Merge pull request 'feat(automation): watcher that addresses claude-review: comments on Claude PRs' (#129) from claude/busy-boyd-b16092 into master

Reviewed-on: #129

.gitignore

@@ -32,6 +32,10 @@ automation/watcher.config.json
 automation/logs/
 automation/.watcher.lock
 
+# Claude PR-review-comment watcher (real token + lock stay local; shares logs/)
+automation/pr-review-watcher.config.json
+automation/.pr-review-watcher.lock
+
 # OS
 .DS_Store
 Thumbs.db

automation/README.md

@@ -70,6 +70,7 @@ A [`PULL_REQUEST_TEMPLATE.md`](../.forgejo/PULL_REQUEST_TEMPLATE.md) carries the
 | Forgejo helper | `App/lib/forgejo.ts` | Needs `FORGEJO_URL`, `FORGEJO_REPO`, `FORGEJO_TOKEN` env (token scope: `write:issue`) |
 | Issue watcher (active) | `automation/claude-issue-watcher.sh` on pms1 | Bash port; runs 24/7 via cron. Config + logs under `~/issue-watcher/` |
 | Issue watcher (Windows, disabled) | `automation/claude-issue-watcher.ps1` | PowerShell original. `PelagiaClaudeIssueWatcher` task is **disabled** (pms1 is the sole worker; two pollers would race) |
+| PR review-comment watcher | `automation/claude-pr-review-watcher.sh` on pms1 | Addresses `claude-review:` comments on Claude-raised PRs. Own cron entry, own clone (`~/pelagia-pr-review`), own config + lock. See below |
 | Forgejo helper | `App/lib/forgejo.ts` | Needs `FORGEJO_URL`, `FORGEJO_REPO`, `FORGEJO_TOKEN` env (token scope: `write:issue`) |
 | Deploy workflow | `.forgejo/workflows/deploy.yml` | Triggers on `v*` tags; runs on the `host` runner |
 | Runner | pms1 `~/forgejo-runner`, pm2 process `forgejo-runner` | Registered as `pms1-host` with labels `host`, `docker` |
@@ -93,6 +94,72 @@ activates automatically once signed in. (An `ANTHROPIC_API_KEY` env var also sat
 The Windows variant (`.ps1` + `register-watcher-task.ps1`) is the portable fallback;
 re-enable its task only if pms1 is unavailable, and disable one before enabling the other.
 
