Skip to content

feat(verify): read a bare page address as the document itself - #4

Merged
Yigtwxx merged 1 commit into
mainfrom
page-target
Sep 17, 2026
Merged

Yigtwxx merged 1 commit into
mainfrom
page-target

Conversation

@Yigtwxx

@Yigtwxx Yigtwxx commented Sep 17, 2026

Copy link
Copy Markdown
Owner

What this changes

A single URL pasted into the TUI bar or given to check --url on any host that is not a social platform is now read as the document itself: fetched up the fetch ladder, cut into one paragraph per block, and cited by its links unless it prints a reference list. Bluesky / Hacker News / Reddit / Mastodon posts are read as before; a profile, subreddit or front page keeps the "a page, not a post" answer.

Why

The bar's own placeholder says "paste a file path or URL", but every non-social address was refused before any request (IngestError: … is a page, not a post) and the run showed as failed. Design and decisions: docs/superpowers/specs/2026-09-17-page-target-design.md.

  • A page the ladder cannot read ends the run in the ladder's own words (UNVERIFIED (blocked, browser not permitted): step 3 browser: not permitted (to permit it: config set permissions.install_browser ask)), never a verdict (rule 2).
  • A page with no links is reported, not passed as clean; a page whose citation markers were set aside for its links says so (MARKERS_SET_ASIDE) (rule 6).
  • The page's own address — anchor, tracking query, printed permalink, redirect target or Wayback copy — is never one of its sources (rule 1).
  • The network permission is read before the page is fetched (rule 4).
  • lxml becomes a direct dependency (already pulled in by scrapling); lxml-stubs in dev.
  • New open items 16.1–16.5 (no cap on link-heavy pages; bibliography-heading regex widening deferred to an eval run).

Checklist

  • ruff check and ruff format --check are clean
  • pytest passes
  • Type annotations on every new function signature
  • No network access in unit tests
  • If a new reported state was added, spec section 15 is updated too — no new state; every refusal reuses an existing Outcome
  • This change does not make the tool sound more confident than its evidence

🤖 Generated with Claude Code

https://claude.ai/code/session_017G4e6oqMrpALsNEQpHoopg

A single URL pasted into the TUI bar or given to `check --url` was routed
as a social post and refused for every host but the four platforms with a
reader, although the bar itself invites a URL. A non-social address is now
the page at that address: fetched up the same ladder its sources are, cut
into one paragraph per block, and cited by its links unless it prints a
reference list. A page the ladder cannot read ends the run in the ladder's
own words with the setting to type; a page with no links is reported, not
passed as clean; the page's own address is never one of its sources, by
anchor, tracking query, printed permalink, redirect or Wayback copy.

Spec: docs/superpowers/specs/2026-09-17-page-target-design.md

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_017G4e6oqMrpALsNEQpHoopg
@Yigtwxx
Yigtwxx merged commit 8dd98c3 into main Sep 17, 2026
7 checks passed
@Yigtwxx
Yigtwxx deleted the page-target branch September 17, 2026 12:24
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant