#!/usr/bin/env bash
# research-speaker -- Proactive web research on a newly identified meeting speaker
#
# Usage: research-speaker <room_dir> <speaker-slug> <display-name> <project-context> [--list-queries] [--approve TOKEN] [--apply]
# Example: research-speaker room/ tyler-chen "Tyler Chen" "EdTech startup building AI tutoring platform"
#
# Performs web research on a speaker in the context of the project/room.
# Outputs research summary to stdout for user confirmation.
# With --apply flag, updates the speaker's PROFILE.md with findings.
#
# Phase 386 plan 24 (owner decision 2026-11-06): nothing is sent until the user
# approves the exact searches. --list-queries prints the 3 searches and the
# approval token and sends nothing. A run sends only with --approve TOKEN, after
# a credential check, and records each search in the room audit ledger. --apply
# needs the same token. No content scan runs. The check and the record use
# lib/core/web-path-gate.cjs (needs node; without node nothing is sent).
#
# Exit 0 on success (or research unavailable -- not fatal)
# Exit 1 on failure (missing profile, invalid args)

set -euo pipefail

# --help flag
if [[ "${1:-}" == "--help" || "${1:-}" == "-h" ]]; then
  cat <<'USAGE'
research-speaker -- Proactive web research on a newly identified meeting speaker

USAGE:
  research-speaker <room_dir> <speaker-slug> <display-name> <project-context> [--list-queries] [--approve TOKEN] [--apply]

ARGUMENTS:
  room_dir         Path to the room directory (e.g., room/)
  speaker-slug     Lowercase hyphenated name (e.g., tyler-chen)
  display-name     Full display name in quotes (e.g., "Tyler Chen")
  project-context  Brief project/venture description for search context

OPTIONS:
  --list-queries   Print the exact searches and the approval token. Sends nothing.
  --approve TOKEN  Send the searches. TOKEN must be the token of the exact list.
                   Without a matching token nothing is sent.
  --apply          Apply research findings to PROFILE.md (requires prior user
                   confirmation and the same --approve TOKEN)

EXAMPLES:
  # Research and output findings for review
  research-speaker room/ tyler-chen "Tyler Chen" "EdTech startup building AI tutoring"

  # Show the exact searches first (sends nothing)
  research-speaker room/ tyler-chen "Tyler Chen" "EdTech startup building AI tutoring" --list-queries

  # Send them after the user approves
  research-speaker room/ tyler-chen "Tyler Chen" "EdTech startup building AI tutoring" --approve TOKEN

  # Apply confirmed findings to profile
  research-speaker room/ tyler-chen "Tyler Chen" "EdTech startup building AI tutoring" --approve TOKEN --apply

OUTPUT:
  Without --apply: Prints research summary markdown to stdout
  With --apply: Updates PROFILE.md and prints confirmation

EXIT CODES:
  0  Success (including "research unavailable" -- not fatal)
  1  Failure (missing profile, invalid arguments)
USAGE
  exit 0
fi

# Validate arguments
if [[ $# -lt 4 ]]; then
  echo "ERROR: At least 4 arguments required: room_dir, speaker-slug, display-name, project-context" >&2
  echo "Run with --help for usage." >&2
  exit 1
fi

ROOM_DIR="$1"
SPEAKER_SLUG="$2"
DISPLAY_NAME="$3"
PROJECT_CONTEXT="$4"
APPLY_FLAG=""
LIST_MODE=false
APPROVE_TOKEN=""
shift 4
while [[ $# -gt 0 ]]; do
  case "$1" in
    --apply) APPLY_FLAG="--apply" ;;
    --list-queries) LIST_MODE=true ;;
    --approve)
      if [[ $# -ge 2 ]]; then APPROVE_TOKEN="$2"; shift; fi
      ;;
  esac
  shift
done
SCRIPT_DIR="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)"
GATE_CLI="${SCRIPT_DIR}/../lib/core/web-path-gate.cjs"

# Validate room_dir exists
if [[ ! -d "$ROOM_DIR" ]]; then
  echo "ERROR: Room directory does not exist: $ROOM_DIR" >&2
  exit 1
fi

# Find the speaker's profile (search across all role directories)
PROFILE_PATH=""
while IFS= read -r path; do
  PROFILE_PATH="$path"
  break
done < <(find "${ROOM_DIR}/team" -path "*/${SPEAKER_SLUG}/PROFILE.md" 2>/dev/null)

if [[ -z "$PROFILE_PATH" ]]; then
  echo "ERROR: No profile found for speaker '${SPEAKER_SLUG}' in ${ROOM_DIR}/team/" >&2
  exit 1
fi

PROFILE_DIR=$(dirname "$PROFILE_PATH")

# Read existing profile for role and dates
SPEAKER_ROLE=$(grep -m1 '^role:' "$PROFILE_PATH" | sed 's/^role: *//' || echo "unknown")
FIRST_MEETING=$(grep -m1 '^first_meeting:' "$PROFILE_PATH" | sed 's/^first_meeting: *//' || echo "unknown")
RESEARCH_STATUS=$(grep -m1 '^research_status:' "$PROFILE_PATH" | sed 's/^research_status: *//' || echo "pending")