+## PR review-comment watcher
+
+Where the issue watcher turns *issues* into PRs, the **PR review-comment watcher**
+([`automation/claude-pr-review-watcher.sh`](claude-pr-review-watcher.sh)) closes the
+loop on the other side: it addresses **review comments left on the PRs Claude already
+raised**. This is how you iterate on an automated PR without dropping into an
+interactive session — leave a comment, Claude pushes a follow-up commit.
+
+**How to use it (as a reviewer):** on any open Claude-raised PR, leave a comment that
+starts with the marker **`claude-review:`** — the text after the marker is the
+instruction. It works in three places:
+
+- the **PR conversation** (a normal PR comment),
+- a **review summary** (the overall body of a submitted review),
+- an **inline / on-file comment** (Claude is given the file, line, and diff hunk).
+
+Example inline comment on `App/lib/foo.ts`:
+
+> `claude-review:` this should null-check `order.vendor` before dereferencing it, and add a test for the null case.
+
+**What the watcher does each run (every 10 min via cron):**
+
+1. Lists open PRs Claude raised — head branch starts with `prBranchPrefix` (`claude/`)
+   or the PR is labelled `claude-pr`.
+2. Collects every `claude-review:` comment **from repo collaborators only** (write
+   access; the repo owner is always included). Comments from anyone else, and the
+   bot's own comments, are ignored. This is the safety gate — only trusted users can
+   make Claude push code.
+3. Skips comments already handled in a previous run (tracked by a hidden
+   `<!-- ppms-review-bot handled: … -->` marker the bot stamps on its acknowledgements,
+   so a 10-minute poll never redoes the same comment).
+4. Checks out the **PR's own branch** in `~/pelagia-pr-review`, runs headless Claude
+   Code with the collected instructions (+ the same `pelagia_test` / port-3100 test
+   environment the fixer uses), then pushes the new commit(s) to **the same branch** —
+   updating the open PR in place.
+5. Acknowledges: posts a reply listing what it addressed (with the handled marker) and
+   adds a 🚀 reaction to each handled PR-conversation comment.
+
+If Claude judges a comment unclear, out of scope, or too risky to do unattended
+(migrations, payments, permissions), it makes no commit for it and the watcher posts a
+"produced no change — a human may need to take these" reply. The comments are still
+marked handled so the poll doesn't loop on them; re-comment with a clearer
+`claude-review:` instruction to retry.
+
+**Deploy on pms1** (mirrors the issue watcher):
+
+```sh
+# 1. Place the script + config alongside the issue watcher
+cp automation/claude-pr-review-watcher.sh        ~/pr-review-watcher/
+cp automation/pr-review-watcher.config.example.json ~/pr-review-watcher/pr-review-watcher.config.json
+# 2. Edit the config: real token (scope write:repository,write:issue), claudeExe = `which claude`
+# 3. Add a crontab entry, OFFSET from the issue watcher so the two don't run at the same minute:
+#    5,15,25,35,45,55 * * * * PATH=<nvm bin>:$PATH ~/pr-review-watcher/claude-pr-review-watcher.sh >> ~/pr-review-watcher/logs/cron.log 2>&1
+```
+
+- **Token scope:** needs `write:repository` (push to the PR branch) **plus**
+  `write:issue` (post comments + reactions) — one scope more than the issue watcher.
+- **Own everything:** separate clone (`~/pelagia-pr-review`), config
+  (`pr-review-watcher.config.json`), and lock (`.pr-review-watcher.lock`) so it never
+  races the issue watcher. Logs land in the same `logs/` dir
+  (`pr-review-<date>.log`, per-PR `claude-pr-<n>-*.log`).
+- Same **auth preflight** as the issue watcher — no-ops until Claude Code is signed in
+  on pms1 (or `ANTHROPIC_API_KEY` is set).
+- A Windows `.ps1` port is not provided yet (pms1 is the sole worker); port it from
+  `claude-issue-watcher.ps1` only if you need a failover.
+
 ## Test database (for autofix verification)
 
 So the fix stage can verify against realistic data without touching production:

