deep-research

name	deep-research
description	Conduct deep web research using the openbrowser-ai agent: decompose a query, investigate sub-questions across multiple sources, and produce a cited markdown report plus structured JSON under local_docs/research/. Trigger when the user asks to: research a topic, do a deep dive, investigate, gather evidence, compare options, write a literature review, build a briefing, or produce a cited report.
allowed-tools	Bash(openbrowser-ai:*) Bash(curl:*) Bash(uv:*) Bash(irm:*) Bash(mkdir:*) Bash(date:*) Read Write

Deep Research

Drive openbrowser-ai to investigate a topic across multiple web sources and produce a cited markdown report plus structured JSON. Two modes:

• flat synthesis (default) -- decompose query into 3-7 sub-questions, dispatch one parallel sub-agent per sub-question (each owns one tab), merge into one cited report.
• drilldown (auto-detected from prompt phrasing: "deep dive", "exhaustive", "recursive", "drilldown", "thorough") -- same as flat, plus a second wave of parallel sub-agents on findings flagged needs_depth=true. Hard cap depth=2, max 3 follow-up sub-agents per parent.

Output paths (relative to current project root):

• local_docs/research/YYYY-MM-DD-<slug>.md
• local_docs/research/YYYY-MM-DD-<slug>.json

Architecture (mandatory): the orchestrating Claude session (the one running this skill) MUST dispatch parallel sub-agents via /dispatching-parallel-agents, one sub-agent per sub-question. Each sub-agent owns exactly ONE tab. Sub-agents do not open additional tabs. The orchestrator merges per-agent findings into one report.

Why one tab per sub-agent and not asyncio.gather over tabs in a single -c call: a single Python coroutine driving N tabs through one daemon serializes navigation events at the CDP layer, contends for the LLM-extraction worker, and cannot make independent decisions about pagination or follow-up clicks per tab. Dispatching real Claude sub-agents (each with its own context window and its own browser tab) gives true parallelism, independent reasoning per tab, and isolates failures so one bad page doesn't poison the rest.

Hard rules:

• One sub-agent = one tab. Sub-agents must NOT call navigate(url, new_tab=True) to spawn additional tabs.
• All sub-agents share the same daemon (and so the same Chrome process). Tabs are isolated; navigation in one tab does not affect another.
• Each sub-agent writes its findings to its own JSON file under local_docs/research/_partial/<slug>-NN.json. The orchestrator reads and merges these.
• The orchestrator never drives tabs itself. It only plans, dispatches, merges, renders, verifies, cleans up.

If a first-wave sub-agent returns <2 findings, the orchestrator dispatches a Step 2b retry sub-agent with broader search strategy (alternative engines, query reformulation, lower thresholds). Still -c-only: the skill never calls openbrowser-ai -p.

Variables persist across -c calls in the daemon namespace.

Session reuse: Step 0 checks openbrowser-ai daemon status. If a daemon is already running (warm browser), the skill reuses it and operates in NEW tabs (never disturbs the user's existing tabs). If no daemon, the skill auto-starts one on first -c call.

Every factual claim in the report carries a footnote citation [N]. Verifier fails the run if uncited prose is found.

Setup

Verify install:

openbrowser-ai --help

Install if missing:

# macOS / Linux
curl -fsSL https://openbrowser.me/install.sh | sh

# Windows PowerShell
irm https://openbrowser.me/install.ps1 | iex

No LLM API key required. The skill drives the daemon via openbrowser-ai -c only, which executes raw CDP / JS through the daemon's Python namespace and never invokes a model. (The -p "prompt mode" of the CLI is a separate code path that loads get_llm() and requires an OpenAI / Anthropic / Google key per cli.py:434-490. This skill explicitly avoids -p.)

Set the headless env var so the daemon starts without a visible browser window (the default in daemon/server.py is already headless: True, but a user config file can override it; this env var wins over config):

export OPENBROWSER_HEADLESS=true

Prepare output dir at the project root (NOT user home):

mkdir -p local_docs/research

Workflow

Step 0 -- Session check

Enforce headless mode and detect whether a daemon is already running. If yes, reuse it (operate in NEW tabs only). If no, the next -c call auto-starts one.

