Match an identifier against the local registry, tolerating vague input.

Always check here before researching a model: regional Bedrock inference-profile prefixes, version suffixes, spacing, and vague platform names are normalized deterministically. status is resolved, needs_confirmation (show candidates to the user and ask), or not_found.

Return a validated local registry model profile.

Interpret reviewed lifecycle facts at a fixed date without live research.

Return explicit same/different/unsupported/unknown comparison states.

Hard-filter and deterministically rank compatible registry models.

Scan a local Python application into a normalized coupling inventory.

prompt_sources optionally names prompt files (relative to the application root) that automatic discovery missed; they become explicit high-confidence prompt sources.

Opt in to an OpenRouter pricing inquiry; canonical registry facts are unchanged.

Estimate recurring token cost from checked-in canonical pricing.

Conservatively analyze prompt characteristics without inference.

Prepare a prompt migration specification without rewriting the prompt.

Statically validate prompt assumptions against target registry facts.

Analyze provider invocation and adjacent tool/output couplings.

Prepare a target invocation contract without editing application files.

Generate an actionable application-level migration manifest without writing files.

Generate a human-readable report from the integrated migration workflow.

Bind a deterministic evaluation corpus to a migration manifest.

Run source and target with user-owned local provider credentials.

Compare one paired source/target evaluation result deterministically.

Return a structured categorized regression report for an evaluation run.

Return bounded, review-only recommendations from observed regression evidence.

Plan a registry update from supplied evidence without editing canonical files.

Scan locally and bound explicit user-scoped research to missing/stale facts.

Run the deterministic scope/policy gate over one research artifact.

Check that an independent evidence review actually covers the research artifact.

Deterministically combine research and review; agent agreement is not evidence.

Finalize a run workspace into an immutable, expiring session overlay.

Research and review artifacts must already exist in the workspace; any stage still needing an agent fails closed instead of doing hidden work.

Plan a migration over the canonical registry plus one run's session overlay.

Start a guided migration run; the preferred entry point for a full migration.

Matches both models against the local registry first (tolerating vague or platform-decorated identifiers). If either identifier needs confirmation, nothing is written and the result carries candidates to show the user. Otherwise the run workspace is created (default <application>/.llm-migrate/runs/<run-id>/, or output_dir when given), a bounded research request is written only when knowledge is missing or stale, and next_steps says exactly which tools to call next.

Ready-to-run researcher and reviewer prompts for a run's research request.

Renders one bounded, scope-isolated prompt pair per remaining scope from <run_dir>/request.yaml. Run each prompt with a separate agent (never let one agent research two scopes or review its own research), write the YAML artifacts to the stated paths, validate them, then call build_session_registry.

Per-file adaptation worklist for a started migration run. Call it once.

Derived from the run's migration plan (using its session overlay when one was built). shared_prompt_guidance applies to every prompt task; each task carries only its own guidance. Adapt each prompt minimally per that guidance and call submit_adapted_prompt; write each complete adapted file and call submit_adapted_file (or pass unchanged=true when no change is needed). Submission results confirm acceptance, so re-listing between submissions is unnecessary; finalize_migration reports remaining gaps.

Questions plus evidence-backed options for every unresolved blocker.

For each blocker of a started run, returns the question to ask the user and 2-5 registry-backed options (retarget / redesign / correction / accept-with-rationale), each with consequences, evidence URLs, and the exact record_blocker_decision call to make once the user chooses. Present questions, options, and evidence VERBATIM, one blocker at a time; never choose on the user's behalf. Also reports how previously recorded decisions applied, including stale ones.

Record the user's decision for one blocker; decisions are durable.

blocker_id and option_id must come from get_blocker_resolutions. Accept options REQUIRE the user's own free-text rationale (they are refused without one and are never a default). Retarget and correction decisions update the run's migration.yaml identity immediately; redesign decisions inject a required, evidence-linked adaptation task on every plan regeneration; accept decisions downgrade the blocker to a prominently reported accepted decision. The result lists the remaining unresolved blockers and the exact next step.

Validate and store one refined prompt for the target model.

Adapt minimally: keep the original wording and structure except where a listed model difference or evidence-linked guidance item requires a change, and cite that evidence in changes. Validation runs on the DECODED runtime prompt values, and a submission whose decoded values equal the original's — or differ only by whitespace or letter case — is rejected: serialization tricks, escapes, quoting and cosmetic edits are never an adaptation. If the prompt needs no change, pass unchanged=true with an EMPTY adapted_prompt (combining it with content is an error) to record a reviewed no-change deliverable; the claim is refused when the prompt's decoded values still reference the source model. Blockers are rejected, and a submission that drops the original's structural sections (XML-like tags or prompt components) is rejected unless allow_restructure is true and the justification is recorded in changes. Accepted prompts are written beneath <run>/output/prompts/ mirroring the application layout, and the rationale/changes appear verbatim in the final migration report.

Store one finalized post-adaptation application file for review.

adapted_content must be the complete file, not a diff. Submissions are checked deterministically (path containment, Python syntax, model-id consistency) and written beneath <run>/output/files/; the application tree itself is never modified. If the file genuinely needs no change for the target model, pass unchanged=true with an empty adapted_content: the original is copied as the deliverable and the coverage gap closes without resending the file. Prompt source files are rejected here; submit them with submit_adapted_prompt instead.

Write the run's manifest, report, and adaptation change log under output/.

The report includes, per adapted file, what changed and why, plus the affected files that still lack an adaptation deliverable. Re-run it any time; it always reflects the current submissions.