[RaghavChamadiya]

What

Two Extract Method plans that Fix first hands out would not compile if applied as written:

• A staged Python plan (assemble_test_impact) whose first helper takes a shared parameter object: the Fix first step showed the helper header and call only, so the call named ctx and _AssembleTestImpactContext, which exist nowhere.
• A TypeScript plan (extractSources) lifting a loop body after if (tc.status !== "done" || !tc.result) continue;: the helper takes tc, and the narrowing of tc.result does not survive into it, so result.results and the host's later result.targets stop type-checking.

How

• Fix first steps carry the plan's notes. A step whose helper takes the parameter object shows the object's declaration above the helper and the line building it above the call. A plan missing either is not offered as a Fix first step. The agent prompt fences multi-line code and prints the notes, and the card lists them.
• The Extract Method planner refuses a TypeScript span when a test above or around it narrows a value no helper parameter can carry: a property the span reads, a property compared with a literal, or a module-level name. It also refuses a span holding an assert the code after it relies on, and a span assigning a name declared outside it that it neither takes nor returns. One refused stage refuses the whole staged split. Reasons are recorded as narrowing_lost and outer_declaration.
• A parameter narrowed by an enclosing if or an assert gets the existing "type it as narrowed" note. An output declared before the span gets a note to declare it in the helper.

Verification

Bar, set before measuring: re-plan every open Extract Method plan on a copy of our own index and on one held-out corpus (Zulip), blind-label at least 30 changed plans with opaque ids for "would this compile and preserve behaviour", and offer 0 broken plans afterwards.

• Re-plan, same commit and stored history, main against this branch: ours 746 to 736 open plans (723 unchanged, 11 changed, 2 new, 12 held); Zulip 309 to 299 (271 unchanged, 26 changed, 2 new, 12 held).
• Blind labels on the final plans: 57 offered plans (24 ours, 33 Zulip; every changed and new plan, every plan whose first helper takes a parameter object, plus a random 8): 57 compile and preserve behaviour, 0 broken.
• Earlier rounds of this branch surfaced 5 broken plans in the same sample (enclosing-if narrowing, an assert on a module-level name, an assert moved out of the host, two outer declarations). Those cases are the second commit.
• Held plans: all 24 (12 per corpus) labelled blind from their pre-change form. 8 were broken and 16 would have worked. This is the recall cost of refusing what the planner cannot prove, mostly property tests that TypeScript would not have needed.
• assemble_test_impact step applied by hand as now served: its 13 tests pass. extractSources as served before failed tsc with TS18048 and is now refused.
• Unit tests for the new rules, the step shape and the prompt rendering. Targeted Python suites pass (health, agent prompts, server, analysis, cli; the one local failure is the known Windows path-separator test in test_uninstall_cmd.py). UI vitest and tsc pass, and the generated HTTP types check passes.

HEALTH_ANALYZER_VERSION is not bumped: it was already bumped once since v0.58.0.

[repowise-bot]

🔍 1 thing to check

1. 2 files that usually change with this PR's files are not in it: tests/unit/health/test_refactoring_extract_method.py, a test, changed in 17 of the 26 commits that touched packages/core/src/repowise/core/analysis/health/refactoring/extract_method.py; packages/core/src/repowise/core/analysis/health/fix_first/build.py changed in 10 of the 14 commits that touched packages/types/src/fix-first.ts

➖ Health of changed files: 8.56 → 8.55 (-0.01; structure -0.02)
_{Average full file score (structure and history) of 13 changed code files, weighted by size before this PR; not the repository score.}

Check	Result
Coverage	98.6% of 220 changed lines covered · 3 uncovered in 2 files
Tests	8 of 9 changed code files reached by a test · 1 reached by no test · which tests
Security	No new findings on changed lines
Doc drift	No drift in docs this PR edits

Coverage: 3 uncovered changed lines

Patch coverage 98.6% (217 of 220 changed executable lines covered)

dc301a2 · cobertura · 6 of 15 changed files measured, 9 out of scope

File	Uncovered changed lines	Covered
packages/core/src/repowise/core/analysis/health/refactoring/narrowing.py	337, 339	109 of 111
packages/core/src/repowise/core/analysis/health/refactoring/extract_method.py	476	65 of 66

Report from this run for dc301a2.

🧾 Evidence: 1 to look at, 6 not known

Item	Status	Evidence
Components touched	ok · inferred	3 components: docs, packages, tests
Declared owners	ok · declared	@RaghavChamadiya, @swati510
Suggested reviewers	unknown · not known	Not known: not gathered on a pull request
Tests reaching the change	unknown · not known	0 of 9 changed files reached by a test; 9 not known
Contracts touched	ok · measured	No workspace contract touched; no breaking signature change
Consumers in other repos	unknown · not known	Not known: not gathered on a pull request
Decisions crossed	unknown · not known	Not known: decision records are read only for an AI review
Health delta	look · measured	Findings: 2 introduced, 0 worsened, 0 resolved
AI provenance	unknown · not known	Provenance: unknown, known for 0 of 3 commits; AI share withheld: coverage 0% is below 30%
Diff size rank	ok · measured	Larger than 75.6% of recent commits
Outcomes since merge	unknown · not known	Not known: a pull request is not a merged change

_{AI review is paused: this repository reached its monthly spending limit. Bot settings}

---

📊 See the full report for this PR
_{Blast radius, every caller of the contracts it changes, and health before and after. No sign-in.
Plain markdown for agents}

_{Settings and updates}

_{Updated 2026-10-22 03:24 UTC, since the last push: health 8.5 to 8.6, 5 findings gone
Silence one PR with [skip repowise] in the title · Bot settings
⭐ Star Repowise · 📥 Install on another repo}