[PIsberg]

Stacked on #1225 and #1226: review the last commit (d65a6d8c0). The first two commits are theirs and drop out of this diff as they merge. All three change docs/CHANGELOG.md, which is why they are chained.

Partly addresses #1172. What is still not followed stays there (listed below).

Why

A lambda reached the common pool's workers only while every stage between it and the terminal operation was woven. sorted, distinct, limit, skip, parallel(), sequential() and the as*Stream conversions were not. So list.stream().map(...).parallel().sorted().toList(), a common shape, ran its map lambda unattributed, and a race in it was seen only on the caller's own elements. The lambdas of flatMap, takeWhile, dropWhile and the primitive-to-primitive maps (IntStream.mapToLong and the rest) ran unattributed too.

What changed

• Stages that run no lambda carry the slots behind them to the stream they return: sorted(), distinct, limit, skip, parallel(), sequential() on all four stream kinds, a Stream's unordered(), and asLongStream/asDoubleStream. A Stream's parallel, sequential and unordered are declared by BaseStream, so their call sites expect a BaseStream back, and those hooks return exactly that.
• The remaining lambda stages wait in a slot like map's: flatMap, takeWhile and dropWhile on all four kinds, and the six primitive-to-primitive mapTo*.
• A sorted(Comparator) comparator runs as a combine: a parallel sort compares elements other subtasks produced, after a join the agent does not see, so it is ordered after the elements that built them, like a reduce operator.
• 47 additive hooks, @since 1.14.2, and 47 weaver rows.

How it was verified

• Tests first: four new cases in AgentParallelStreamHooksTest failed to compile on the missing hooks. All 23 cases in the class pass now.
• Mutations, each restored afterwards. Dropping the sorted() carry turns aLambdaStaysLentBehindTheCarryingStages red. Running the comparator as a plain element instead of a combine turns aSortedComparatorIsOrderedAfterTheElementsItCompares red: a correct parallel sort was then reported as a race.
• Corpus, through the real agent: idiom_parallelMapThenSorted_plainCounter (MUST_FIRE) and its locked twin (MUST_STAY_SILENT). The idiom lane passed 88/88 and every other lane is green.
• Library suite 3967 and agent suite 118 green. The static gate chain is green. The first library run failed one case, JavadocDescribesRatherThanRestatesTest, on a new @param comparator the comparator; I reworded it and that class and the stream class pass. The agent suite was skipped in that first run because the reactor stopped, so I ran it on its own afterwards (118 green).

Docs

The docs/CHANGELOG.md #1172 entry, docs/agent/limitations.md and docs/analysis/corpus-eval-future-improvements.md now list what is followed and what is not. No VibeTags guardrail was added: the hooks class's contract-frozen rule covers the new hooks, and no external contract changed.

Checklist

• A behaviour change ships with a test that fails without the fix
• A new or changed detector is tested against the buggy code and its correctly synchronized twin (no detector changed; the corpus pair covers both directions of the feed)
• Docs the change makes wrong are updated in this same change
• Nothing was hand-edited between VIBETAGS-START / VIBETAGS-END markers
• Public API changes are intentional and carry the right version bump: 47 additive hooks, @since 1.14.2
• Commits an agent authored carry the Co-Authored-By trailer
• A new dependency was proposed with a reason (none added)
• CI is green, or the failure is explained below

Still not followed (tracked in #1172)

A mapMulti lambda, a stage behind a primitive stream's unordered() or onClose, a pipeline ended by iterator(), and a Collector's own functions.

🤖 Generated with Claude Code