Motivation
Users want a way to ask OpenCode "did this actually get done?" against a tracker, checklist, or prior OpenCode output — a verification pass, not a fresh review. Today you'd have to hand-craft a /opencode:rescue prompt for every verification.
Proposed UX
/opencode:verify --tracker TODO.md
/opencode:verify --plan path/to/plan.md
/opencode:verify --last # verify the diffs produced by the last task run
Modes:
--tracker <file>: read the tracker file, compare against the current tree, report which items are actually done.
--plan <file>: check that the plan's listed changes exist in the tree / git history.
--last: use the last task job's changedFiles + result summary and ask OpenCode to verify the claimed changes were actually made and work.
Implementation sketch
- New command
plugins/opencode/commands/verify.md.
- New handler
handleVerify that constructs context from the tracker/plan/prior-job data and runs the standard review flow under the plan agent.
- Reuse
jobDataPath to load the previous job's result when --last is used.
Upstream reference
Derived from openai/codex-plugin-cc#10.
Motivation
Users want a way to ask OpenCode "did this actually get done?" against a tracker, checklist, or prior OpenCode output — a verification pass, not a fresh review. Today you'd have to hand-craft a
/opencode:rescueprompt for every verification.Proposed UX
Modes:
--tracker <file>: read the tracker file, compare against the current tree, report which items are actually done.--plan <file>: check that the plan's listed changes exist in the tree / git history.--last: use the lasttaskjob'schangedFiles+ result summary and ask OpenCode to verify the claimed changes were actually made and work.Implementation sketch
plugins/opencode/commands/verify.md.handleVerifythat constructs context from the tracker/plan/prior-job data and runs the standard review flow under theplanagent.jobDataPathto load the previous job's result when--lastis used.Upstream reference
Derived from openai/codex-plugin-cc#10.