OPENBROWSER_HEADLESS=true is set here so the daemon spawned by the first -c call inherits it, even if the user's config file sets headless: false. Already-running daemons are unaffected (their browser was opened at start time).

export OPENBROWSER_HEADLESS=true

if openbrowser-ai daemon status 2>&1 | grep -qi 'running\|listening\|pid'; then
    echo "Reusing existing daemon -- will work in new tabs"
    export DEEP_RESEARCH_REUSED=1
else
    echo "No daemon running -- will start fresh headless session"
    export DEEP_RESEARCH_REUSED=0
fi

Snapshot existing tabs so cleanup leaves them untouched:

openbrowser-ai -c - <<'EOF'
state = await browser.get_browser_state_summary()
_preexisting_tab_ids = {t.target_id for t in state.tabs} if state.tabs else set()
print(f"Pre-existing tabs: {len(_preexisting_tab_ids)}")
EOF

Step 1 -- Plan

Decompose the user query into sub-questions and pick the mode. Daemon namespace persists _plan across later -c calls.

openbrowser-ai -c - <<'EOF'
import json, re, datetime, os

QUERY = """<USER_QUERY>"""  # paste exact user query here

# Daemon CWD often != shell CWD. Hard-code the absolute project root.
# Set this to the shell CWD at the start of the run; do NOT rely on os.getcwd().
PROJECT_ROOT = "<ABSOLUTE_PATH_TO_PROJECT_ROOT>"  # e.g. /Users/foo/myproject

# Auto-detect mode
DRILL_RE = re.compile(r"\b(deep ?dive|exhaustive|recursive|drill ?down|thorough)\b", re.I)
mode = "drilldown" if DRILL_RE.search(QUERY) else "flat"

# Slug = first 60 chars, lowercase, non-alnum -> '-', collapse repeats
def slugify(s):
    s = re.sub(r"[^a-z0-9]+", "-", s.lower())[:60]
    return s.strip("-") or "research"

today = datetime.date.today().isoformat()
slug = slugify(QUERY)
research_dir = os.path.join(PROJECT_ROOT, "local_docs", "research")
os.makedirs(research_dir, exist_ok=True)
base = os.path.join(research_dir, f"{today}-{slug}")
md_path, json_path = f"{base}.md", f"{base}.json"

# Bump suffix if collision
n = 2
while os.path.exists(md_path):
    md_path, json_path = f"{base}-{n}.md", f"{base}-{n}.json"
    n += 1

# Decompose: 3-7 sub-questions. Keep tight, non-overlapping, each answerable from web.
# This is a heuristic split; replace with your own decomposition for the actual query.
sub_questions = [
    # e.g.  "What is X?",
    #       "Who are the main actors / vendors / authors?",
    #       "What are recent (last 12 months) developments?",
    #       "What are the trade-offs / criticisms?",
    #       "What concrete numbers / benchmarks exist?",
]

assert 3 <= len(sub_questions) <= 7, "need 3-7 sub-questions"

_plan = {
    "query": QUERY,
    "mode": mode,
    "generated_at": datetime.datetime.now().astimezone().isoformat(),
    "sub_questions": sub_questions,
    "md_path": md_path,
    "json_path": json_path,
}
print(json.dumps(_plan, indent=2))
EOF

Edit the QUERY, PROJECT_ROOT, and sub_questions list before running. PROJECT_ROOT MUST be an absolute path: the daemon runs in its own working directory (usually wherever the daemon was first started), so relative paths land in the wrong place. Use the shell pwd output as the value. Verify output looks right before continuing.

Step 2a -- Dispatch parallel sub-agents (one tab per agent)

The orchestrator (the Claude session running this skill) MUST invoke /dispatching-parallel-agents and dispatch one sub-agent per sub-question. All sub-agents share the same openbrowser-ai daemon. Each sub-agent owns exactly one tab, the one it opens at the start of its run.

Hard rules for the orchestrator:

• Send ONE message containing N parallel Agent tool calls, where N = len(_plan["sub_questions"]).
• Each sub-agent gets the prompt template below, parameterized with: SUB_QUESTION, AGENT_INDEX (zero-padded 2 digits, used in output filename), PROJECT_ROOT (absolute path).
• After dispatch, wait for all sub-agents to return. Do not begin Step 4 until every partial JSON file under local_docs/research/_partial/ is on disk.
• The orchestrator does NOT drive any tab in this step.

