Skip to content

feat(capsule): state completeness at the top of every capsule - #46

Merged
wolfgang-aura merged 1 commit into
mainfrom
feat/capsule-completeness-status
Oct 5, 2026
Merged

wolfgang-aura merged 1 commit into
mainfrom
feat/capsule-completeness-status

Conversation

@wolfgang-aura

Copy link
Copy Markdown
Owner

Capsules now say up front whether they are complete, so an AI reading one can tell the user.

  • .llm.md: a new ## Completeness section under the header. It reads Status: COMPLETE or Status: INCOMPLETE with counts, plus one sentence asking the reader to tell the user and name what's missing.
  • HTML: the same status as a visible amber banner above the capture summary, shown only when incomplete. It is visible text on purpose. Hidden instructions for an AI would look like prompt injection.
  • Only real gaps count: main text, quoted posts, images, videos with neither file nor poster, video posters, and possibly truncated embedded posts. A video kept as poster plus source link, a quote tombstone, and a missing avatar are not gaps.
  • The Warnings list is deduplicated, and informational lines (companion-file video bytes, no transcript, duplicate hashes, tombstones) move to a separate Notes list. The manifest warnings array is unchanged.

Tests: new smoke check covering the incomplete and clean cases; npm test, lint and format check pass. The banner was checked by screenshot in the regenerated sample.

🤖 Generated with Claude Code

An AI reading a capsule had to find gaps in a warnings list below the
header, mixed with notes that are not gaps (video bytes kept in the
companion file, duplicate hashes) and with repeated lines. A model asked to
summarize would usually skip it.

The .llm.md now opens with a Completeness section: COMPLETE, or INCOMPLETE
with counts of what was not captured and what may be truncated, plus one
sentence asking the reader to tell the user. The HTML shows the same status
as a visible banner, only when incomplete. It is written as a stated fact
and not hidden, so agents that treat file contents as data still see it.

Only real gaps count. A video kept as poster plus source link, a quote
tombstone and a missing avatar do not. The Warnings list is deduplicated and
informational lines move to a separate Notes list; the manifest warnings
array is unchanged.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
@wolfgang-aura
wolfgang-aura merged commit 5b7f9be into main Oct 5, 2026
1 check passed
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