feat: add RCA commands and surface root cause analysis in checks get [AI-190]

[thebiglabasky]

Summary

RCA in checks get

• Add RCA column (Yes/-) to the error groups table in checks get <id> so users can see at a glance which error groups have root cause analysis available
• Add Error Group ID column to the error groups table for easy reference when triggering RCA
• Render full RCA detail (classification, root cause, user impact, code fix, evidence, references) when drilling into checks get <id> --error-group <egId>
• Transform JSON output to expose latestRootCauseAnalysis (singular, most recent) + rootCauseAnalysisCount instead of the raw array, to avoid surfacing potentially contradictory historical analyses to agents

RCA trigger and get commands

• Add checkly rca run --error-group <id> to trigger a root cause analysis, with --watch to poll until complete
• Add checkly rca get <rcaId> to retrieve an existing analysis, with --watch to wait if still generating
• Both commands support --output detail|json|md formats
• Polling uses HTTP 202 to detect in-progress analyses (404 is now a real "not found")

RCA API client

• Add rca.trigger(errorGroupId) — POST to /v1/root-cause-analyses/error-groups/{errorGroupId}
• Add rca.get(id) — GET from /v1/root-cause-analyses/{id}

Bug fix

• Fix Windows CI failure in init command test — path assertion was hardcoded with Unix separators

Screenshot

Test plan

• Run checkly rca run --error-group <egId> --watch — verify it triggers and polls until complete
• Run checkly rca get <rcaId> — verify it retrieves a completed analysis
• Run checkly rca get <rcaId> --watch — verify it polls on 202 and displays result on completion
• Run checkly checks get <checkId> — verify RCA and Error Group ID columns appear in error groups table
• Run checkly checks get <checkId> --error-group <egId> with an error group that has RCA — verify full RCA detail renders
• Run checkly checks get <checkId> --output json — verify latestRootCauseAnalysis and rootCauseAnalysisCount in error groups
• npx vitest --run passes all tests
• Windows CI passes

🤖 Generated with Claude Code

[MichaelHogers]

Three observations from testing with CC:

1. --result is a dead end. No link back to the error group or its RCA. Adding Error group: checkly checks get --error-group would help, especially for agents navigating programmatically. (need to double check this manually)

2. RCA: - is ambiguous. Could mean not entitled, not triggered, or not available. For non-entitled accounts the API just returns an empty array — user has no idea RCA exists. A tip pointing to checkly account plan when the entitlement is missing would be a natural upsell moment.

3. CLI can't trigger RCA (future). The public POST endpoint exists but CLI is read-only. A --analyze flag would close the loop.

2 is worth looking into imo ? might be easy to pull in given the earlier plan/entitlement work

[thebiglabasky]

Three observations from testing with CC:

1. --result is a dead end. No link back to the error group or its RCA. Adding Error group: checkly checks get --error-group would help, especially for agents navigating programmatically. (need to double check this manually)
2. RCA: - is ambiguous. Could mean not entitled, not triggered, or not available. For non-entitled accounts the API just returns an empty array — user has no idea RCA exists. A tip pointing to checkly account plan when the entitlement is missing would be a natural upsell moment.
3. CLI can't trigger RCA (future). The public POST endpoint exists but CLI is read-only. A --analyze flag would close the loop.

2 is worth looking into imo ? might be easy to pull in given the earlier plan/entitlement work

I'll check 1.

2 is a good point: it is verified by the agent in general, but worth distinguishing if possible. The problem I anticipate is that this requires an extra call to your entitlements to have that context, and that's kind of parallel to the call getting the error groups... So I don't necessarily think there's a simple solution to this. I'd lean towards leaving as is, and the attempt at calling the RCA itself, returns a 402 payment required hence push to upgrade as agents seeing this should naturally turn to the account plan command?

3 is done in a separate branch sitting on top of this one already 🤓

[thebiglabasky]

@MichaelHogers re: 1

The fields exist in the database but are not exposed in the public API — the Joi schema with stripUnknown strips them out.

So for linking --result back to an error group, there are two options:

A) Add errorGroupId/errorGroupIds to the public check results schema — small API change, adds the field to the response. Then the CLI can show "Error group: checkly checks
get --error-group " in the result detail navigation hints. This is a backend change.

I created an issue for it. So I'll need to make a BE change before we can fix this. I think it's worth doing so the command makes sense.