automation/claude-pr-review-watcher.sh

@@ -0,0 +1,277 @@
+#!/usr/bin/env bash
+# Claude PR-review-comment watcher -- Linux port (runs on pms1 via cron).
+#
+# Sibling to claude-issue-watcher.sh. Where that watcher turns *issues* into PRs,
+# this one addresses *review comments* left on the PRs Claude already raised.
+#
+# Per run:
+#   1. List open PRs that Claude raised (head branch starts with prBranchPrefix,
+#      or labelled `claude-pr`).
+#   2. On each, collect every comment carrying the marker `claude-review:` --
+#      from the PR conversation, from review summaries, and from inline (on-file)
+#      review comments -- but ONLY from repo collaborators (write access).
+#   3. Skip comments already handled in a previous run (tracked by a hidden marker
+#      in the bot's acknowledgement comments).
+#   4. Run headless Claude Code on the PR's own branch with those instructions;
+#      it edits + verifies, the watcher pushes the new commit(s) to the SAME branch
+#      (updating the PR in place), then acknowledges each comment (reply + reaction).
+#
+# Config: pr-review-watcher.config.json next to this script (or pass a path as $1).
+# See automation/README.md > "PR review-comment watcher".
+
+set -uo pipefail
+
+SCRIPT_DIR="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)"
+CONFIG="${1:-$SCRIPT_DIR/pr-review-watcher.config.json}"
+[ -f "$CONFIG" ] || { echo "Config not found: $CONFIG (copy pr-review-watcher.config.example.json and fill in the token)"; exit 1; }
+
+cfg() { jq -r "$1" "$CONFIG"; }
+FORGEJO_URL=$(cfg .forgejoUrl)
+REPO=$(cfg .repo)
+TOKEN=$(cfg .token)
+WORKDIR=$(cfg .workDir)
+BASE_BRANCH=$(cfg .baseBranch)
+PR_BRANCH_PREFIX=$(cfg '.prBranchPrefix // "claude/"')
+MARKER=$(cfg '.marker // "claude-review:"')
+MAX_PRS=$(cfg '.maxPrsPerRun // 1')
+MAX_COMMENTS=$(cfg '.maxCommentsPerPr // 20')
+CLAUDE=$(cfg .claudeExe)
+TURNS=$(cfg '.claudeMaxTurns // 150')
+API="$FORGEJO_URL/api/v1"
+
+# Hidden marker the bot stamps on its acknowledgement comments. The "handled:"
+# line lists every comment key it has addressed, so subsequent runs skip them.
+HANDLED_TAG='ppms-review-bot handled:'
+ACK_REACTION='rocket'
+
+LOG_DIR="$SCRIPT_DIR/logs"
+mkdir -p "$LOG_DIR"
+LOG_FILE="$LOG_DIR/pr-review-$(date +%F).log"
+log() { echo "$(date +%T) $*" | tee -a "$LOG_FILE"; }
+
+# --- single-instance lock (separate from the issue watcher's) ---
+exec 9>"$SCRIPT_DIR/.pr-review-watcher.lock"
+if ! flock -n 9; then log "Another PR-review watcher run is active; exiting."; exit 0; fi
+
+# --- preflight: idle until Claude Code is authenticated on this host ---
+if [ ! -f "$HOME/.claude/.credentials.json" ] && [ -z "${ANTHROPIC_API_KEY:-}" ]; then
+    log "Claude Code not authenticated yet (no ~/.claude/.credentials.json or ANTHROPIC_API_KEY); skipping."
+    exit 0
+fi
+
+# --- Forgejo API helpers (curl + jq) ---
+api() { # METHOD PATH [JSON_BODY]
+    local method=$1 path=$2 body=${3:-}
+    if [ -n "$body" ]; then
+        curl -fsS -X "$method" "$API$path" -H "Authorization: token $TOKEN" \
+            -H "Content-Type: application/json" --data "$body"
+    else
+        curl -fsS -X "$method" "$API$path" -H "Authorization: token $TOKEN"
+    fi
+}
+# Soft variant: never aborts the run on a single failed call (e.g. reactions
+# unsupported on a given comment type). Returns empty + logs instead.
+api_soft() { api "$@" 2>/dev/null || { log "api_soft: $1 $2 failed (ignored)"; printf ''; }; }
+
+add_pr_comment() { # NUMBER TEXT
+    api POST "/repos/$REPO/issues/$1/comments" "$(jq -nc --arg b "$2" '{body:$b}')" >/dev/null
+}
+react() { # COMMENT_ID  (PR-conversation comments only; best-effort)
+    api_soft POST "/repos/$REPO/issues/comments/$1/reactions" \
+        "$(jq -nc --arg c "$ACK_REACTION" '{content:$c}')" >/dev/null
