ARTEX/.github/scripts/bot.sh

#!/usr/bin/env bash
# Lightweight triage and reply bot for issues and pull requests.
# Runs inside GitHub Actions; no server involved.
#
# Triggers (see bot.yml):
#   - issue opened: exactly one triage comment per issue
#   - comment containing @artex-bot: answers that comment (works in PR
#     threads too, since issue_comment fires there)
#   - pull request opened: one first-look review comment
# The bot never posts an approving review and never replies to itself,
# to the maintainer, or to other bots (enforced in bot.yml).
set -uo pipefail

MAX_CHARS=8000
FOOTER='---
🤖 automated reply from the repo bot. A human maintainer reads everything. Ping me in any comment with `@artex-bot`.'

log() { echo "[bot] $*"; }

[ -z "${DEEPSEEK_API_KEY:-}" ] && { log "DEEPSEEK_API_KEY secret is not set; skipping."; exit 0; }

ev=$(cat "$GITHUB_EVENT_PATH")
event_name="$GITHUB_EVENT_NAME"
repo="$GITHUB_REPOSITORY"

ask_model() { # $1 = system prompt, $2 = user payload; prints the answer
  local body="${2:0:$MAX_CHARS}"
  local req
  req=$(jq -n --arg sys "$1" --arg body "$body" \
    '{model: "deepseek-chat",
      messages: [{role: "system", content: $sys}, {role: "user", content: $body}],
      max_tokens: 900, temperature: 0.2}')
  curl -sS --max-time 120 https://api.deepseek.com/chat/completions \
    -H "Authorization: Bearer $DEEPSEEK_API_KEY" \
    -H 'Content-Type: application/json' \
    -d "$req" | jq -r '.choices[0].message.content // empty'
}

repo_context() { git ls-files | head -120; }

post_comment() { # $1 = issue or PR number, $2 = markdown text
  { echo "$2"; echo; echo "$FOOTER"; } > /tmp/bot-comment.md
  if gh issue view "$1" --json pull_request --jq '.pull_request' | grep -q .; then
    gh pr comment "$1" --body-file /tmp/bot-comment.md
  else
    gh issue comment "$1" --body-file /tmp/bot-comment.md
  fi
}

eyes() { # $1 = issue or PR number
  gh api -X POST "repos/$repo/issues/$1/reactions" -f content=eyes >/dev/null 2>&1 || true
}

# The model ends its triage reply with a "LABELS: a, b" line. Extract the
# whitelisted labels, strip the line from the comment, and apply the labels.
LABEL_COLORS=(bug=d73a4a question=d876e3 feature=a2eeef P1=b60205 P2=fbca04 P3=c5def5)

ensure_label() { # $1 = label name
  local entry color
  for entry in "${LABEL_COLORS[@]}"; do
    [ "${entry%%=*}" = "$1" ] && color="${entry#*=}"
  done
  gh label create "$1" --color "${color:-cccccc}" --force >/dev/null 2>&1 || true
}

apply_labels() { # $1 = issue number, $2 = comma separated label names from the model
  local wanted=() item ok w
  IFS=',' read -ra parts <<<"$2"
  for item in "${parts[@]}"; do
    item=$(echo "$item" | tr -d '[:space:]' | tr '[:upper:]' '[:lower:]')
    ok=0
    for w in bug question feature p1 p2 p3; do [ "$item" = "$w" ] && ok=1; done
    [ "$ok" = 1 ] || continue
    case "$item" in p1|p2|p3) item=${item^^} ;; esac
    wanted+=("$item")
  done
  if [ "${#wanted[@]}" -gt 0 ]; then
    for item in "${wanted[@]}"; do ensure_label "$item"; done
    gh issue edit "$1" --add-label "$(IFS=,; echo "${wanted[*]}")" >/dev/null 2>&1 || log "label apply failed"
    log "applied labels: ${wanted[*]}"
  fi
}

split_labels() { # stdin: model reply; stdout: reply without the LABELS line; the label names land in /tmp/bot-labels-line (survives the command-substitution subshell)
  awk '
    /^LABELS:[ \t]*/ { sub(/^[^:]*:[ \t]*/, ""); print > "/dev/stderr"; next }
    { print }
  ' 2>/tmp/bot-labels-line
}

