feat(extract): wire connection detection into CLI with performance instrumentation (#239) - #252

In
`@packages/riviere-cli/src/features/extract/__fixtures__/extraction-test-fixtures.ts`:
- Around line 54-58: parseExtractionOutput currently throws TestAssertionError
with only result.error.message while parseFullExtractionOutput throws a richer
message including a custom JSON preview; make their error formatting consistent
by changing parseExtractionOutput (and the other parse helper at the other
location) to throw TestAssertionError with a combined message that includes both
the structured Zod error (result.error.message) and a short JSON preview of the
invalid input (use JSON.stringify(parsed) truncated/previewed), preserving the
existing TestAssertionError usage and the extractionOutputSchema/result
identifiers so the message mirrors parseFullExtractionOutput’s format for
uniform debugging.

In
`@packages/riviere-cli/src/features/extract/entrypoint/extract.connections.spec.ts`:
- Around line 68-105: Add a file-level afterEach hook to restore mocks so
vi.spyOn(console, 'error') in the tests does not leak into other tests;
specifically, add an afterEach that calls the Vitest restore function (e.g.,
vi.restoreAllMocks()) near the top of the test file (alongside createTestContext
/ setupCommandTest) so any spying on console.error is cleaned up after each
spec.

In `@packages/riviere-cli/src/features/extract/entrypoint/extract.ts`:
- Around line 51-52: The CLI --patterns flag is parsed but not used; add a
boolean patterns property to the ConnectionDetectionOptions type in
ConnectionDetectionOptions (used in detect-connections.ts), propagate that field
through detectConnectionsQuery (ensure the query builder accepts and forwards
options.patterns), and update detectConnections (the function that inspects
source files) to branch into pattern-based detection when options.patterns is
true (apply the existing pattern matching logic or add pattern matching steps
there and ensure return shape matches current ConnectionDetectionResult). Also
update any callers that construct ConnectionDetectionOptions to pass the parsed
patterns flag through.

In
`@packages/riviere-cli/src/features/extract/queries/detect-connections-query.ts`:
- Around line 10-13: DetectConnectionsInput duplicates the existing
ConnectionDetectionOptions type; replace the local interface by importing and
reusing ConnectionDetectionOptions so changes stay in sync. Update
detect-connections-query.ts to import the type (e.g., import type {
ConnectionDetectionOptions } from the extract package) and change any references
from DetectConnectionsInput to ConnectionDetectionOptions (remove the duplicate
interface declaration and adjust function/method signatures that use
DetectConnectionsInput such as the detect-connections query entry point).

In
`@packages/riviere-cli/src/platform/infra/cli-presentation/format-extraction-stats.spec.ts`:
- Around line 128-139: Rename the test to reflect that it checks truncation to
zero rather than precision: update the test description string in the spec that
calls formatTimingLine from "formats timing with sub-millisecond precision" to
something like "formats timing by truncating sub-millisecond values to 0.00s"
(or "truncates sub-millisecond timings to 0.00s") so the intent matches the
assertion.

In
`@packages/riviere-extract-ts/src/domain/connection-detection/detect-connections.ts`:
- Around line 47-50: The timing variable currently named filteringMs incorrectly
includes both ComponentIndex construction and file filtering; change the name
(e.g., indexingAndFilteringMs or componentIndexAndFilteringMs) to reflect that
it measures the duration between performance.now() assigned to filteringStart
and the later timestamp after new ComponentIndex(components) and
computeFilteredFilePaths(...) complete, update the variable declaration and all
usages, and also update the ConnectionTimings type/field names to match the new
identifier so semantics remain correct (references: filteringStart,
ComponentIndex, computeFilteredFilePaths, filteringMs, ConnectionTimings).

