docs: align bridge framing with revised single-dimension certification model#1795
Conversation
…n model Maintainer walked back the two-badge proxy-vs-state-local split from #1782 and proposed a single-dimension framing (Wire Conformance / Live Integration Verified) where every seller faces the same verifiability gap, and the bridge is one of two mechanisms for closing the seed→read loop, not a special path for one seller class. JSDoc on AdcpServerConfig.testController: - Drops "only upstream-proxy sellers" as primary framing - "Pick by where your read handlers fetch from, not by seller class" - "Either path earns wire-conformance credit; it is *not* a separate certification category" skills/build-seller-agent/SKILL.md: - New "Test surfaces" section frames the verifiability gap as universal - Names the two implementations (state-local store vs TestControllerBridge) - Hedges on certification names while #1782 settles No SDK behavior change. Marker contract, trust-boundary docs, and dual-emit warn from #1786/#1787/#1788 all stay as-is — they describe mechanism without committing to certification taxonomy. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
…n + in-repo JSDoc link
Docs review pass — folded in commit c687bc7Docs-expert review caught one material edit and one minor: Material: the new SKILL.md section had three paraphrases of the same decision rule (bullets, then symptom-form paragraph, then hedge blockquote) — an LLM-agent reader would spend tokens reconciling them and might tag the section "not actionable yet." Fixed:
Minor: JSDoc had no in-repo cross-link — IDE-hover readers had to jump to GitHub for the wiring mechanics. Added Docs-expert verdict: lands-well; JSDoc rewrite was already materially better than what it replaced. SKILL.md hedge framing is now scoped to certification names, not mechanism. CI re-running; admin merging once green. |
Summary
Maintainer walked back the two-badge proxy-vs-state-local-seller split from #1782 and proposed a single-dimension framing: every seller faces the same verifiability gap; the bridge is one of two mechanisms for closing the seed→read loop, not a special path for one seller class. This PR folds the revised framing back into the JSDoc on
AdcpServerConfig.testController(which currently names "upstream-proxy sellers" as the audience) and intoskills/build-seller-agent/SKILL.md.Also posted matching revision to the upstream taxonomy proposal at
adcontextprotocol/adcp#4593.Changes
JSDoc (
src/lib/server/create-adcp-server.ts:1676-1707):Seller-agent skill (
skills/build-seller-agent/SKILL.md):TestControllerBridge), explains the decision ruleNo SDK behavior change
The marker contract from #1786, the JSDoc trust-boundary section from #1787, and the construction-time dual-emit warn from #1788 all stay as-is — they describe mechanism without committing to a certification taxonomy. Only the framing prose around "what kind of seller this is for" shifts.
Test plan
tsc --noEmitcleanprettier --checkclean🤖 Generated with Claude Code