# Technology Research Guide

> **Guide Origin**: Official | **ArcKit Version**: [VERSION]

`/arckit:research` investigates market, SaaS, open-source, and government marketplace options to support build vs buy decisions.

> **Agent Architecture**: This command delegates to the `arckit-research` autonomous agent. The agent runs as a subprocess with its own context window, performing dozens of WebSearch and WebFetch calls for vendor pricing, reviews, and compliance data without polluting your main conversation. The slash command launches the agent and relays its summary back to you.

---

## Scenario Matrix

| Scenario | Prompt seed | Focus |
|---------|-------------|-------|
| Build vs buy | “Research options for <capability> and recommend build vs buy” | Compares custom development vs product |
| Supplier shortlist | “Research G-Cloud services for <need> including pricing tiers” | Public sector supplier discovery |
| Standards review | “Research regulatory obligations for <domain>” | Highlights policies, certifications, compliance |
| Migration | “Research tooling to migrate from <legacy> to <target>” | Guides modernisation approaches |
| Risk investigation | “Research security/operational risks for <technology>” | Informs risk register and mitigations |

Add constraints (budget, data residency, clearance) in the prompt for tailored results.

---

## Command

```bash
/arckit:research Research <topic> for <project>
```

Outputs: `projects/<id>/ARC-<id>-RSCH-v1.0.md` plus optional CSV of suppliers.

> **Auto-versioning**: Re-running this command when a document already exists automatically increments the version (minor for refreshed content, major for changed scope) instead of overwriting.

---

## Long runs: Remote Control + push notifications

`/arckit:research` frequently exceeds 10 minutes once the agent starts pulling vendor pricing pages, G-Cloud listings, and product reviews. To avoid babysitting the terminal, pair it with [Claude Code Remote Control](https://code.claude.com/docs/en/remote-control):

```bash
claude remote-control
```

Drive the session from claude.ai/code or the mobile app, then enable `/config → Push when Claude decides` so your phone gets a notification on completion or when the agent reaches a decision point (vendor shortlist confirmation, build vs buy direction). ArcKit's minimum Claude Code floor (v2.1.287) already covers the v2.1.110 RC requirement.

Caveats: Pro/Max plans only (no API keys, no Bedrock/Vertex/Foundry), push is a single on/off so chatty agents can over-notify, and the local `claude` process must keep running.

## Session search budget

Claude Code v2.1.212 added a session-wide cap of **200 WebSearch calls**, after which further searches are refused. `/arckit:research` is deliberately search-heavy — the agent is instructed to discover the current market landscape rather than rely on general knowledge — so a single large-scope run uses a meaningful slice of that budget, and several research commands back to back in one session can exhaust it.

If you hit the cap, either start a fresh session (`/clear` also resets the parallel subagent budget) or raise it:

```bash
export CLAUDE_CODE_MAX_WEB_SEARCHES_PER_SESSION=500
claude
```

The cap exists to stop runaway search loops, so raise it deliberately rather than as a default. `WebFetch` calls are not counted against it.

`WebFetch` has a session cache instead: a URL fetched in the last 15 minutes is served from memory rather than re-fetched. `CLAUDE_CODE_WEBFETCH_CACHE_TTL_MS` (Claude Code v2.1.233+) adjusts that window in milliseconds — raise it when a long run re-reads the same policy pages, lower it when the source is changing underneath you. Before v2.1.239 expired entries stayed in memory for the whole session; that leak is fixed.

---

## Output Highlights

- Option catalogue with pros, cons, pricing, support model.
- Alignment to requirements and principles.
- Risk considerations (lock-in, data sovereignty, accessibility).
- Recommendation with rationale and next steps (PoC, procurement, or custom build).

---

## Follow-on Actions

- Feed supplier data into `/arckit:sow` and `/arckit:evaluate`.
- Update Wardley Maps with evolution stage insights.
- Add identified risks to `/arckit:risk` and mitigations to project backlog.
- Cite findings in business case and design reviews.