Sub-agent prompt template (copy into each Agent tool call's prompt argument):

You are a deep-research sub-agent. Your job: investigate ONE sub-question
in ONE Chrome tab and write findings to a JSON file.

Sub-question: <SUB_QUESTION>
Agent index: <AGENT_INDEX>
Project root: <PROJECT_ROOT>

Hard constraints:

- You own ONE tab. The tab is the one you open at the start of this task.
- NEVER pass new_tab=True to navigate(). Reuse your one tab for every page.
- Do not switch to other tabs.
- Do not call openbrowser-ai daemon stop. The orchestrator owns daemon lifecycle.
- Visit at least 3 result URLs from at least 3 different domains.
- For each URL, extract one finding with: claim (one sentence), exact url,
  supporting quote (verbatim, max 40 words), domain, confidence (low|medium|high),
  needs_depth (bool, true if a deeper follow-up would meaningfully sharpen the claim).
- Return at least 2 findings; 3 is ideal.
- Write the findings JSON array to:
  <PROJECT_ROOT>/local_docs/research/_partial/agent-<AGENT_INDEX>.json

Workflow (run via openbrowser-ai -c - heredocs, all in this one bash session):

1. Open exactly one tab on Google search:
   openbrowser-ai -c - <<'EOF'
   await navigate("https://www.google.com/search?q=<URL_ENCODED_SUB_QUESTION>", new_tab=True)
   await wait(2)
   state = await browser.get_browser_state_summary()
   tab_id = state.tabs[-1].target_id
   _MY_TAB = tab_id[-4:]
   print(f"my tab: {_MY_TAB}")
   EOF

2. Scrape the SERP for top 5 result URLs.

3. For each of the top 3 results, navigate IN THE SAME TAB (no new_tab=True),
   wait 2s, capture body innerText, pick the sentence with the most word-overlap
   vs the sub-question (40-280 chars, >3-letter keyword overlap >= 1).

4. Write the findings JSON array to the partial file.

5. Print a one-line summary: "agent <AGENT_INDEX>: <N> findings written".

Return: a one-line confirmation that the partial JSON file was written, and
its absolute path. Do not include the findings in your text reply -- the
orchestrator will read them from disk.

The orchestrator's pre-step (run once before dispatching agents):

openbrowser-ai -c - <<'EOF'
import os
PROJECT_ROOT = "<ABSOLUTE_PATH_TO_PROJECT_ROOT>"
partial_dir = os.path.join(PROJECT_ROOT, "local_docs", "research", "_partial")
os.makedirs(partial_dir, exist_ok=True)
# Wipe stale partials from prior runs of the same query
for f in os.listdir(partial_dir):
    if f.startswith("agent-") and f.endswith(".json"):
        os.remove(os.path.join(partial_dir, f))
print(f"partial dir ready: {partial_dir}")
EOF

After all sub-agents return, the orchestrator collects findings:

openbrowser-ai -c - <<'EOF'
import os, json
global _findings
PROJECT_ROOT = "<ABSOLUTE_PATH_TO_PROJECT_ROOT>"
partial_dir = os.path.join(PROJECT_ROOT, "local_docs", "research", "_partial")

_findings = []
for fname in sorted(os.listdir(partial_dir)):
    if not (fname.startswith("agent-") and fname.endswith(".json")):
        continue
    path = os.path.join(partial_dir, fname)
    try:
        with open(path) as fh:
            arr = json.load(fh)
        if isinstance(arr, list):
            # normalize exact_url -> url (sub-agents may use either key)
            for item in arr:
                if "exact_url" in item and "url" not in item:
                    item["url"] = item.pop("exact_url")
            _findings.extend(arr)
    except Exception as e:
        print(f"skip {path}: {e}")

print(f"merged {len(_findings)} findings from {len(os.listdir(partial_dir))} partials")
EOF

If any sub-question's partial file is missing or has <2 findings, fall through to Step 2b for that one.

Step 2b -- Retry weak sub-questions with broader sub-agents

If a sub-question came back with <2 findings, do NOT use openbrowser-ai -p. The -p mode requires an LLM API key (OpenAI / Anthropic / Google) because it runs the full Browser Agent loop with LLM-driven navigation, see cli.py:434-490 get_llm(). This skill is -c-only by design: the daemon executes raw CDP / JS and needs no API key.

