fix(test-classifier): resolve repo from origin (not gh's parent) + harden comment-id capture (#64)

On a FORK, `gh repo view` resolves to the PARENT repo (a navapbc/ai-chatbot
checkout reports vercel/chatbot), so the dispatcher looked up the wrong PR, ran
the suite against the wrong diff, posted to the wrong repo (review-comment 422
'path could not be resolved' → issue-comment fallback on the upstream), and
recorded a Testing Events row attributed to the wrong slug.

Fix: resolve the owner/name slug from the `origin` remote URL (HTTPS or SSH),
cached, with an AI_REVIEW_REPO=owner/name override. Thread it through all four
gh call sites (--pr lookup, auto-discovery, comment posting, metrics row) via
`-R <slug>` so they all agree on the fork.

Also harden the comment-id capture: on a 422, `gh ... --jq` can print the raw
error JSON to stdout, which slipped past the old non-empty guard and got stored
as comment_id — poisoning the metrics row (the 'Validation Failed' text seen in
the Sheet). Add ai_review::valid_comment_capture requiring a NUMERIC id; reject
otherwise. Both this and the repo fix independently prevent the corruption.

Docs: add a prominent WARNING that AI_RUN_SUITE=1 (OBSERVED) runs the change's
code on your machine with no sandbox — only use it on branches you trust, never
an untrusted --pr; plus a troubleshooting row for the fork/AI_REVIEW_REPO case.

Author: fg-nava

testing/classifier/.skills/test-classifier/scripts/test-classifier-dispatcher.sh

@@ -216,6 +216,42 @@ require_gh_cli() {
   fi
 }
 
+# Resolve the owner/name slug of the repo the PR lives in, from the `origin`
+# remote — NOT `gh repo view`. On a fork, `gh repo view` resolves to the PARENT
+# repo (e.g. a navapbc/ai-chatbot checkout reports vercel/chatbot), so posting
+# and the metrics row would target the wrong repo. `origin` is the fork you are
+# actually working in and whose PR number you passed. Override with AI_REVIEW_REPO
+# (owner/name) for the rare cross-remote case. Cached after first resolution.
+AI_REVIEW_REPO_SLUG=""
+ai_review::repo_slug() {
+  if [[ -n "${AI_REVIEW_REPO_SLUG}" ]]; then
+    printf '%s' "${AI_REVIEW_REPO_SLUG}"; return 0
+  fi
+  if [[ -n "${AI_REVIEW_REPO:-}" ]]; then
+    AI_REVIEW_REPO_SLUG="${AI_REVIEW_REPO}"
+    printf '%s' "${AI_REVIEW_REPO_SLUG}"; return 0
+  fi
+  local url
+  url="$(git remote get-url origin 2>/dev/null || true)"
+  if [[ -z "${url}" ]]; then
+    echo "ERROR: no 'origin' remote — cannot determine which repo the PR lives in." >&2
+    echo "       Set one (git remote add origin <url>) or export AI_REVIEW_REPO=owner/name." >&2
+    return 1
+  fi
+  # Normalize both forms to owner/name:
+  #   https://github.com/owner/name(.git)   git@github.com:owner/name(.git)
+  local slug="${url}"
+  slug="${slug#*github.com[:/]}"   # strip scheme/host up to owner
+  slug="${slug%.git}"              # strip trailing .git
+  if [[ ! "${slug}" =~ ^[^/]+/[^/]+$ ]]; then
+    echo "ERROR: could not parse owner/name from origin URL: ${url}" >&2
+    echo "       Export AI_REVIEW_REPO=owner/name to set it explicitly." >&2
+    return 1
+  fi
+  AI_REVIEW_REPO_SLUG="${slug}"
+  printf '%s' "${AI_REVIEW_REPO_SLUG}"
+}
+
 ai_review::discover_pr_context() {
   # If --against was passed, it wins for the diff range — no base-ref lookup
   # needed. But we still record --pr (when given) as the PR number so comment
@@ -241,8 +277,10 @@ ai_review::discover_pr_context() {
   # If --pr was given, look up that PR's base ref.
   if [[ -n "${PR_NUMBER}" ]]; then
     require_gh_cli "PR number was specified via --pr"
+    local repo_slug
+    repo_slug="$(ai_review::repo_slug)" || exit 1
     local base
-    base="$(gh pr view "${PR_NUMBER}" --json baseRefName --jq '.baseRefName' 2>/dev/null || true)"
+    base="$(gh pr view "${PR_NUMBER}" -R "${repo_slug}" --json baseRefName --jq '.baseRefName' 2>/dev/null || true)"
     if [[ -z "${base}" ]]; then
       echo "ERROR: could not look up PR #${PR_NUMBER} via gh CLI." >&2
       echo "       Verify the PR number exists and you have access to it." >&2
@@ -266,9 +304,12 @@ ai_review::discover_pr_context() {
   fi
 