# Check if already researched
if [[ "$RESEARCH_STATUS" == "complete" || "$RESEARCH_STATUS" == "completed" ]]; then
  RESEARCH_DATE=$(grep -m1 '^research_date:' "$PROFILE_PATH" | sed 's/^research_date: *//' || echo "unknown")
  echo "Already researched on ${RESEARCH_DATE}. Skipping."
  exit 0
fi

# Read room/STATE.md for venture context if available
VENTURE_NAME=""
VENTURE_DOMAIN=""
VENTURE_STAGE=""
if [[ -f "${ROOM_DIR}/STATE.md" ]]; then
  VENTURE_NAME=$(grep -m1 '^venture_name:' "${ROOM_DIR}/STATE.md" | sed 's/^venture_name: *//' 2>/dev/null || echo "")
  VENTURE_DOMAIN=$(grep -m1 '^domain:' "${ROOM_DIR}/STATE.md" | sed 's/^domain: *//' 2>/dev/null || echo "")
  VENTURE_STAGE=$(grep -m1 '^venture_stage:' "${ROOM_DIR}/STATE.md" | sed 's/^venture_stage: *//' 2>/dev/null || echo "")
fi

# Build search context from project_context and STATE.md
SEARCH_CONTEXT="${PROJECT_CONTEXT}"
if [[ -n "$VENTURE_DOMAIN" ]]; then
  SEARCH_CONTEXT="${SEARCH_CONTEXT} ${VENTURE_DOMAIN}"
fi

# Extract keywords for search queries
VENTURE_KEYWORDS=$(echo "$SEARCH_CONTEXT" | tr ' ' '\n' | sort -u | head -5 | tr '\n' ' ')

TODAY=$(date +%Y-%m-%d)

# URL-encode a string for search queries
urlencode() {
  local string="$1"
  python3 -c 'import sys,urllib.parse; print(urllib.parse.quote(sys.argv[1]))' "$string" 2>/dev/null \
    || echo "$string" | sed 's/ /+/g; s/"/%22/g'
}

# Construct the 3 search strings (a line break becomes a space, so one search is one line)
RAW1=$(printf '%s' "\"${DISPLAY_NAME}\" ${VENTURE_DOMAIN:-${PROJECT_CONTEXT}}" | tr '\r\n' '  ')
RAW2=$(printf '%s' "\"${DISPLAY_NAME}\" ${SPEAKER_ROLE} linkedin OR crunchbase OR twitter" | tr '\r\n' '  ')
RAW3=$(printf '%s' "\"${DISPLAY_NAME}\" ${VENTURE_KEYWORDS}" | tr '\r\n' '  ')
RS_STRINGS=("$RAW1" "$RAW2" "$RAW3")
QUERY1=$(urlencode "$RAW1")
QUERY2=$(urlencode "$RAW2")
QUERY3=$(urlencode "$RAW3")

# Phase 386 plan 24: ask before anything is sent (owner decision 2026-11-06).
# Order: node check, credential check, list mode, token check. Only then curl.
if ! command -v node >/dev/null 2>&1 || [[ ! -f "$GATE_CLI" ]]; then
  echo "Research not sent: the approval check needs node and lib/core/web-path-gate.cjs. Nothing was sent."
  exit 0
fi
if ! printf '%s\n' "${RS_STRINGS[@]}" | node "$GATE_CLI" credential --surface research_speaker >/dev/null 2>&1; then
  echo "Research not sent: a search looks like it holds a credential. Nothing was sent."
  exit 0
fi
EXPECTED_TOKEN=$(printf '%s\n' "${RS_STRINGS[@]}" | node "$GATE_CLI" token 2>/dev/null || echo "")
if [[ -z "$EXPECTED_TOKEN" ]]; then
  echo "Research not sent: the approval token could not be built. Nothing was sent."
  exit 0
fi
if [[ "$LIST_MODE" == true ]]; then
  echo "Searches that would be sent (3):"
  i=1
  for s in "${RS_STRINGS[@]}"; do
    echo "${i}. ${s}"
    i=$((i + 1))
  done
  echo "Approval token: ${EXPECTED_TOKEN}"
  echo "Nothing was sent."
  exit 0
fi
if [[ -z "$APPROVE_TOKEN" || "$APPROVE_TOKEN" != "$EXPECTED_TOKEN" ]]; then
  echo "Approval is needed before research is sent. Run with --list-queries to see the exact searches and the token. Nothing was sent."
  exit 0
fi

# Attempt web search using curl (Google search as fallback)
SEARCH_RESULTS=""
SEARCH_AVAILABLE=false

# Try fetching search results; one audit record per search sent
qi=0
for query in "$QUERY1" "$QUERY2" "$QUERY3"; do
  result=$(curl -s -m 10 "https://www.google.com/search?q=${query}&num=5" \
    -H "User-Agent: Mozilla/5.0" 2>/dev/null || echo "")
  outcome="failed"
  if [[ -n "$result" ]]; then
    SEARCH_AVAILABLE=true
    SEARCH_RESULTS="${SEARCH_RESULTS}${result}"
    outcome="ok"
  fi
  printf '%s\n' "${RS_STRINGS[$qi]}" | node "$GATE_CLI" record --room "$ROOM_DIR" --provider google_search --outcome "$outcome" >/dev/null 2>&1 \
    || echo "Audit record not written for search $((qi + 1))." >&2
  qi=$((qi + 1))
done