Identify weak sub-questions, then dispatch a fresh round of parallel sub-agents (same /dispatching-parallel-agents pattern as Step 2a) with a broader prompt. The retry sub-agents reuse the same single-tab discipline and write to agent-retry-NN.json partials.

openbrowser-ai -c - <<'EOF'
from collections import Counter
global _retry_subqs
counts = Counter(f.get("sub_question") for f in _findings)
_retry_subqs = [sq for sq in _plan["sub_questions"] if counts.get(sq, 0) < 2]
print(f"Retry targets: {len(_retry_subqs)}")
for sq in _retry_subqs:
    print(f"  - {sq}")
EOF

The orchestrator then dispatches one parallel sub-agent per weak sub-question with this retry prompt template (same single-tab rule, broader search strategy):

You are a deep-research RETRY sub-agent. Your job: investigate ONE sub-question
that the first-wave sub-agent could not satisfy. Use ONE Chrome tab and write
findings to a JSON file.

Sub-question: <SUB_QUESTION>
Agent index: <AGENT_INDEX> (filename agent-retry-<AGENT_INDEX>.json)
Project root: <PROJECT_ROOT>

Hard constraints:

- You own ONE tab. NEVER pass new_tab=True to navigate() after the first call.
- Use openbrowser-ai -c only. NEVER use openbrowser-ai -p (it requires an LLM
  API key and we explicitly avoid that path).
- The first-wave attempt failed: heuristic sentence picker returned <2 findings.
  This means the page text either had no high-overlap sentences for the question,
  or the SERP returned thin sources. Try one or more of:
  1. Reformulate the search query (try 2-3 alternative phrasings, pick the one
     with the best SERP).
  2. Search a different engine: try Bing or DuckDuckGo if Google was thin.
     URLs: https://www.bing.com/search?q=... or https://duckduckgo.com/?q=...
  3. Lower the sentence-length floor for the heuristic (e.g. 30 chars instead
     of 40), or accept partial-match sentences with overlap >= 1 word.
  4. For factual sub-questions ("when was X released"), check Wikipedia
     directly: https://en.wikipedia.org/wiki/Special:Search?search=...
- Same JSON shape as Step 2a sub-agents.
- Write to: <PROJECT_ROOT>/local_docs/research/_partial/agent-retry-<AGENT_INDEX>.json
- Return at least 2 findings. If still <2 after the broader strategy, write
  whatever you got (even 0-1) and surface the limitation in your reply.

After all retry sub-agents return, merge their partials into _findings:

openbrowser-ai -c - <<'EOF'
import os, json
global _findings
PROJECT_ROOT = "<ABSOLUTE_PATH_TO_PROJECT_ROOT>"
partial_dir = os.path.join(PROJECT_ROOT, "local_docs", "research", "_partial")
extra = []
for fname in sorted(os.listdir(partial_dir)):
    if not (fname.startswith("agent-retry-") and fname.endswith(".json")):
        continue
    path = os.path.join(partial_dir, fname)
    try:
        with open(path) as fh:
            arr = json.load(fh)
        if isinstance(arr, list):
            for item in arr:
                if "exact_url" in item and "url" not in item:
                    item["url"] = item.pop("exact_url")
            extra.extend(arr)
    except Exception as e:
        print(f"skip {path}: {e}")
_findings.extend(extra)
print(f"Retry added {len(extra)} findings -> total {len(_findings)}")
EOF

_findings lives in the daemon namespace for the next steps.

Step 3 -- Drilldown (drilldown mode only): second wave of parallel sub-agents

Skip this step in flat mode. In drilldown mode, dispatch a second wave of /dispatching-parallel-agents -- one sub-agent per needs_depth=true finding, max 3 follow-ups per parent sub-question. Each follow-up sub-agent owns ONE tab, exactly like Step 2a.

The orchestrator first picks targets:

openbrowser-ai -c - <<'EOF'
global _drill_targets
if _plan["mode"] != "drilldown":
    print("flat mode, skipping drilldown")
    _drill_targets = []
else:
    by_sq = {}
    for f in _findings:
        if f.get("needs_depth"):
            by_sq.setdefault(f["sub_question"], []).append(f)
    _drill_targets = []
    for sq, fs in by_sq.items():
        _drill_targets.extend(fs[:3])
    print(f"Drilldown targets: {len(_drill_targets)}")
    for t in _drill_targets:
        print(f"  - {t['claim'][:80]}  ({t['url']})")
EOF

Then dispatch one parallel sub-agent per drilldown target. Use the same single-tab-per-agent rule and the same partial-file output convention, but with this drilldown prompt template:

You are a deep-research drilldown sub-agent. Your job: investigate ONE claim
in ONE Chrome tab and write supporting/contradicting evidence to a JSON file.

