Mark adressed AI review comments as resolved

.github/seidroid/ai-review/README.md

@@ -5,7 +5,7 @@ use. Two workflows live in `.github/workflows/`:
 
 | Workflow | Trigger (in the caller) | What it does |
 |----------|-------------------------|--------------|
-| `ai-review.yml` | `pull_request` and PR comment events | Three-pass review (OpenAI Codex ∥ Cursor → Claude synthesizes), posting **one** PR review + an `AI Review` check run. It reviews automatically once; an active allowed-team member can request another review with an exact `@seidroid review` comment. |
+| `ai-review.yml` | `pull_request` and PR comment events | Three-pass review (OpenAI Codex ∥ Cursor → Claude synthesizes), posting **one** PR review + an `AI Review` check run. By default it reviews automatically once; callers can enable re-review on every push, and an active allowed-team member can request another review with an exact `@seidroid review` comment. Re-reviews resolve previous seidroid inline threads whose findings were addressed or superseded by a new inline comment. |
 | `ai-assistant.yml` | `issue_comment`, `pull_request_review_comment`, `pull_request_review` | Conversational responder: mention `@seidroid` on a PR and the bot answers in-thread. |
 
 ## Base prompts (edit these)
@@ -56,6 +56,7 @@ jobs:
       # allowed-team: my-org/my-team   # default: sei-protocol/sei-core
       # extra-instructions: "Flag added allocations in the hot path."
       # prebuild-script: "go mod download"   # warm Codex's offline sandbox
+      # re-review-on-push: true         # review again after every PR push
 ```
 
 | Input | Default | Notes |
@@ -71,6 +72,7 @@ jobs:
 | `runs-on` | `ubuntu-latest` | Runner label. |
 | `claude-model` | `''` | Optional Claude model override. |
 | `approve-on-success` | `true` | If true, APPROVE on a clean verdict; else COMMENT. |
+| `re-review-on-push` | `false` | Re-run the review on every `pull_request.synchronize` event, even after seidroid has already reviewed the PR. |
 | `timeout-minutes` | `15` | Per-job timeout. |
 
 ## Using the assistant workflow

.github/seidroid/ai-review/review.md

@@ -23,6 +23,17 @@ seidroid reviews and replies to their inline comments. On a re-review:
 - report an earlier finding again if it is still present, briefly noting why the reply or
   subsequent change did not resolve it.
 
+Each previous inline thread may include a `thread_id` and its resolved/unresolved state.
+For every unresolved thread whose finding the current changes fully address, add that exact
+ID to `resolved_thread_ids`.
+
+If an unresolved finding is still present, report it again as a current `inline_comments`
+entry and put its old thread ID in that entry's `supersedes_thread_ids`. This replaces the
+stale thread with the new comment instead of leaving two unresolved copies. Do not put
+already-resolved threads or threads without an ID in either field. The workflow validates
+all IDs and resolves old threads only after posting the new review (and, for superseded
+threads, only when the replacement was successfully posted inline).
+
 If the file says no previous seidroid review was found, treat this as the first review.
 
 ## STEP 2 — Read the PR changes (review ONLY what the PR changes)
@@ -59,7 +70,10 @@ with a brief note. Be concise and specific.
 - Only anchor to a line that actually appears in the PR diff. If you are not confident a
   finding maps to a changed line, do NOT force it — put it in bucket B instead.
 - `severity`: `"blocker"`, `"suggestion"`, or `"nit"`.
-- `body`: concise comment text.
+- `body`: concise comment text. Do not include a severity prefix such as `[blocker]`; the
+  workflow adds exactly one prefix from `severity`.
+- `supersedes_thread_ids`: exact IDs of unresolved previous threads that this new inline
+  comment replaces. Use `[]` for a new finding.
 
 **B) NOT tied to a single line** (cross-cutting, missing tests, design, general
 observations) → `blockers` (must-fix) or `non_blockers` (suggestions/nits). Each entry is
@@ -73,7 +87,8 @@ one short bullet.
 - `"success"` → clean; nothing of note, safe to merge.
 
 Write `summary`: a one- or two-sentence overall summary. Use empty arrays (`[]`) for any
-bucket with no findings.
+bucket with no findings, including `resolved_thread_ids` when no previous inline finding
+was addressed.
 
 ## Untrusted content