Discover and verify useful RSS sources

Why use this

1. A reading list becomes noisy when discovery is driven only by broad topic matches. A specialist reader needs to distinguish original reporting, occasional announcements and republished material before adding another stream to an already crowded reader.

2. This job could turn a vague interest into a shortlist whose relevance and technical accessibility can be reviewed together. Keeping rejected candidates and reasons would also prevent the same attractive but unusable endpoint from returning in every discovery pass.

Why an agent / simpler alternative

Useful judgment

Judge whether a publisher adds a distinct perspective and whether sample items match the requested subject.

Simpler alternative

Manually add known feeds when the desired publishers are already clear.

Unsuitable when

The reader needs licensed full-text redistribution rather than discovery of feed locations.

Proposed inputs

1. Name

   Reading brief

   Owner

   Reader

   Minimum scope

   Topics, language, exclusions and examples of useful articles

   Quality check

   Separate must-have subjects from casual interests.

2. Name

   Candidate feed samples

   Owner

   Research owner

   Minimum scope

   Publisher URLs, endpoint text and recent item metadata

   Quality check

   Keep the original URL and distinguish RSS/XML from an HTML landing page.

Access and setup — proposed or unknown

Proposed minimum

1. Read-only candidate pages or supplied feed snapshots; produce a shortlist before subscribing.

Compatibility unknowns

1. Unknown: runtime feed parsing, authentication and FreshRSS connector availability.

External actions requiring approval

1. Adding subscriptions, importing reader settings or redistributing article content.

Expected deliverable and review checks

Proposed artifact

Candidate table with publisher, endpoint, topic rationale, sample-item dates and verification status.

Critical checks

1. A successful HTTP response alone does not establish a usable feed.

2. Duplicate syndication and stale publication dates are visible.

Missing output evidence

This overlay adds no new output inspection or execution validation. Preserve the canonical implementation evidence and its outcome limits.

Failure signals and proposed responses

1. Signal

   The endpoint returns a styled webpage instead of feed entries.

   Why it matters

   A plausible URL can produce an empty subscription.

   Proposed response

   Mark format verification unresolved and retain the publisher page for manual discovery.

2. Signal

   Several sources reproduce the same stories.

   Why it matters

   The reading list grows without adding information.

   Proposed response

   Compare sample titles and original publishers; recommend one source or explain the overlap.

Effort and cost — unknown

Setup work

1. Define relevance criteria and an endpoint-status vocabulary.

Ongoing review

1. Recheck stale feeds and prune redundant coverage.

Unknowns

1. Unknown: reader limits, discovery usage charges and maintenance frequency.

First check — proposed, unexecuted

Status

proposed_unexecuted

Synthetic fixture

Invented snapshots contain one valid specialist feed, one HTML endpoint and two feeds sharing identical items.

Acceptance checks

1. Only the valid distinct source is recommended without qualification.

2. Invalid format and overlapping coverage receive different explanations.

Stop conditions

1. Live subscriptions or access to private reader history becomes necessary.

Limits

1. Editorial selection criteria do not establish continuing feed availability or permission to reuse articles.