Parent claim: <CLAIM>
Original source URL: <URL>
Agent index: <AGENT_INDEX> (use prefix "drill-" -> filename agent-drill-<AGENT_INDEX>.json)
Project root: <PROJECT_ROOT>

Hard constraints:

- You own ONE tab. NEVER pass new_tab=True to navigate() after the first call.
- Find at least 2 additional sources from at least 2 different domains
  (different from the original source URL above).
- needs_depth must be false on every finding you return (depth cap reached).
- Same JSON shape as Step 2a sub-agents: claim, url, quote, domain, confidence,
  needs_depth, plus add sub_question = "<PARENT_SUB_QUESTION>".
- Write to: <PROJECT_ROOT>/local_docs/research/_partial/agent-drill-<AGENT_INDEX>.json

After all drilldown sub-agents return, merge their partials into _findings:

openbrowser-ai -c - <<'EOF'
import os, json
global _findings
PROJECT_ROOT = "<ABSOLUTE_PATH_TO_PROJECT_ROOT>"
partial_dir = os.path.join(PROJECT_ROOT, "local_docs", "research", "_partial")
extra = []
for fname in sorted(os.listdir(partial_dir)):
    if not (fname.startswith("agent-drill-") and fname.endswith(".json")):
        continue
    path = os.path.join(partial_dir, fname)
    try:
        with open(path) as fh:
            arr = json.load(fh)
        if isinstance(arr, list):
            for f in arr:
                f["needs_depth"] = False  # cap reached
                if "exact_url" in f and "url" not in f:
                    f["url"] = f.pop("exact_url")
            extra.extend(arr)
    except Exception as e:
        print(f"skip {path}: {e}")
_findings.extend(extra)
print(f"Drilldown added {len(extra)} findings -> total {len(_findings)}")
EOF

Step 4 -- Aggregate and dedup

Merge findings, dedup by URL, build the source index that footnote numbers will point to.

openbrowser-ai -c - <<'EOF'
from urllib.parse import urlparse

# `global` so reassignment of daemon-namespace var doesn't shadow it
global _findings, _sources

# Normalize exact_url -> url (sub-agents may use either key)
for f in _findings:
    if "exact_url" in f and "url" not in f:
        f["url"] = f.pop("exact_url")

# Dedup by URL, keep first occurrence
seen_urls = {}
deduped = []
for f in _findings:
    u = (f.get("url") or "").strip()
    if not u:
        continue  # drop uncited findings entirely
    if u in seen_urls:
        continue
    seen_urls[u] = True
    deduped.append(f)

# Build sources index, 1-based ids
_sources = []
url_to_id = {}
for f in deduped:
    u = f["url"]
    if u not in url_to_id:
        sid = len(_sources) + 1
        url_to_id[u] = sid
        domain = f.get("domain") or urlparse(u).netloc
        _sources.append({"id": sid, "url": u, "domain": domain, "title": f.get("title", "")})
    f["source_id"] = url_to_id[u]

_findings = deduped
print(f"After dedup: {len(_findings)} findings, {len(_sources)} unique sources")
EOF

Step 5 -- Render report

Write paired markdown + JSON. Every claim line in markdown ends with [N] cite. Summary cites top 3-5 sources.

openbrowser-ai -c - <<'EOF'
import json, datetime
from collections import defaultdict

# Group findings by sub-question (preserve plan order)
groups = defaultdict(list)
for f in _findings:
    groups[f["sub_question"]].append(f)

lines = []
lines.append(f"# Research: {_plan['query']}")
lines.append("")
gen = datetime.datetime.fromisoformat(_plan["generated_at"]).strftime("%Y-%m-%d %H:%M %Z").strip()
lines.append(f"Generated: {gen}  *  Mode: {_plan['mode']}  *  Sources: {len(_sources)}")
lines.append("")

