RxJS 7 to RxJS Next migration

Treat this as reviewed project work, not a codemod run. The objective is an intentional RxJS Next contract backed by the project's own build and behavior evidence. A transformed file, a clean diff, or one passing test is not proof that the migration is complete.

Work through all eight stages in order. Re-enter an earlier stage whenever new evidence changes scope, coverage, or lifecycle intent. Keep the developer involved at every explicit pause.

Non-negotiable rules

• Read and obey repository-local instructions before acting. Protect existing work and stay within the authorized repository and paths.
• Establish the RxJS 7 baseline before changing dependencies or source.
• Never infer platform sharing versus producer-per-direct-subscription behavior from syntax, filenames, operator names, or a currently passing output.
• Use the installed engine's versioned registry and schemas as authority. Do not reconstruct capability mappings from this Skill or from memory.
• Run the deterministic engine without writes first. A refusal is a successful safety outcome, not permission to improvise a mechanical rewrite.
• Preserve the project's test framework by default. Select a framework adapter only when the developer asked for that separate conversion.
• Do not weaken, skip, delete, or replace a behavioral expectation merely to make the migrated suite green.
• Do not add RxJS 7 compatibility shims, string-named platform methods, local substitute operators, or runtime generators unless the developer separately authorizes that product work.
• Keep changes in small coherent batches. Review and verify each batch before widening the scope.
• Do not call a migration complete while a required command is unknown, unapproved, skipped, or red without an explicitly accepted blocker.

Working record

Create or locate a checked-in migration contract manifest and a concise human-readable report. Use the schema exported by the installed @rxjs/migrate; schema validity and migration readiness are separate checks. Use assets/migration-report.md for the report.

For every migration unit, preserve source locations and record:

• its current RxJS 7 behavioral claim;
• the target lifecycle selected by the developer;
• its evidence classification and supporting tests;
• approval state, diagnostics, and any intentional divergence; and
• verification or a named blocker.

Stage 1: Establish authority and scope

1. Locate repository instructions, package and workspace metadata, lockfiles, CI configuration, RxJS versions, build/test commands, and current working tree state.
2. Identify the repository root, migration boundary, permitted writes, package manager, network policy, and candidate verification commands.
3. Record the installed @rxjs/migrate version and canonical Skill digest. If the installed Skill does not match its package, stop and synchronize it before trusting these instructions.
4. State the proposed read, write, dependency, and command scope before broad changes.

Pause for the developer if the target repository or allowed scope is unclear, existing changes overlap the migration, or destructive, networked, external, or broader actions require new authority.

Exit only when the scope, allowed writes, package manager, and candidate gates are recorded.

Stage 2: Assess usage, lifecycle risk, and coverage

Inventory installed RxJS versions and import forms, then locate Observable construction, custom producers, Subjects, subscription ownership, repeated subscriptions, retry/refresh/cache paths, schedulers and timing, test helpers, interop inputs, cancellation, teardown, and error handling. Include direct and transitive operator usage.

For each lifecycle-sensitive path, link exact source evidence and mark it as:

• covered by an existing baseline;
• requiring a characterization test;
• explicitly unsupported; or
• an uncovered risk the developer accepts.

Produce a risk and coverage report before proposing rewrites. Read references/assessment-and-contract.md for the inventory and characterization protocol.

Exit only when every in-scope lifecycle-sensitive use has one recorded coverage disposition. Missing coverage is not evidence that behavior is safe.

Stage 3: Establish the green RxJS 7 baseline

Run the agreed build, type, lint, unit, integration, and behavior checks against the unchanged RxJS 7 project. Record exact commands, environment facts, exit codes, and concise results.

Add focused characterization tests where Stage 2 found material gaps. Protect observable values, completion/error behavior, producer multiplicity, cancellation, abort reasons, teardown ordering, timing, and externally visible subscription effects as applicable. Every characterization test must pass on RxJS 7 before migration begins.

Pause if a starting gate fails. Diagnose it and ask the developer whether to fix it first, narrow scope, or record it as an accepted pre-existing failure. Never silently reclassify it as a migration regression or ignore it.