+}
+
+# --- prepare the dedicated work clone ---
+host_no_scheme=$(printf '%s' "$FORGEJO_URL" | sed 's#^https\?://##')
+owner=${REPO%%/*}
+CLONE_URL="http://${owner}:${TOKEN}@${host_no_scheme}/${REPO}.git"
+[ "${FORGEJO_URL#https}" != "$FORGEJO_URL" ] && CLONE_URL="https://${owner}:${TOKEN}@${host_no_scheme}/${REPO}.git"
+
+if [ ! -d "$WORKDIR/.git" ]; then
+    log "Cloning $REPO into $WORKDIR"
+    if ! git clone -q "$CLONE_URL" "$WORKDIR"; then log "git clone failed"; exit 1; fi
+    git -C "$WORKDIR" config user.name "Claude (review-bot)"
+    git -C "$WORKDIR" config user.email "claude-autofix@pelagiamarine.com"
+fi
+
+# --- identity + authorization set ---
+# The bot's own login (so its acknowledgements are never treated as instructions).
+BOT_LOGIN=$(api GET "/user" | jq -r '.login // ""')
+# Collaborators = users with write access. The repo owner is always allowed.
+COLLAB=$(api GET "/repos/$REPO/collaborators?limit=100" \
+    | jq -c --arg owner "$owner" '[.[].login] + [$owner] | unique')
+log "Authorized commenters: $(printf '%s' "$COLLAB" | jq -r 'join(", ")') (bot=$BOT_LOGIN)"
+
+# --- find Claude-raised open PRs (head branch under the prefix, or labelled claude-pr) ---
+prs=$(api GET "/repos/$REPO/pulls?state=open&limit=50" \
+    | jq -c --arg pfx "$PR_BRANCH_PREFIX" \
+        '[ .[] | select((.head.ref | startswith($pfx)) or (((.labels//[])|map(.name))|index("claude-pr"))) ] | sort_by(.number)')
+prs=$(printf '%s' "$prs" | jq -c ".[:$MAX_PRS]")
+n_prs=$(printf '%s' "$prs" | jq 'length')
+log "Found $n_prs Claude-raised open PR(s) to scan for '$MARKER' comments"
+
+# Pull the instruction text that follows the marker out of a comment body.
+instr_of() { # BODY  -> text after the first marker occurrence, trimmed
+    jq -rn --arg b "$1" --arg m "$MARKER" \
+        '$b | split($m) | .[1:] | join($m) | gsub("^\\s+|\\s+$";"")'
+}
+
+p=0
+while [ "$p" -lt "$n_prs" ]; do
+    pr=$(printf '%s' "$prs" | jq -c ".[$p]")
+    p=$((p+1))
+    num=$(printf '%s' "$pr" | jq -r .number)
+    title=$(printf '%s' "$pr" | jq -r .title)
+    branch=$(printf '%s' "$pr" | jq -r .head.ref)
+    log "-- PR #$num ($branch): $title"
+
+    # ---- gather candidate comments from the three sources ----
+    conv=$(api GET "/repos/$REPO/issues/$num/comments?limit=100")
+    reviews=$(api GET "/repos/$REPO/pulls/$num/reviews?limit=100")
+
+    # Keys already addressed in a prior run (scanned from the bot's ack comments).
+    handled=$(printf '%s' "$conv" | jq -c --arg tag "$HANDLED_TAG" \
+        '[ .[].body // "" | select(contains($tag)) | scan("(?:conv|summary|inline):[0-9]+") ] | unique')
+
+    sel='select(.body != null) | select(.body | contains($m))
+         | select(.user.login as $u | ($collab | index($u)))
+         | select(.user.login != $bot)'
+
+    conv_tasks=$(printf '%s' "$conv" | jq -c --arg m "$MARKER" --argjson collab "$COLLAB" --arg bot "$BOT_LOGIN" "
+        [ .[] | $sel | { key:(\"conv:\"+(.id|tostring)), kind:\"conv\", id:.id, user:.user.login,
+                         loc:\"PR conversation\", body:.body } ]")
+
+    summary_tasks=$(printf '%s' "$reviews" | jq -c --arg m "$MARKER" --argjson collab "$COLLAB" --arg bot "$BOT_LOGIN" "
+        [ .[] | select(.body != \"\") | $sel
+          | { key:(\"summary:\"+(.id|tostring)), kind:\"summary\", id:.id, user:.user.login,
+              loc:\"review summary\", body:.body } ]")
+
+    # Inline (on-file) comments live under each review.
+    inline_tasks='[]'
+    for rid in $(printf '%s' "$reviews" | jq -r '.[].id'); do
+        rc=$(api_soft GET "/repos/$REPO/pulls/$num/reviews/$rid/comments")
+        [ -z "$rc" ] && continue
+        t=$(printf '%s' "$rc" | jq -c --arg m "$MARKER" --argjson collab "$COLLAB" --arg bot "$BOT_LOGIN" "
