bestblogs-explain

bestblogs-explain — Explanation layer (profile / source contribution / scoring)

When to use

• The user asks "Why did BestBlogs recommend this one?"
• The user wants to see theirinterest profile summary。
• The user wants to knowwhich subscription sources contribute the most to my morning digest, as a basis for deciding whether to keep following them.
• The user wants to know a certain resource'sscore。

Primitives

Primitive	CLI	Description
bestblogs.interest_profile.get	bestblogs explain profile --json	Profile summary
bestblogs.source.contribution.list	bestblogs explain sources --days 7 --json	Contribution of followed sources to the morning digest over the last N days
bestblogs.resource.score.get	bestblogs explain score <id> --json	Resource score

Related fields (inlined in discovery-type responses)

Note: Each candidate in discovery-type interfaces already has inlined candidateSource / selectionReason / fallbackApplied / personalized. To explain "why this one was recommended", first look at these fields—you don't necessarily need to call the explanation interface separately. Only go through the explain primitive when you need more depth (profile dimensions / source level / scoring dimensions).

Actions

Profile summary

bestblogs explain profile --json

Fields of interest:

• onboardingCompleted
• languagePreference
• driftDetected — drift detection (when true, it is recommended to remind the user to reset their interests)
• explicitInterests[] — tags actively chosen by the user
• implicitInterests[] — behavior inference
• recentActiveTopics[] — active topics in the last 7 days

Source contribution

bestblogs explain sources --days 7 --json

Returns a source list sorted by count in descending order:

{
  "success": true,
  "data": {
    "days": 7,
    "contributions": [
      { "sourceId": "SRC_xxx", "sourceName": "阮一峰的网络日志", "count": 4 },
      ...
    ]
  }
}

Resource score

bestblogs explain score RAW_xxx --json

Returns including score / resourceType / language / qualified / sourceName。

Error handling

• 401 → not logged in (profile and sources require login;score can be anonymous for public resources)
• Empty list (contribution is 0) → tell the user "no sources were recommended in the last N days", and suggest adjusting --days or following more high-quality sources

Combination suggestions

• User complains "recommendations have been mediocre lately" → call in sequence profile(check drift) + sources(see which sources are dragging it down) + suggest the user go to bestblogs-subscription to unfollow sources with low contribution (this skill is not available in this period; prompt the user to do it on the website)
• When the Agent outputs candidates to the user: display candidateSource + selectionReason when the user asks "why does this one score high" explain score