already_triaged() { # $1 = issue number
  gh api "repos/$repo/issues/$1/comments" --jq '.[].body' 2>/dev/null \
    | grep -q 'automated reply from the repo bot'
}

answer=""

case "$event_name" in
  issues)
    number=$(jq -r '.issue.number' <<<"$ev")
    title=$(jq -r '.issue.title' <<<"$ev")
    body=$(jq -r '.issue.body // ""' <<<"$ev")
    maintainer_opened=false
    [ "$GITHUB_ACTOR" = "BaaRai" ] && maintainer_opened=true
    if [ "$maintainer_opened" = false ] && already_triaged "$number"; then log "issue #$number already triaged"; exit 0; fi
    [ "$maintainer_opened" = false ] && eyes "$number"
    answer=$(ask_model \
      'You are the triage bot of an open-source Go + Next.js repository (an autonomous penetration testing agent system; AGPL-3.0). Classify this report as bug, question, or feature request. If it is a question, answer it concisely using the repository file list as context. If it is a bug report, say what reproduction steps or version information are missing. If it is a feature request, restate it in one sentence. Maximum 180 words, plain markdown, reply in the language of the report, no heading. End your reply with one final line of the exact form "LABELS: x, y" choosing only from: bug, question, feature, and for bugs a priority: P1 (security impact or data loss), P2 (core functionality broken), P3 (minor or cosmetic); omit the priority for questions and features; use "LABELS: none" if nothing fits.' \
      "Issue title: $title

Issue body:
$body

Repository file list:
$(repo_context)")
    [ -z "$answer" ] && { log "model returned nothing"; exit 0; }
    clean_text=$(split_labels <<<"$answer")
    reply_labels=$(tail -1 /tmp/bot-labels-line 2>/dev/null)
    [ -z "$reply_labels" ] && reply_labels="none"
    if [ "$maintainer_opened" = true ]; then
      log "maintainer issue #$number: labels only, staying silent"
      apply_labels "$number" "$reply_labels"
      exit 0
    fi
    post_comment "$number" "$clean_text"
    apply_labels "$number" "$reply_labels"
    ;;

  issue_comment)
    number=$(jq -r '.issue.number' <<<"$ev")
    body=$(jq -r '.comment.body' <<<"$ev")
    question="${body#*@artex-bot}"            # strip up to the mention
    echo "$body" | grep -qi '@artex-bot' || { log "no mention; skipping"; exit 0; }
    eyes "$number"
    answer=$(ask_model \
      'You are the helper bot of an open-source Go + Next.js repository (an autonomous penetration testing agent system; AGPL-3.0). Someone pinged you with a question about this repository. Answer it concisely and only from what the repository file list suggests; say plainly when you are not sure and suggest opening a proper issue. Maximum 180 words, plain markdown, reply in the language of the question.' \
      "Question:
$question

Repository file list:
$(repo_context)")
    [ -z "$answer" ] && { log "model returned nothing"; exit 0; }
    post_comment "$number" "$answer"
    ;;

  pull_request)
    number=$(jq -r '.pull_request.number' <<<"$ev")
    title=$(jq -r '.pull_request.title' <<<"$ev")
    body=$(jq -r '.pull_request.body // ""' <<<"$ev")
    eyes "$number"
    diff=$(gh pr diff "$number" 2>/dev/null | head -c 4000)
    answer=$(ask_model \
      'You are the first-look bot of an open-source Go + Next.js repository (an autonomous penetration testing agent system; AGPL-3.0). Summarize what this pull request changes in two or three sentences, then a short checklist: does it include or update tests, does it touch documentation where needed, does anything look inconsistent with the described goal. You are commenting for information only; you do not approve or request changes. Maximum 180 words, plain markdown.' \
      "Pull request title: $title

Pull request description:
$body

Diff (truncated):
$diff")
    [ -z "$answer" ] && { log "model returned nothing"; exit 0; }
    post_comment "$number" "$answer"
    ;;

  *)
    log "unhandled event $event_name; skipping"
    ;;
esac

log "done"