+            [ .[] | $sel
+              | { key:(\"inline:\"+(.id|tostring)), kind:\"inline\", id:.id, user:.user.login,
+                  loc:(\"inline \"+(.path//\"?\")+\":\"+((.line // .original_line // 0)|tostring)),
+                  hunk:(.diff_hunk // \"\"), body:.body } ]")
+        inline_tasks=$(jq -nc --argjson a "$inline_tasks" --argjson b "$t" '$a + $b')
+    done
+
+    all=$(jq -nc --argjson a "$conv_tasks" --argjson b "$summary_tasks" --argjson c "$inline_tasks" '$a + $b + $c')
+    fresh=$(printf '%s' "$all" | jq -c --argjson h "$handled" '[ .[] | select(.key as $k | ($h|index($k)) | not) ]')
+    fresh=$(printf '%s' "$fresh" | jq -c ".[:$MAX_COMMENTS]")
+    n=$(printf '%s' "$fresh" | jq 'length')
+    if [ "$n" -eq 0 ]; then log "   no new '$MARKER' comments"; continue; fi
+    log "   $n new '$MARKER' comment(s) to address"
+
+    # ---- check out the PR branch in the work clone ----
+    git -C "$WORKDIR" fetch origin -q
+    if ! git -C "$WORKDIR" checkout -B "$branch" "origin/$branch" -q 2>>"$LOG_FILE"; then
+        log "   checkout of origin/$branch failed; skipping PR #$num"; continue
+    fi
+    git -C "$WORKDIR" clean -fdq
+
+    # ---- build the prompt ----
+    keys=$(printf '%s' "$fresh" | jq -r '[.[].key] | join(" ")')
+    prompt_file=$(mktemp)
+    {
+        printf '%s\n' "You are addressing REVIEW COMMENTS on PR #$num of the Pelagia Portal (PPMS), a Next.js 15"
+        printf '%s\n' "purchase-order management system. The web app lives in App/ -- read App/CLAUDE.md first."
+        printf '%s\n' "You are already checked out on the PR branch '$branch'. Inspect what the PR changed with:"
+        printf '%s\n' "    git -C . log --oneline origin/$BASE_BRANCH..HEAD && git diff origin/$BASE_BRANCH...HEAD"
+        printf '\n## PR #%s: %s\n\n' "$num" "$title"
+        printf '%s\n\n' "## Review comments to address (each begins with '$MARKER')"
+        i=0
+        while [ "$i" -lt "$n" ]; do
+            item=$(printf '%s' "$fresh" | jq -c ".[$i]")
+            i=$((i+1))
+            u=$(printf '%s' "$item" | jq -r .user)
+            loc=$(printf '%s' "$item" | jq -r .loc)
+            body=$(printf '%s' "$item" | jq -r .body)
+            hunk=$(printf '%s' "$item" | jq -r '.hunk // ""')
+            instr=$(instr_of "$body")
+            printf '### Comment %s -- %s (by %s)\n' "$i" "$loc" "$u"
+            if [ -n "$hunk" ] && [ "$hunk" != "null" ]; then
+                printf 'Code under review:\n```\n%s\n```\n' "$hunk"
+            fi
+            printf 'Instruction: %s\n\n' "$instr"
+        done
+        printf '%s\n' "## Test environment available to you"
+        printf '%s\n' "- App/.env points DATABASE_URL at a TEST database (pelagia_test) -- a daily mirror of"
+        printf '%s\n' "  production, safe to read and write. It is NOT production. Email is console-logged and"
+        printf '%s\n' "  storage is local in this dev mode."
+        printf '%s\n' "- Run integration tests after loading the env:"
+        printf '%s\n' "    cd App && set -a && . ./.env && set +a && pnpm test:integration"
+        printf '%s\n' "- If you need runtime verification you MAY start a dev server ON PORT 3100 ONLY:"
+        printf '%s\n' "    cd App && pnpm dev -p 3100      (production runs on 3000 -- NEVER touch 3000)"
+        printf '%s\n' "  Stop ONLY your own server by port ('fuser -k 3100/tcp'); NEVER a broad 'pkill -f next'."
+        printf '%s\n' ""
+        printf '%s\n' "## Your job (PR policy: every code change ships with tests + docs)"
+        printf '%s\n' "1. Make the focused changes the review comments ask for -- nothing more."
+        printf '%s\n' "2. If you change code under App/app|lib|components|hooks, add or update a test (the PR check"
