|
|
578
|
+
if (OMP_SESSION_DIR_RE.test(basename(root))) return [root];
|
|
579
|
+
let ents;
|
|
580
|
+
try { ents = readdirSync(root, { withFileTypes: true }); }
|
|
581
|
+
catch { return []; }
|
|
582
|
+
const found = [];
|
|
583
|
+
for (const e of ents) if (e.isDirectory()) found.push(...findOmpSessionDirs(join(root, e.name)));
|
|
584
|
+
return found;
|
|
585
|
+
}
|
|
586
|
+
|
|
587
|
+
// ---------------------------------------------------------------------------
|
|
588
|
+
// readMembers — the one entry point
|
|
589
|
+
// ---------------------------------------------------------------------------
|
|
590
|
+
|
|
591
|
+
// The adapter's public entry point. A root that resolves to no session
|
|
592
|
+
// directory anywhere is refused rather than silently skipped: silently
|
|
593
|
+
// returning no rows for a typo'd path is the blackout #1302's ruling exists
|
|
594
|
+
// to prevent.
|
|
595
|
+
export function readMembers(roots) {
|
|
596
|
+
const rows = [];
|
|
597
|
+
for (const root of [].concat(roots ?? [])) {
|
|
598
|
+
const dirs = findOmpSessionDirs(root);
|
|
599
|
+
if (dirs.length === 0) {
|
|
600
|
+
throw new Error(`member-record: no omp session directory found anywhere under this root, refusing to silently contribute nothing: ${root}`);
|
|
601
|
+
}
|
|
602
|
+
for (const dir of dirs) rows.push(...readOmpSession(dir));
|
|
603
|
+
}
|
|
604
|
+
return rows;
|
|
605
|
+
}
|