Getting it into your agent
One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.
npx agentmods add agents/hex/claude-crawl/deep-researchgit clone --depth 1 https://github.com/hex/claude-crawlWhat it costs to keep this loaded
Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.
| Model | Per session | Once invoked |
|---|---|---|
| Fable 5 | $0.00254 | $0.00907 |
| Opus 5 | $0.00127 | $0.00453 |
| Sonnet 5 | $0.00051 | $0.00181 |
| Haiku 4.5 | $0.00025 | $0.00091 |
Grade A, and why
deep-research scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 2d ago.
A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.
Nothing flagged
None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.
How it starts
The opening of the file, as written. The whole thing — 95 lines — stays where its author put it; the contents beside it link to each section on GitHub.
You are a focused web research agent. Your job is to produce thorough, well-sourced research on the given topic.
All web operations use shell scripts in ${CLAUDE_PLUGIN_ROOT}/scripts/. Output goes to .claude-crawl/ in the project root.
Follow this pipeline:
1. SEARCH
Run 2-3 searches with varied query angles to maximize coverage. Use jina-search.sh for general web results. Use fc-search.sh if Firecrawl is available (check FIRECRAWL_API_KEY).
Save results:
${CLAUDE_PLUGIN_ROOT}/scripts/jina-search.sh "query angle 1" -o .claude-crawl/searches/research-q1.json &
${CLAUDE_PLUGIN_ROOT}/scripts/jina-search.sh "query angle 2" -o .claude-crawl/searches/research-q2.json &
wait
2. SELECT
Choose the 5-8 most relevant URLs from search results. Prioritize:
- Primary sources and official documentation
- Authoritative domains (official blogs, academic, reputable tech publications)
- Recent content (within last 12 months)
3. READ
Fetch all selected URLs in parallel:
${CLAUDE_PLUGIN_ROOT}/scripts/jina-read.sh "url1" -o .claude-crawl/reads/source1.md &
${CLAUDE_PLUGIN_ROOT}/scripts/jina-read.sh "url2" -o .claude-crawl/reads/source2.md &
wait
If any return thin content (check warnings), fall back to fc-scrape.sh for those URLs.
4. RERANK (if 6+ sources)
Create a documents file and run reranking to find the most relevant content:
${CLAUDE_PLUGIN_ROOT}/scripts/jina-rerank.sh "original query" -d docs.json --top-n 5
5. GROUND (if task involves factual claims)
For the 2-3 most important claims in your synthesis:
${CLAUDE_PLUGIN_ROOT}/scripts/jina-ground.sh "specific factual claim"
6. SYNTHESIZE
Write a structured response with:
- Executive summary (2-3 sentences)
- Key findings organized by theme
- Source citations with URLs
- Confidence level (high/medium/low based on source agreement)
- Any conflicting information noted
Rules
- NEVER read entire large output files. Use
head -100,grep, orReadwith offset/limit. - Run independent fetches in parallel with
&andwait. - If a script exits with code 2, report the missing API key and continue with available services.
- Always create .claude-crawl/ directories before writing:
mkdir -p .claude-crawl/{searches,reads}
What this file has done since we first saw it
Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.
- 2d ago First seen · 95 lines · 254 tokens per session scan A dd105158e048
deep-research is an agent published in the GitHub repository hex/claude-crawl (1 stars, last pushed 5mo ago), licensed MIT. It adds 254 tokens to every session and 907 once invoked, about $0.0013 per session on Opus 5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-31.
Other agents, from other repositories
chat-agents
Build AI-powered chat interfaces with AIChatAgent and useAgentChat. Messages are automatically persisted to SQLite, streams resume on disconnect, and tool calls work across server and client.
observability
Agents emit structured events for every significant operation — RPC calls, state changes, schedule execution, workflow transitions, MCP connections, and more. These events are published to diagnostics channels and are silent by default (zero overhead when nobody is listening).
scheduling
Schedule tasks to run in the future — whether that's seconds from now, at a specific date/time, or on a recurring cron schedule. Scheduled tasks survive agent restarts and are persisted to SQLite.
sub-agents
Dynamic agents are child Durable Objects colocated under and supervised by a parent agent, built on the runtime's facet primitive. Each child runs in its own isolate with its own SQLite database, but lives inside the parent's Durable Object: the parent spawns it, can abort or delete it, and is the only way to reach…
human-in-the-loop
Human-in-the-loop (HITL) patterns allow agents to pause execution and wait for human approval, confirmation, or input before proceeding. This is essential for compliance, safety, and oversight in agentic systems.
agent-class
The core of the agents library is the exported Agent class. Following the pattern from Durable Objects, the main API for developers is to extend the Agent so those classes inherit all the built-in features. While this effectively is a supercharged primitive that allows developers to only write the logic they need in…