Step 2: Country discovery

Builds a single candidate universe, then runs isolated research sub-agents (one per region batch) to check each country for both Remote and Sponsorship suitability together. No country is included based on reputation alone; every finding must be traceable to a real, checkable source.

Flow 

flowchart TD
  Start([Step 2 begins]) --> IncludeCheck{Include list\ngiven in Step 1?}
  IncludeCheck -->|yes| BaseInclude[Base = Include list]
  IncludeCheck -->|no| AskDefault([Show shipped default list,\nask: proceed with default,\nor provide your own? Wait.])
  AskDefault -->|proceeds with default| BaseDefault[Base = default-preferred-countries.md]
  AskDefault -->|provides own list| BaseOwn[Base = provided list]

  BaseInclude --> ExcludeStep
  BaseDefault --> ExcludeStep
  BaseOwn --> ExcludeStep

  ExcludeStep[Remove Exclude list\nfrom base] --> ShowList[Show final country list]
  ShowList --> Batch[Batch by region\nor by ~15 countries if unbatched]
  Batch --> Agents[One isolated sub-agent per batch,\nresearches Remote AND Sponsorship together]
  Agents --> Append[Append results to\ncf-step2-candidates.md]
  Append --> TZCheck{Max timezone\nprovided?}
  TZCheck -->|yes| MarkTZ[Mark out-of-range countries'\nRemote suitability as not suitable]
  TZCheck -->|no| Done
  MarkTZ --> Done([Continue automatically\ninto Step 3])

What it reads

  • Criteria and preferences from Step 1 (cf-step1-criteria.md)
  • profile.md and situational-profile.md
  • data/default-preferred-countries.md, shipped with the plugin, read only if no Include list was given

Part A: Build the candidate universe

No research needed; pure filtering logic, applied immediately:

  1. Base list: the Step 1 Include list if one was given; otherwise Claude shows you the shipped data/default-preferred-countries.md grouped by continent and asks whether to proceed with it or provide your own countries instead, waiting for your answer either way.
  2. If an Exclude list was given, those countries are removed from the resulting base.

Other than the default-list question above, this all happens silently; Claude never announces which source it’s using as the base or asks whether to use it. The resulting final country list is shown before research begins.

Part B: Research sub-agents

The final list is batched by region, since data/default-preferred-countries.md is organized under region headings; one sub-agent per region present in the final list, when the base came from it. If, for some reason, the list has no usable region grouping (e.g. the base came from your own Include list instead), it falls back to groups of roughly 15 countries. One isolated sub-agent is spawned per batch.

Each sub-agent’s prompt embeds, as literal text (never by reference), the exact countries in its batch, the candidate’s target role and skillset from profile.md, the minimum monthly salary from situational-profile.md (if specified), and the candidate’s citizenship and any noted immigration friction. This keeps the prompt self-contained, since sub-agents have no access to the conversation that computed these values.

Each agent checks both tracks for every country in its batch, not just one track:

Remote suitability: is remote hiring for this role a common, well-established practice? If a salary minimum was specified, does typical remote pay meet it? Evidence from remote job boards, hiring reports, salary surveys, or company remote policies.

Sponsorship suitability: does a real, named work-visa pathway exist where a local employer is the sponsor, a standard employer-sponsored visa or work permit tied to a job offer? Digital-nomad, remote-worker, long-stay, or self-qualifying visas don’t count, even if prominent in search results; they don’t involve a local employer sponsoring the candidate. Is the occupation on a shortage list? That’s a positive factor (and a hard requirement only for the minority of visas that explicitly say so); its absence alone doesn’t disqualify a country unless the identified pathway requires it. Citizenship-specific friction is factored in. Evidence from official immigration sources, visa program pages, or recruiter guides. A visa’s salary threshold is noted as a fact, never used by itself to decide suitability; that judgment against real market data happens later, in Step 5.

Each country gets independent Suitable / Reason / Source / Date findings for each track; “no evidence found” is reported explicitly rather than guessed. If isolation can’t be guaranteed, Claude shows all the batch prompts and waits for you to bring back the results manually.

Part C: Timezone marking

If a maximum timezone difference was provided in Step 1, Claude calculates each country’s UTC distance from your current location (from situational-profile.md) and overwrites the Remote “Suitable” value for any out-of-range country with a stated reason; it does not remove the country from the list, and Sponsorship suitability is never affected by timezone. Skipped entirely if no limit was given.

Output

  • cf-step2-candidates.md: every researched country, with independent Remote and Sponsorship findings (suitability, reasoning, source, date). Claude does not reproduce these per-country findings in chat; only a brief summary (countries researched, Suitable counts per track) and confirmation that the file is saved for Step 3.

Stop condition

Claude only stops here if Part A’s default-list question was triggered (no Include list given); it waits for you to confirm the default or provide your own countries. Otherwise, once cf-step2-candidates.md exists (and Part C has run, if applicable), Claude continues automatically into Step 3 within the same response, without waiting for a new message.

Claude Compass — MIT License