Exit only with a green baseline or explicit approval for each named pre-existing failure, plus green characterization tests.

Stage 4: Classify and approve the target contract

Divide the scope into stable migration units. For each unit, record one target lifecycle using the values accepted by the installed manifest schema and one evidence classification. Use repository evidence and references/assessment-and-contract.md to explain the choice.

The engine may identify risk but cannot choose lifecycle intent. unresolved is a stop state. Pause for the developer when either platform-shared behavior or producer-per-direct-subscription behavior is plausible, or when Subject semantics, cancellation, scheduler ordering, error behavior, interop, or a public behavior change is not proved.

Intentional divergences require a concrete old claim, proposed Next claim, user impact, evidence, and developer approval. Validate the manifest structurally, then run the installed readiness assessment. Do not proceed with a batch containing unresolved intent or required pending approval.

Exit only when the developer has approved every ambiguous choice and intentional divergence needed for the next batch.

Stage 5: Plan and dry-run bounded engine changes

Inspect the installed capability registry and select only entries whose version, source form, arity, preconditions, and executable evidence match the approved unit. Do not infer support from a familiar name. Follow references/engine-and-batches.md. When the batch contains RxJS 7 marble tests, also read references/test-migration.md.

Run rxjs-migrate without --write using explicit provenance and, when required, the approved cold or platform mode. Preserve the current test framework unless a supported adapter was separately selected. Capture the versioned JSON report, changed paths, structured diagnostics, and exit code.

Pause when a diagnostic requires review or refuses a transform, when output would escape authorized paths, when source does not match the capability evidence, or when batch size or semantic reach exceeds the reviewed scope. Never conceal unsupported segments inside a partially transformed pipeline.

Exit only after diagnostics and changed-file scope are reviewed and the developer has approved any newly broadened batch.

Stage 6: Apply one small migration batch

Apply only the reviewed dry-run plan. Require an explicit write destination inside the authorized project and preserve source provenance. Do not overwrite existing output without specific approval. Review the resulting diff before additional edits.

Format through the project's normal tools. Ensure ordinary project-owned source and direct test declarations remain; do not leave dynamic generators, hidden case registries, or compatibility assertion layers in the result.

Immediately check parsing, the narrowest useful type boundary, and engine idempotence. Re-running the same transform must produce no new source change and stable diagnostics. If it does not, restore the affected batch to the reviewed starting point and report an engine defect.

Exit only when the batch parses, type-checks at its narrow boundary, is idempotent, and every diagnostic is resolved, carried forward, or escalated.

Stage 7: Build, test, diagnose, and repair

Run focused characterization and migrated tests first, then the agreed project gates. Classify every failure as a migration defect, RxJS Next product gap, intentional divergence, environment/baseline issue, or still-unknown result. Use references/verification-and-closeout.md for the repair rules.

Repair migration defects in a small batch and repeat the relevant gates. For a product gap, preserve the failing evidence and report it; do not disguise it with a local replacement. For an intentional divergence, change an expectation only after the manifest records approved new behavior and user impact.

Pause before any repair that changes public behavior, weakens evidence, expands writes, changes dependencies beyond the reviewed plan, needs external access, or introduces a compatibility layer.

Exit only when agreed gates pass or each remaining red result has a named, evidenced, explicitly accepted blocker.

Stage 8: Close out and hand off

Update the contract manifest and report with exact source and target versions, engine version, capability-registry version, canonical Skill digest, baseline, units, diagnostics, divergences, final commands/results, and blockers. Validate the manifest with the installed schema and assess readiness separately.

Review references/review-checklist.md. Report:

1. what behavior and lifecycle contract changed;
2. what source and tests changed;
3. the exact gates run and their results;
4. migration defects repaired versus product gaps retained;
5. approved divergences and environment limitations; and
6. remaining blockers, owners, evidence, and acceptance.

Claim completion only when readiness is ready, or ready-with-accepted-blockers when the developer intentionally chose that outcome. Otherwise hand off the migration as incomplete with the next concrete decision or action.