   # Let gh extract the fields with --jq (same idiom as the --pr branch above),
-  # rather than scraping the raw JSON with brittle regexes.
-  AI_REVIEW_PR_NUMBER="$(gh pr view --json number --jq '.number' 2>/dev/null || true)"
-  AI_REVIEW_PR_BASE="$(gh pr view --json baseRefName --jq '.baseRefName' 2>/dev/null || true)"
+  # rather than scraping the raw JSON with brittle regexes. Pin to origin's slug
+  # so a fork checkout discovers its OWN PR, not the parent repo's.
+  local repo_slug
+  repo_slug="$(ai_review::repo_slug)" || exit 1
+  AI_REVIEW_PR_NUMBER="$(gh pr view -R "${repo_slug}" --json number --jq '.number' 2>/dev/null || true)"
+  AI_REVIEW_PR_BASE="$(gh pr view -R "${repo_slug}" --json baseRefName --jq '.baseRefName' 2>/dev/null || true)"
 
   if [[ -z "${AI_REVIEW_PR_NUMBER}" ]] || [[ -z "${AI_REVIEW_PR_BASE}" ]]; then
     echo "ERROR: 'gh pr view' could not find an open PR for the current branch." >&2
@@ -585,7 +626,7 @@ post_issue_comment_to_github() {
         "repos/${repo_slug}/issues/${pr_number}/comments" \
         --method POST \
         --field body="${body}" --jq '"\(.id)\t\(.created_at)"' 2>/dev/null || true)"
-  if [[ -z "$resp" || "$resp" == $'\t' ]]; then
+  if ! ai_review::valid_comment_capture "$resp"; then
     echo "ERROR: 'gh api' issue-comment call failed." >&2
     echo "       Check your gh auth status and that your token has 'pull-requests: write'" >&2
     echo "       (or 'issues: write')." >&2
@@ -595,6 +636,17 @@ post_issue_comment_to_github() {
   POSTED_COMMENT_CREATED="${resp#*$'\t'}"
 }
 
+# Validate an "<id>\t<created_at>" capture from a comment-POST response. A real
+# success has a NUMERIC id. On an API error (e.g. a 422), `gh ... --jq` may print
+# the raw error JSON to stdout, which would otherwise be stored verbatim as the
+# comment_id and poison the metrics row — so require a numeric leading field.
+ai_review::valid_comment_capture() {
+  local resp="$1"
+  [[ -n "$resp" && "$resp" != $'\t' ]] || return 1
+  local id="${resp%%$'\t'*}"
+  [[ "$id" =~ ^[0-9]+$ ]]
+}
+
 # Post ONE classification comment to the PR. Args: PR number, comment body.
 #
 # Posts as a file-level pull-request REVIEW comment (subject_type=file) so the
@@ -611,10 +663,7 @@ post_comment_to_github() {
   require_gh_cli "--post-comment was specified"
 
   local repo_slug
-  if ! repo_slug="$(gh repo view --json nameWithOwner --jq '.nameWithOwner' 2>/dev/null)"; then
-    echo "ERROR: could not determine repo from gh CLI." >&2
-    exit 1
-  fi
+  repo_slug="$(ai_review::repo_slug)" || exit 1
 
   # Anchor path: a file in the PR diff. Prefer a changed test file if one is
   # present, since the comment is about test failures, but any changed file
@@ -651,7 +700,7 @@ post_comment_to_github() {
           --field commit_id="${commit_id}" \
           --field path="${anchor_path}" \
           --field subject_type=file --jq '"\(.id)\t\(.created_at)"' 2>/dev/null || true)"
-    if [[ -n "$resp" && "$resp" != $'\t' ]]; then
+    if ai_review::valid_comment_capture "$resp"; then
       POSTED_COMMENT_ID="${resp%%$'\t'*}"
       POSTED_COMMENT_CREATED="${resp#*$'\t'}"
       echo "[test-classifier] Review comment posted. Awaiting the developer's 👍/👎 reaction (and a reply reason on a 👎)."
@@ -733,7 +782,7 @@ submit_metrics_row() {
   fi
 
   local repo_slug
-  repo_slug="$(gh repo view --json nameWithOwner --jq '.nameWithOwner' 2>/dev/null || echo "")"
+  repo_slug="$(ai_review::repo_slug 2>/dev/null || echo "")"
 
   # The classifier may emit multiple classifications; the Testing Events row is
   # per-comment, so collapse to one representative verdict by most-actionable

testing/classifier/docs/LOCAL_TEST_CLASSIFIER.md

@@ -32,6 +32,21 @@ and uses the same `--unpushed` scope rule. This is the ergonomics layer over
   OBSERVED with `AI_RUN_SUITE=1` to have the agent actually locate, install, and
   run your suite and triage the real failures — the same behavior CI uses.
 
+> [!WARNING]
+> **OBSERVED (`AI_RUN_SUITE=1`) runs the change's code on your machine — there
+> is no sandbox.** The agent installs dependencies (running each package's
+> install/postinstall scripts) and executes the test suite **directly in your
+> checkout**, with your shell, your environment, and your credentials. It also
+> leaves build artifacts behind (`node_modules/`, downloaded browsers, etc.).
+>
+> Only use `AI_RUN_SUITE=1` on a branch **you trust** — typically your own
+> work-in-progress. Do **not** point it at an untrusted or unreviewed PR
+> (e.g. `--pr N` for someone else's contribution): installing and running that
+> branch's code is arbitrary code execution on your laptop. For untrusted
+> changes, use the default **INFERRED** mode (omit `AI_RUN_SUITE`), which only
+> reasons over the diff and never executes anything, or let CI run OBSERVED in
+> its disposable runner.
+
 ---
 
 ## Prerequisites
@@ -214,6 +229,7 @@ run (Path B) is still the recorded, metrics-feeding pass; this is your preview.
 | `AI_REVIEW_TOOL … not set` | Export `AI_REVIEW_TOOL=claude` (or `codex`/`copilot`); see `README.md`. |
 | Classification seems to miss recent edits | `--unpushed` excludes *unstaged* changes. `git add` or commit them first. |
 | Everything lands in INFERRED | That's the default. Add `AI_RUN_SUITE=1` to run the suite (OBSERVED). Suites needing services may still fall back to INFERRED, by design. |
+| Posted to / recorded the wrong repo (e.g. the upstream of your fork) | The repo is resolved from your `origin` remote. On a fork whose `origin` isn't the repo the PR lives in, export `AI_REVIEW_REPO=owner/name` to set it explicitly. |
 
 ---