# Summary: pick first finding from each sub-question, max 5 sentences
lines.append("## Summary")
lines.append("")
summary_sents = []
for sq in _plan["sub_questions"]:
    fs = groups.get(sq, [])
    if fs:
        f = fs[0]
        claim = " ".join(f["claim"].split()).rstrip(".")
        summary_sents.append(f"{claim}[{f['source_id']}].")
    if len(summary_sents) >= 5:
        break
lines.append(" ".join(summary_sents) or "_No findings._")
lines.append("")

# Per sub-question section
for sq in _plan["sub_questions"]:
    lines.append(f"## {sq}")
    lines.append("")
    fs = groups.get(sq, [])
    if not fs:
        lines.append("_No findings for this sub-question._")
        lines.append("")
        continue
    for f in fs:
        claim = " ".join(f["claim"].split()).rstrip(".")  # collapse all whitespace incl newlines
        lines.append(f"- {claim}[{f['source_id']}].")
        q = " ".join((f.get("quote") or "").split()).strip()
        if q:
            lines.append(f"  > {q}")
    lines.append("")

# Sources
lines.append("## Sources")
lines.append("")
for s in _sources:
    title = s.get("title") or s["domain"]
    lines.append(f"{s['id']}. [{title}]({s['url']})")
lines.append("")

md = "\n".join(lines)

with open(_plan["md_path"], "w") as fh:
    fh.write(md)

payload = {
    "query": _plan["query"],
    "mode": _plan["mode"],
    "generated_at": _plan["generated_at"],
    "sub_questions": _plan["sub_questions"],
    "findings": _findings,
    "sources": _sources,
}
with open(_plan["json_path"], "w") as fh:
    json.dump(payload, fh, indent=2, ensure_ascii=False)

print(f"Wrote {_plan['md_path']}")
print(f"Wrote {_plan['json_path']}")
EOF

Step 6 -- Verify citations

Fail loud if any prose sentence outside headings, quotes, code, or source list lacks a [N] cite. Fix by adding the missing source then rerendering, never by deleting prose silently.

Do the verify in plain shell python3 (not the daemon). sys.exit inside the daemon namespace kills the daemon.

python3 - "$PROJECT_ROOT" <<'EOF'
import re, sys, os

project_root = sys.argv[1] if len(sys.argv) > 1 else os.getcwd()
research_dir = os.path.join(project_root, "local_docs", "research")

# Find latest report file -- use os.listdir + absolute path (glob fails on non-shell CWD)
files = sorted(
    [os.path.join(research_dir, f) for f in os.listdir(research_dir) if f.endswith(".md")],
    key=os.path.getmtime, reverse=True
)
if not files:
    print("No report found"); sys.exit(1)
md_path = files[0]

with open(md_path) as fh: md = fh.read()
md_no_code = re.sub(r"```.*?```", "", md, flags=re.S)

violations = []
in_sources = False
for i, line in enumerate(md_no_code.splitlines(), 1):
    s = line.strip()
    if not s: continue
    if s.startswith("## Sources"):
        in_sources = True
        continue
    if in_sources: continue
    if s.startswith("#"): continue
    if s.startswith(">"): continue
    if re.match(r"^\d+\.\s+\[", s): continue
    if s.startswith("Generated:"): continue
    if s in ("_No findings._", "_No findings for this sub-question._"): continue
    if not re.search(r"\[\d+\]", s):
        violations.append((i, s[:120]))

if violations:
    print("CITATION VIOLATIONS:")
    for ln, txt in violations:
        print(f"  L{ln}: {txt}")
    sys.exit(1)
print(f"OK: all prose cited in {md_path}")
EOF

Step 7 -- Cleanup

Two cases:

Case A: daemon was already running before the skill started (DEEP_RESEARCH_REUSED=1). Close ONLY the research tabs and leave the daemon + pre-existing tabs alone:

if [ "$DEEP_RESEARCH_REUSED" = "1" ]; then
openbrowser-ai -c - <<'EOF'
state = await browser.get_browser_state_summary()
to_close = [t.target_id for t in state.tabs if t.target_id not in _preexisting_tab_ids]
print(f"Closing {len(to_close)} research tabs, preserving {len(_preexisting_tab_ids)} pre-existing")
for tid in to_close:
    try:
        await close(tab_id=tid[-4:])
    except Exception as e:
        print(f"  skip {tid[-4:]}: {e}")
EOF
fi

Case B: daemon was started fresh by this skill (DEEP_RESEARCH_REUSED=0). Full daemon stop, freeing the Chrome process. See Cleanup section below.

