Repository navigation
Commit verified reasoning and input-modality values to the USAi catalog #541
Description
After a recorded live verification, commit verified reasoning / inputModalities values to the USAi catalog and shipped snapshots, with sources.capabilitiesVerified set to <YYYY-MM-DD>; opencode <version>. Models that fail or were not tested stay absent (unknown). Never render text-only by default.
Blocked on: schema adoption by consumers, and the live image verification run, both tracked as follow-ups to #366. Trigger: both are complete.
AI-assisted tracker entry (OpenCode). Human review required.
Activity
mogul commentedon Oct 25, 2026
Design note from review:
sourcesis rebuilt on everybuild-catalogrun and the builder never emitscapabilitiesVerified, so a hand-added verification record and any committed capability values are erased by a plain regeneration (the guard test then fails, which is the safe direction). This issue must decide how verified capabilities survive regeneration, for example a reviewed verified-capabilities input file that the builder merges, and update the guard test to match.mogul commentedon Nov 5, 2026
🤖: AI agent (First Mate), posting on behalf of @mogul.
Closing as delivered by #575 (merged 2026-11-05, released in Patterns v1.19.0).
integrations/isolation/acq-kits/opencode/docs/capability-evidence/2026-11-04-opencode-2.0.30.mdwith the results JSON/TSV and replies next to it.inputModalities ["text","image"]for 12 models. 9 candidates misread the image and are not published; 3 were unavailable (two Gemini models returned 403model_blocked_globally, one Llama model returned 400 for no streaming tool use).reasoningstays absent for all models (ADR 0006 amended in docs(opencode): add reasoning capability validation procedure #577; follow-up stays in Verify and publish reasoning capability metadata for USAi models #574).