# Parse results (basic extraction)
BACKGROUND="No public information found"
RELEVANCE="Unable to determine relevance -- manual research recommended"
ONLINE_PRESENCE="No links found"
SUGGESTED_AFFILIATION="unknown"
SUGGESTED_EXPERTISE=""

if [[ "$SEARCH_AVAILABLE" == true && -n "$SEARCH_RESULTS" ]]; then
  # Extract titles and snippets from search results (basic HTML parsing)
  # Look for common patterns: LinkedIn titles, company mentions, etc.
  TITLES=$(echo "$SEARCH_RESULTS" | grep -E '<h3[^>]*>' 2>/dev/null | sed 's/.*<h3[^>]*>//;s/<\/h3>.*//' | head -10 || echo "")
  SNIPPETS=$(echo "$SEARCH_RESULTS" | grep -E '<span[^>]*>[^<]\{50,200\}' 2>/dev/null | sed 's/.*<span[^>]*>//;s/<\/span>.*//' | head -10 || echo "")

  if [[ -n "$TITLES" ]]; then
    BACKGROUND="Search results found. Titles include: $(echo "$TITLES" | head -3 | tr '\n' '; ')"
  fi

  # Look for LinkedIn pattern
  LINKEDIN=$(echo "$SEARCH_RESULTS" | grep -oE 'https?://[a-z]+\.linkedin\.com/in/[^ "]+' 2>/dev/null | head -1 || echo "")
  if [[ -n "$LINKEDIN" ]]; then
    ONLINE_PRESENCE="LinkedIn: ${LINKEDIN}"
  fi

  # Look for company/org mentions near the person's name
  if [[ -n "$SNIPPETS" ]]; then
    RELEVANCE="Found in search results related to: ${VENTURE_DOMAIN:-$PROJECT_CONTEXT}"
  fi
fi

# Compose research summary
RESEARCH_SUMMARY=$(cat <<SUMMARY
## Research: ${DISPLAY_NAME}
**Searched:** ${TODAY}
**Context:** ${PROJECT_CONTEXT}
**Role in meetings:** ${SPEAKER_ROLE}
**First meeting:** ${FIRST_MEETING}

### Professional Background
${BACKGROUND}

### Relevance to Project
${RELEVANCE}

### Online Presence
${ONLINE_PRESENCE}

### Suggested Profile Updates
- affiliation: ${SUGGESTED_AFFILIATION}
- expertise_areas: [to be determined from research and meeting contributions]
SUMMARY
)

# If search was not available at all
if [[ "$SEARCH_AVAILABLE" == false ]]; then
  RESEARCH_SUMMARY=$(cat <<SUMMARY
## Research: ${DISPLAY_NAME}
**Searched:** ${TODAY}
**Context:** ${PROJECT_CONTEXT}
**Status:** Web search not available

Web search not available. Populate manually or configure a search MCP.

### Suggested Actions
1. Search for "${DISPLAY_NAME}" on LinkedIn
2. Search for "${DISPLAY_NAME}" in the context of "${PROJECT_CONTEXT}"
3. Update the profile manually at: ${PROFILE_PATH}
SUMMARY
)
fi

# --apply mode: update PROFILE.md
if [[ "$APPLY_FLAG" == "--apply" ]]; then
  # Read current PROFILE.md
  CURRENT_CONTENT=$(cat "$PROFILE_PATH")

  # Update research_status from pending to complete
  UPDATED=$(echo "$CURRENT_CONTENT" | sed "s/^research_status: pending/research_status: complete/")
  UPDATED=$(echo "$UPDATED" | sed "s/^research_status: in-progress/research_status: complete/")

  # Add research_date to frontmatter (after research_status line)
  UPDATED=$(echo "$UPDATED" | sed "/^research_status:/a research_date: ${TODAY}")

  # Replace placeholder context section. The context reaches awk as data
  # through the environment, never as sed program text (quick 261030-eo7, N1:
  # a context could add the sed e or w flag and run a command or write a file).
  # Same result as the old sed line: the first placeholder on each line.
  UPDATED=$(echo "$UPDATED" | RS_REPLACEMENT="Researched on ${TODAY} in context of: ${PROJECT_CONTEXT}" awk '
    BEGIN { pat = "[To be filled after proactive research]"; rep = ENVIRON["RS_REPLACEMENT"] }
    { line = $0; i = index(line, pat); if (i > 0) line = substr(line, 1, i - 1) rep substr(line, i + length(pat)); print line }')

  # Write updated profile
  echo "$UPDATED" > "$PROFILE_PATH"

  echo "Profile updated: ${PROFILE_PATH}"
  echo "Research status: complete"
  echo "Research date: ${TODAY}"
  exit 0
fi

# Default: output research summary to stdout
echo "$RESEARCH_SUMMARY"
exit 0