fix(skill): tighten the stream workflow for the current Tessl reviewer - #94

Summary

• Problem: the current Tessl reviewer scores the released skills/java-streams/SKILL.md at 91 (content 79), so the Tessl skill review check is red on every pull request (fix(skill): released SKILL.md scores 91% under the current Tessl reviewer #88).
• Why it matters: the workflow gate is 100, and the skill review is part of what the registry shows.
• What changed: the opening paragraph is four lines instead of six; the find-first rule lives in the decision table plus a checkpoint after API selection; step 2 and step 4 each carry a one-line snippet; a compact toMap example with merge, null filtering, and an identity mapper joins step 4; the review-output bullets are one imperative sentence. Every rule keeps its meaning.
• What did not change: the references and the evals.
• October follow-up (commit a81acce): the reviewer changed again and scored the first rewrite 94. The second commit states the named-helper rule once (step 5), shows Gatherers.mapConcurrent as a complete pipeline, wraps the toMap merge example in the SeatIndex class it references, drops the OrderChecks snippet that repeated the anyMatch advice, points to the parallel-safety conditions once (step 7), and scopes the generic trigger words in the description to stream stages. No rule was added or removed.

Quality review on this branch: 100 (01a10390-fe22-72b8-bfe2-26ecb066c104), after 94, 97, 97, and 98 on the October drafts and one N/A when a colon in the description broke the YAML front matter.

Change Type

• Skill behavior
• Evals or scoring
• Documentation
• CI, release, or dependency automation
• Repository metadata or contribution process
• Other maintenance

Linked Issue

• Fixes fix(skill): released SKILL.md scores 91% under the current Tessl reviewer #88

User-Visible Behavior

Same guidance, shorter body, one more copy-paste example.

Bug Fix Details

• Root cause: Tessl replaced the single-pass skill review with a full-bundle agent review that weighs conciseness and actionability differently.
• Test, eval, or guardrail added: none; the existing suites are the guardrail and must rerun (below).
• If no test or eval was added, why not: the change is wording and examples.

Ownership Boundary

Unchanged. The example uses Function.identity() inside a collector, which the ownership page lists as stream-owned collector semantics.

Hosted Evidence

Lift-sensitive: this changes the measured skill bundle. Revert strategy: revert the squash commit. Every run below used the runtime text of commit a81acce and the Tessl default solver, which changed from deepseek-v4-flash to deepseek-v4.1-flash before this window.

Every scenario reached 100 with context. The 2.22x bar from the handoff cannot be met: all four main baselines also scored 100 under the new solver, so the main suite shows 1.0x. That comes from the solver, not the skill text, and it needs a maintainer decision (new main scenarios, or accept that the registry score drops) before release. Results are also recorded in the NUMBERING.md notes. The per-scenario criteria-meta.json sidecars exist only on main, so they get updated after this branch is rebased.

Before merge: the ruleset requires an up-to-date branch, and this branch is behind main. Rebase it onto main (no conflicts; skills/ is unchanged on main since the base) and force-push with lease.

Release note: v1.2.1 is a GitHub-only release; its Tessl publish failed at the review gate (run 35561430400), so the registry stays at 1.2.0 until this PR lands as 1.2.2.

Validation

Checks most contributors can run:

• python3 scripts/validate_skill.py skills/java-streams -> passed
• python3 scripts/validate_eval_criteria.py evals evals-reference evals-regression -> passed, 29 scenarios
• python3 -m py_compile scripts/*.py -> passed
• bash -n scripts/*.sh -> passed
• python3 scripts/validate_json_files.py -> passed
• python3 scripts/validate_openai_agent_yaml.py -> passed
• tessl plugin lint . -> passed
• Manual rendered-doc review

Tessl-authenticated checks:

• tessl review run --workspace martinfrancois --threshold 100 skills/java-streams/SKILL.md -> 100
• Hosted evals -> see the Hosted Evidence table

Human Verification

Each rewritten sentence was checked against the original for a dropped rule: none dropped; the find-first exception, the min/max caveat, the parallelStream quote, and the Java 24 gatherer shape are all still stated.

Review Checklist

• The change is scoped to the sections, skill files, evals, or workflows described above.
• Validation that applies to this change is checked above, or any unavailable check is explained.
• If Java stream guidance changed, Java baseline compatibility plus ordering, null handling, and parallelism were considered.
• If evals or benchmark claims changed, the eval scenarios remain fair and do not leak answer keys, run IDs, or fixed score claims into runtime references. (No eval change.)
• If runtime skill text or references changed, hosted checks were widened as described in docs/agents/workflow.md, or any Tessl blocker is documented.
• If a runtime skill/reference change was released, the final report includes the published main eval run plus post-change reference and regression run IDs, or a blocker issue for missing broad suites. (Not released.)
• Main and reference evals were run with both variants when hosted evals were needed; regression evals were run with context only unless reclassification back to reference was being checked. (Main both variants; reference with context only, to save credits.)
• New or moved eval scenarios follow the classifier recommendation, or the PR explains the maintainer-approved override. (None.)
• Every retained eval scenario has a 100% with-context result, or any below-100 result is documented as blocking follow-up rather than classified/reportable coverage.
• PR title or squash title uses Conventional Commits.
• Redaction checked: no tokens, private links, private eval artifacts, local host paths, or proprietary Java source.

AI Assistance (if used)

• AI-assisted PR
• I confirm I understand and reviewed the change (the October follow-up commits are not yet reviewed by the maintainer)