+        printf '%s\n' "   rejects code changes with no test change). Model integration tests on"
+        printf '%s\n' "   App/tests/integration/dashboard-approved-this-month.test.ts."
+        printf '%s\n' "3. Verify: 'cd App && pnpm type-check' (no new errors); run relevant tests."
+        printf '%s\n' "4. Update any docs the change affects (App/README.md, App/CLAUDE.md, Docs/, CHANGELOG.md)."
+        printf '%s\n' "5. Commit ALL changes to the current branch with a conventional message referencing #$num."
+        printf '%s\n' "6. Do NOT push, do NOT switch branches, do NOT open/close PRs. The supervisor pushes."
+        printf '%s\n' "If a comment is unclear, out of scope, or too risky to do unattended (migrations, payments,"
+        printf '%s\n' "permissions), make NO commit for it and explain why in CLAUDE_RESULT.md in the repo root."
+    } > "$prompt_file"
+
+    clog="$LOG_DIR/claude-pr-$num-$(date +%Y%m%d-%H%M%S).log"
+    log "   Running Claude on PR #$num (log: $clog)"
+    ( cd "$WORKDIR" && "$CLAUDE" -p --dangerously-skip-permissions \
+        --max-turns "$TURNS" --output-format text < "$prompt_file" > "$clog" 2>&1 ); rc=$?
+    log "   Claude exited with code $rc for PR #$num"
+    rm -f "$prompt_file"
+
+    note=""
+    if [ -f "$WORKDIR/CLAUDE_RESULT.md" ]; then
+        note=$(cat "$WORKDIR/CLAUDE_RESULT.md")
+        rm -f "$WORKDIR/CLAUDE_RESULT.md"
+        git -C "$WORKDIR" checkout -- . 2>/dev/null
+    fi
+
+    # ---- build the acknowledgement (lists handled keys + quotes each comment) ----
+    ack_items=$(printf '%s' "$fresh" | jq -r '.[] | "- **\(.loc)** (by \(.user))"')
+
+    commits=$(git -C "$WORKDIR" rev-list "origin/$branch..HEAD" --count 2>/dev/null || echo 0)
+    if [ "${commits:-0}" -gt 0 ]; then
+        log "   Claude made $commits commit(s); pushing to $branch"
+        if ! git -C "$WORKDIR" push -u origin "$branch" -q 2>>"$LOG_FILE"; then
+            log "   push failed for PR #$num"
+            add_pr_comment "$num" "[Claude review-bot] Addressed the review comments locally but the push to \`$branch\` failed. See watcher logs on pms1: \`$clog\`.
+<!-- $HANDLED_TAG -->"
+            continue
+        fi
+        body="[Claude review-bot] Addressed the following review comment(s) on \`$branch\` ($commits commit(s) pushed):
+
+$ack_items
+${note:+
+Notes:
+$note
+}
+<!-- $HANDLED_TAG $keys -->"
+        add_pr_comment "$num" "$body"
+        # Best-effort reaction on PR-conversation comments (reactions API is keyed
+        # to issue comments; inline/summary review comments are tracked by the marker).
+        for cid in $(printf '%s' "$fresh" | jq -r '.[] | select(.kind=="conv") | .id'); do react "$cid"; done
+        log "   PR #$num updated and acknowledged"
+    else
+        log "   No commits produced for PR #$num"
+        reason=${note:-"Claude did not produce a change. See watcher logs on pms1: \`$clog\`."}
+        add_pr_comment "$num" "[Claude review-bot] Reviewed the marked comment(s) but produced no change:
+
+$reason
+
+A human may need to take these:
+$ack_items
+<!-- $HANDLED_TAG $keys -->"
+        log "   PR #$num: no change, acknowledged (marked handled to avoid re-running)"
+    fi
+done
+
+log "PR-review watcher run complete."

automation/pr-review-watcher.config.example.json

@@ -0,0 +1,13 @@
+{
+  "forgejoUrl": "https://git.pelagiamarine.com",
+  "repo": "shad0w/pelagia-portal",
+  "token": "<forgejo token with write:repository,write:issue>",
+  "workDir": "/home/shad0w/pelagia-pr-review",
+  "baseBranch": "master",
+  "prBranchPrefix": "claude/",
+  "marker": "claude-review:",
+  "maxPrsPerRun": 1,
+  "maxCommentsPerPr": 20,
+  "claudeExe": "/home/shad0w/.nvm/versions/node/<ver>/bin/claude",
+  "claudeMaxTurns": 150
+}