Always: remove the per-agent partial JSON files. They are intermediate state, not part of the report:

rm -rf "<ABSOLUTE_PATH_TO_PROJECT_ROOT>/local_docs/research/_partial" 2>/dev/null || true

Tips

• Headless by default: OPENBROWSER_HEADLESS=true is set in Step 0 so the daemon always starts without a visible browser window. The daemon default in daemon/server.py is already headless: True, but a user config file can override it. The env var wins over config. If you need a visible window for debugging, unset or override: export OPENBROWSER_HEADLESS=false.
• One tab per sub-agent: the orchestrator dispatches via /dispatching-parallel-agents and each sub-agent must own exactly one tab. Multiple-tab sub-agents serialize navigation at the CDP layer and lose the parallelism benefit. Enforce in the agent prompt: "NEVER pass new_tab=True to navigate() after the first call."
• Single message, multiple Agent calls: to actually parallelize, the orchestrator must put all N Agent tool calls in ONE message. Sequential Agent invocations across messages run serially.
• Heredoc quoting: always use <<'EOF' (single-quoted) so $, backticks, and ! inside Python don't expand in the shell.
• Daemon namespace: _plan, _findings, _sources persist across orchestrator -c calls. Sub-agents share the same daemon and so see the same namespace, but each operates in its own tab. Don't restart the daemon mid-run.
• Partial files as the contract: sub-agents write to local_docs/research/_partial/agent-NN.json, the orchestrator reads them back. Sub-agents do NOT communicate findings via stdout. If a partial file is missing or empty, that sub-agent failed; rerun it or fall through to Step 2b.
• exact_url normalization: sub-agents may write exact_url instead of url. All merge steps (Step 2a, 2b, 3) normalize this automatically. Step 4 has a final safety pass. Never rely on url being set before the merge step runs.
• Daemon restart recovery: if the daemon restarts mid-session, _plan/_findings/_sources are lost from namespace. The partial JSON files on disk survive. Re-run the merge step (Step 2a collect block) with PROJECT_ROOT hardcoded to reload _findings from disk. Re-run Step 1 with the same QUERY and PROJECT_ROOT to restore _plan; the slug-collision bump will avoid overwriting the existing output file.
• No -p, ever: this skill never shells out to openbrowser-ai -p. The -p mode runs the full Browser Agent loop with LLM-driven navigation and requires an OpenAI / Anthropic / Google API key (cli.py:434-490 get_llm()). The -c daemon path needs no API key. If you see -p anywhere in skill code, that is a bug.
• Source diversity: if dedup leaves <60% of findings (heavy domain repetition), rerun the affected sub-agent with explicit "prefer different domains than: " appended to its prompt.
• Drilldown cost: drilldown dispatches up to len(sub_questions) * 3 extra parallel sub-agents. Reserve for genuinely deep topics.
• Rerun semantics: rerunning Step 5 alone re-renders from current _findings, useful after manual edits.
• Slug collisions: Step 1 auto-bumps -2, -3 if file exists, so rerunning the same query never overwrites.

Cleanup

Mandatory. Run after every workflow, success or failure.

If this skill started the daemon (DEEP_RESEARCH_REUSED=0), stop it fully:

if [ "$DEEP_RESEARCH_REUSED" = "0" ]; then
    openbrowser-ai daemon stop
    openbrowser-ai daemon status
fi

If daemon was reused (DEEP_RESEARCH_REUSED=1), Step 7 already closed only the research tabs. Do NOT call daemon stop, that kills the user's pre-existing browser session.

Force-kill fallback (only if daemon was started by skill and stop failed):

[ "$DEEP_RESEARCH_REUSED" = "0" ] && pkill -f 'openbrowser.*daemon' || true

Trap for failure-safe cleanup at script top:

trap '[ "$DEEP_RESEARCH_REUSED" = "0" ] && openbrowser-ai daemon stop >/dev/null 2>&1 || true' EXIT

Anti-patterns:

• Do NOT daemon stop when reusing an existing session; that kills the user's open tabs.
• Do NOT rely on the 600s idle timeout; that wastes a Chrome process for 10 minutes.
• Do NOT use done() as a substitute; it only ends the agent loop, browser stays open.
• Do NOT mix --mcp mode with the daemon; separate profiles, browser contention.