Skip to content

feat(discover): doc-approval review chain as the multi-capability goal - #77

Merged
enricopiovesan merged 1 commit into
mainfrom
discover-docapproval-goal
Sep 10, 2026
Merged

feat(discover): doc-approval review chain as the multi-capability goal#77
enricopiovesan merged 1 commit into
mainfrom
discover-docapproval-goal

Conversation

@enricopiovesan

Copy link
Copy Markdown
Collaborator

Feedback on the goal set: the examples weren't representative, and the no-plan card read as broken.

New goal set

Goal Chain Status
Price a quote (goal 1, auto-plans on load) core.calculate-price@1.2.0 ✅ executes now
Review a document for approval doc-approval.analyze → doc-approval.recommend — document text → extracted type / parties / amounts / confidence → approve-or-route recommendation + rationale plans now, executes once registry#424 + registry#425 land (doc-approval.*@1.4.0 with a pure_read risk class)
Check a card number validation.validate-luhn@1.2.0 ✅ executes now

Verified: browserLocalPlan chains analyze → recommend from a { document } fact (mappings: facts.document → analyze, then all 5 of analyze's outputs → recommend's inputs). Until the @1.4.0 capabilities are on the live catalog, clicking that card plans and then fails closed on the per-node authorization check — honest, not a crash, and it starts working automatically when the registry PRs merge (no further website change needed; a follow-up just promotes it to goal 1 + adds its E2E).

Dropped: the abstract uncertainty.score chain and the no-plan card (renderNoPlan stays as defensive handling).

Supersedes the still-open #76 — same intent, better multi-capability example. Close #76 in favour of this.

Verification

npm test 5/5 (+1 networked skip) · npm run build clean · npm run test:e2e 2/2 (price + luhn plan+execute).

🤖 Generated with Claude Code

…o-plan card

Feedback: the examples weren't representative and the no-plan card read as
broken. New three-goal set, all real:

- "Price a quote" — core.calculate-price@1.2.0 (goal 1, auto-plans on load).
- "Review a document for approval" — doc-approval.analyze -> doc-approval.recommend,
  a named-reference-app 2-capability governed chain (document text -> extracted
  type/parties/amounts/confidence -> approve-or-route recommendation + rationale).
  Verified: browserLocalPlan chains analyze -> recommend from a {document} fact.
  EXECUTES once doc-approval.analyze@1.4.0 / recommend@1.4.0 land on the live
  catalog with a pure_read risk class (traverse-framework/registry#424, #425);
  until then this card plans and then fails closed on the authorization check.
- "Check a card number" — validation.validate-luhn@1.2.0.

Dropped: the abstract uncertainty.score chain and the no-plan card. renderNoPlan
stays as defensive handling for a live goal that returns zero candidates.

Supersedes the still-open #76 (same intent, better multi-cap example).
E2E updated: goal-1 (price) + luhn plan+execute; doc-approval spec added when
the registry PRs merge. npm test 5/5 (+1 skip), test:e2e 2/2, build clean.

Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
@enricopiovesan
enricopiovesan merged commit 160d36d into main Sep 10, 2026
5 checks passed
@enricopiovesan
enricopiovesan deleted the discover-docapproval-goal branch September 10, 2026 02:00
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