In `@packages/riviere-cli/src/features/extract/entrypoint/extract.ts`:
- Line 82: Replace the implicit-return arrow callback passed to
formatDryRunOutput(...).forEach so it uses a block body that explicitly calls
console.log; locate the expression using formatDryRunOutput and change the
forEach callback from an expression-bodied arrow to a block-bodied arrow that
calls console.log(line) inside the block to silence the lint warning.
- Around line 139-143: The stats printing uses a concise arrow in
formatExtractionStats(stats).forEach((line) => console.error(line)) which
creates an implicit return; change the forEach callback to a block body to avoid
that by rewriting the callback to use braces and an explicit statement (for
example: forEach((line) => { console.error(line) })); update the call site that
references options.stats, countLinksByType, enrichmentResult.components.length,
and formatExtractionStats to use the block-bodied arrow so the console.error is
executed as a statement rather than implicitly returned.

In
`@docs/continuous-improvement/post-merge-reflections/2026-02-05-issue-239-m1-d1-8-d1-9-wire-connection-e.md`:
- Line 1: The document title and its storage directory are inconsistent: the
top-level heading "# Pre-Merge Reflection:
issue-239-m1-d1-8-d1-9-wire-connection-e" should match the directory name
"post-merge-reflections"; either change the heading to "Post-Merge Reflection:
issue-239-m1-d1-8-d1-9-wire-connection-e" to keep the file in its current
directory, or move the file into a "pre-merge-reflections/" directory if the
pre-merge label is intentional—update the heading or relocate the file
accordingly so naming is consistent.

In @.claude/commands/pre-merge-reflection.md:
- Around line 198-201: The document contains a duplicated heading "Pipeline
Improvement Proposals" (the detailed subsection and the later top-level
section); consolidate by removing or renaming the later instance to avoid
confusion—e.g., rename the second heading to "Summary of Improvement Proposals"
and update any cross-references or the surrounding explanatory text so it
clearly states that one is the detailed template (the original "Pipeline
Improvement Proposals" subsection) and the other is a short summary, ensuring
consistency in naming and intent wherever the header "Pipeline Improvement
Proposals" appears.

In @.claude/commands/pre-merge-reflection.md:
- Line 84: This is inconsistent: the command and file are named
/pre-merge-reflection and pre-merge-reflection.md but the template header and
output path use "Post-Merge"; update the template header "Post-Merge Reflection"
to "Pre-Merge Reflection" and change any output directory references from
post-merge-reflections/ to pre-merge-reflections/, and search for and update any
other references to "Post-Merge Reflection" so all uses (file
pre-merge-reflection.md, command /pre-merge-reflection, header text, and output
directory) consistently say "Pre-Merge Reflection".

In
`@docs/continuous-improvement/post-merge-reflections/2026-02-05-issue-239-m1-d1-8-d1-9-wire-connection-e.md`:
- Line 30: Check the pipeline timeline vs the narrative: confirm whether there
were actually 4 architecture review iterations for the CS-010 rule or only 3
(review `#3`, review `#7`, review `#10` shown). If there were only 3, update the
sentence that says "4 architecture review iterations" to "3 architecture review
iterations" and adjust the explanatory sentence about detectConnections
accordingly; if there was a fourth review outside the table, add that missing
review entry to the timeline (with its identifier and brief note) or append a
parenthetical clarifying that one review occurred off-timeline. Reference the
detectConnections mention and the sentence claiming "4 architecture review
iterations" when making the edit.
- Around line 180-186: The "Recommended Follow-Ups" list has inconsistent status
markers: items "AP-009: Inconsistent Patterns Between Related Functions" and
"RFC-016: Mock Cleanup After vi.spyOn" use ✅ but the remaining items ("RFC:
'Prefer for-of over forEach for side-effect-only iteration'", "Create task: Add
pre-implementation placement check to task workflow", "Create task: Add 'single
function call = inline in entrypoint' guidance to separation-of-concerns") lack
any indicator; update the list under the "Recommended Follow-Ups" heading to use
a consistent checkbox convention (e.g., ✅ for completed and ❌ or ⬜ for pending)
and apply it to each list item so every entry clearly shows completion status
(adjust the three pending items to start with ❌ or ⬜).